Behaviour difficulties in children with special education needs ...

Behaviour difficulties in children with special education needs and disabilities: assessing risk,

promotive and protective factors at individual and school levels

A thesis submitted to the University of Manchester for the degree of

PhD in the Faculty of Humanities

2012

Jeremy Oldfield

School of Education

1

Table of Contents

LIST OF TABLES…………………………………..……………………………….…6 LIST OF FIGURES……………………………………………………………….……8 ABSTRACT…………………………………………………………….………………9 DECLARATION………………………………………………………………………10 COPYRIGHT STATEMENT…………………………………………………………10 ABBREVIATIONS……………………………………………………………………11 ACKNOWLEDGEMENTS……………………………………………………...……13 CHAPTER 1 1.1 Introduction to chapter ................................................................................ 14

1.2 Behaviour Problems .................................................................................... 16

1.2.1 Definitions and Terminologies .............................................................. 16

1.2.2 Prevalence rates and time trends in behaviour difficulties .................... 29

1.2.3 Outcomes associated with childhood behaviour difficulties .................. 32

1.2.4 Summary Statements ........................................................................... 35

1.3 Special Educational Needs and Disabilities (SEND) ................................... 36

1.3.1 Overview of the concept of SEND ........................................................ 36

1.3.2 Justification of SEND population within the study................................. 46

1.3.3 Summary statements ........................................................................... 49

1.4 Risk, Promotion and Protection .................................................................. 50

1.4.1 Risk ...................................................................................................... 50

1.4.2 Promotion and Protection ..................................................................... 54

1.4.3 Summary statements ........................................................................... 60

1.5 Aim of study and Research Questions ........................................................ 61

1.5.1 Aim ....................................................................................................... 61

1.5.2 Research Questions ............................................................................. 63

CHAPTER 2

2.1 Introduction to chapter ................................................................................ 64

2.2 Individual level risk and promotive factors .................................................. 66

2.2.1 Age ....................................................................................................... 66

2.2.2 Relative age in the school year ............................................................ 67

2.2.3 Gender ................................................................................................. 69

2.2.4 Ethnicity ................................................................................................ 70

2.2.5 Academic achievement ........................................................................ 71

2.2.6 Attendance ........................................................................................... 72

2

2.2.7 Positive relationships ............................................................................ 74

2.2.8 Bullying ................................................................................................. 75

2.2.9 Special Educational Needs and Disabilities .......................................... 77

2.2.10 Other individual level risk factors ........................................................ 79

2.3 Family level risk factors ............................................................................... 80

2.3.1 Socio-economic status ......................................................................... 80

2.3.2 Other familial level risk factors .............................................................. 81

2.4 School level risk factors .............................................................................. 82

2.4.1 School Urbanicity ................................................................................. 82

2.4.2 School size ........................................................................................... 83

2.4.3 School level socio economic status ...................................................... 84

2.4.4 School EAL composition....................................................................... 85

2.4.5 Proportion of SEND pupils .................................................................... 86

2.4.6 School level achievement ..................................................................... 88

2.4.7 School level attendance ....................................................................... 89

2.4.8 School level behaviour difficulties ......................................................... 90

2.4.9 Other school level risk and promotive factors ....................................... 92

2.5 Neighbourhood level risk and promotive factors ......................................... 93

2.5.1 Neighbourhood disadvantage ............................................................... 93

2.5.2 Other neighbourhood level risk and promotive factors ......................... 94

2.6 The relative importance of the individual versus the school ecological level in accounting for behaviour difficulties .............................................................. 95

2.7 Summary Statements ................................................................................. 98

CHAPTER 3

3.1 Introduction to chapter ................................................................................ 99

3.2 Multiple risk factors ..................................................................................... 99

3.3 What is cumulative risk? ........................................................................... 101

3.4 Measurement of cumulative risk .............................................................. 102

3.5 Evaluation of cumulative risk .................................................................... 103

3.6 The functional form of cumulative risk models .......................................... 106

3.7 Cumulative risk across different ecological levels ..................................... 109

3.8 Conclusions .............................................................................................. 110

3.9 Chapter Summary and research questions ............................................... 112

3

CHAPTER 4

4.1. Introduction to chapter ............................................................................. 113

4.2 Protective factors and resilience ............................................................... 113

4.3 Measuring risk contexts for protective factor research .............................. 115

4.4 Measuring outcomes for protective factor research .................................. 117

4.5 Approaches to protective factor research ................................................. 118

4.5.1 Variable focused approaches ............................................................. 118

4.5.2. Person focused approaches .............................................................. 123

4.6 Protective factors and behaviour difficulties .............................................. 123

4.7 School level protective factors .................................................................. 127

4.8 Conclusions .............................................................................................. 129

4.9 Summary Statements ............................................................................... 130

CHAPTER 5

5.1 Introduction to chapter .............................................................................. 131

5.2 Theoretical framework .............................................................................. 132

5.2.1 Ecological Systems Theory ................................................................ 132

5.2.2 Framing the present study within Ecological Systems Theory ........... 135

5.3 Context of the Present Study .................................................................... 138

5.3.1What is Achievement for All? .............................................................. 138

5.3.2 Achievement for All evaluation study .................................................. 140

5.3.3 Overview of Methodology ................................................................... 141

5.3.4 How is the present study different from the AfA evaluation study? .... 142

5.4 Design ....................................................................................................... 144

5.4.1 Epistemology ...................................................................................... 144

5.4.2 Post-Positivist Approach .................................................................... 145

5.4.3 Quantitative Methodology ................................................................... 146

5.4.4 Survey design..................................................................................... 147

5.4.5 Teacher, parent or self-report ............................................................. 149

5.4.6 Longitudinal study .............................................................................. 150

5.4.7 Variables in the study ......................................................................... 151

5.5 Participants ............................................................................................... 155

5.5.1 Local Authorities ................................................................................. 155

5.5.2 Schools .............................................................................................. 155

5.5.3 Pupils ................................................................................................. 155

5.5.4 Attrition ............................................................................................... 156

4

5.5.5 Characteristics of final sample ............................................................ 157

5.6 Materials ................................................................................................... 164

5.6.1 Overview ............................................................................................ 164

5.6.2 Wider Outcomes Survey for Teachers (WOST) ................................. 164

5.6.3 Scoring ............................................................................................... 165

5.6.4 Psychometric properties of the WOST ............................................... 165

5.7 Procedure ................................................................................................. 170

5.8 Ethics ........................................................................................................ 173

5.9 Analytical Strategy .................................................................................... 176

5.9.1 Introduction ......................................................................................... 176

5.9.2 Primary and Secondary schools ......................................................... 176

5.9.3 What are Multi-Level Models? ............................................................ 177

5.9.4 How to interpret the outcome: Key concepts in Multi-Level Models ... 180

5.10 Chapter Summary ................................................................................... 183

CHAPTER 6 6.1 Introduction to chapter .............................................................................. 184

6.2 Missing Data ............................................................................................. 184

6.3 Data Assumptions for Multi-Level Models ................................................. 186

6.4 Risk and Promotive Factors ...................................................................... 193

6.4.1 Introduction to the section .................................................................. 193

6.4.2 Descriptive Statistics .......................................................................... 193

6.4.3 Research question 1a......................................................................... 197

6.4.4 Research Question 1b. ....................................................................... 199

6.4.5 Research Question1c ......................................................................... 216

6.4.6 Summary Statements ......................................................................... 218

6.5 Cumulative effects of risk factors: ............................................................. 219

6.5.1 Introduction to section ........................................................................ 219

6.5.2 Cumulative Risk Score ....................................................................... 219

6.5.3 Descriptive Statistics .......................................................................... 221

6.5.4 Research Question 2a ........................................................................ 222

6.5.5 Research Question 2b ........................................................................ 224

6.5.6 Research Question 2c. ....................................................................... 228


6.6 Protective Factors ..................................................................................... 232

6.6.1. Introduction to section ....................................................................... 232

6.6.2 Research Question 3a ........................................................................ 233

5

6.6.3 Research Question 3b ........................................................................ 239


CHAPTER 7 7.1 Introduction to the Chapter ....................................................................... 246

7.2 Summary of results ................................................................................... 246

7.2.1 Research Question 1 .......................................................................... 246



7.3 Discussion of results in relation to previous literature. .............................. 249

7.3.1 Research question 1a: ....................................................................... 249

7.3.2 Research question 1b: ....................................................................... 251

7.3.3 Research question 1c: ........................................................................ 271

7.3.4 Research question 2a: ....................................................................... 273

7.3.5 Research question 2b ........................................................................ 275

7.3.6 Research question 2c ......................................................................... 277

7.3.7 Research Questions 3a and 3b .......................................................... 279

7.4 Limitations ................................................................................................. 285

7.3.1. Methodological limitations ................................................................. 285

7.3.2. Conceptual limitations ....................................................................... 293

7.5 Future research ........................................................................................ 297

7.6 Chapter summary ..................................................................................... 304

CHAPTER 8 8.1. Introduction to chapter ............................................................................. 305

8.2 Implications ............................................................................................... 305

8.2.1 Implications from Research Question 1 .............................................. 305



8.2.4 Summary statements of key implications ........................................... 312

8.3 Contribution to Knowledge ........................................................................ 314

8.4 Summary of study ..................................................................................... 321

8.5 Summary Statements ............................................................................... 323

REFERENCES…………… ………………………………………………………324

APPENDICES………………………………………………………………………352

Word Count 86,663

6

LIST OF TABLES

Table 1.1 Percentage of SEND by Primary need

44

Table 4.1: Examples of protective factors that foster resilience for children at risk

124

Table 5.1 An overview of how the present study differs from the AfA evaluation

142

Table 5.2 Pupil level predictor variables: descriptions and sources of data collection

152

Table 5.3 School level predictor variables: descriptions and sources of data collection

154

Table 5.4 Participant attrition rates by total sample and school type

157

Table 5.5 Sample proportion within each group of the categorical predictor variables at the individual and school level (primary schools)

159

Table 5.6 Means and standard deviations of continuous predictor variables, with national averages and effect size comparisons (primary schools)

160

Table 5.7 Sample proportion within each group of the categorical predictor variables at the individual and school level (secondary schools)

161

Table 5.8 Means and standard deviations of continuous predictor variables, with national averages and effect size comparisons (secondary schools)

162

Table 5.9 Confirmatory factor analysis fit indices for the WOST

167

Table 6.1 Summary of data assumptions and requirements

192

Table 6.2 Mean and standard deviations for the behaviour difficulties score at baseline and follow up within the primary and secondary data sets

193

Table 6.3 Bivariate correlations between the behaviour difficulties mean score (at baseline and follow up) and all predictor variables within the primary and secondary data sets

194

Table 6.4 Empty multi-level model for primary school data

197

Table 6.5 Empty multi-level model for secondary school data 198

7

Table 6.6a Full multi-level model for primary school data

201

Table 6.6b Full multi-level model for primary school data (SEND category excluded)

203

Table 6.7a Full multi-level model for secondary school data

208

Table 6.7b Full multi-level model for secondary school data (SEND category excluded)

210

Table 6.8 Comparison between empty and full model for primary school data

216

Table 6.9 Comparison between empty and full model for secondary school data

217

Table 6.10 Total number and percentage of participants per each risk group for the primary and secondary school models

221

Table 6.11 Mean cumulative risk scores for the primary and secondary models

221

Table 6.12 Cumulative risk multi-level model for the primary school data

222

Table 6.13 Cumulative risk multi-level model for the secondary school data

223

Table 6.14 The mean and standard deviation of behaviour difficulties at baseline and follow up for each risk group

224

Table 6.15 Empty, Cumulative and Quadratic multi-level model for primary school data

226

Table 6.16 Empty, Cumulative and Quadratic multi-level model for secondary school data

227

Table 6.17 The independent additive model for the primary school data

229

Table 6.18 The independent additive model for the secondary school data

230

Table 6.19 Protective factor model for primary school data

234

Table 6.20 Protective factor model for secondary school data 236

8

LIST OF FIGURES

Figure 4.1 An example of a compensatory model

119

Figure 4.2 An example of a protective-stabilising model

121

Figure 4.3. An example of a protective-reactive model

122

Figure 5.1

Bronfenbrenner’s Ecological Systems Framework 134

Figure 5.2

Predictor variables within the present study incorporated into Bronfenbrenner’s Ecological Systems Framework

137

Figure 5.3 Achievement for All diagram

139

Figure 5.4 An example of a two-level data set

178

Figure 6.1 Venn diagram displaying risk factors for primary and secondary schools

215

Figure 6.2 Venn diagram displaying promotive factors for primary and secondary schools

215

Figure 6.3 Protective factor interaction graphs for primary schools (school level academic achievement)

240

Figure 6.4 Protective factor interaction graphs for primary schools (school percentage of pupils at school action)

241

Figure 6.5 Protective factor interaction graphs for secondary schools (school urbanicity)

243

9

ABSTRACT Jeremy Oldfield University of Manchester September 2012 Behaviour difficulties in children with special education needs and disabilities: assessing risk, promotive and protective factors at individual and schools levels

Behaviour difficulties displayed in childhood and adolescence have pervasive and long term effects into adulthood and across various domains of functioning (Healey, et al. 2004, Woodward, et al. 2002). The numbers of children who suffer with them remain worryingly high (Green et al. 2005). Children with special educational needs and disabilities (SEND) are considered particularly at risk of displaying behaviour difficulties; however, despite representing around one fifth of the school population (DfE 2011), little research to date has explicitly investigated these problems in this population.

The present study therefore aimed to investigate risk, promotive and protective factors for behaviour difficulties in children with SEND across multiple ecological levels. Data were collected through a concurrent research project evaluating Achievement for All (Humphrey et al. 2011). The sample comprised children identified with SEND in years 1, 5, 7 and 10, from ten local authorities deemed representative of England. A final sample consisted of 2660 primary pupils nested in 248 primary schools and 1628 secondary pupils nested within 57 secondary schools. Predictor variables were measured at the individual and school levels at baseline, along with a teacher reported measure of behaviour difficulties which was assessed again eighteen months later.

Analyses were carried out using multi-level modelling revealing that primary schools accounted for 15% and secondary schools 13% of the total variance in behaviour difficulties, with the remainder being at the individual level. Significant risk factors for these problems across both school types were: being male; eligibility for FSM; and being a bully. Risk factors specific to primary schools included being autumn born, being older in the school, having poor positive relationships, and attending schools with lower levels of academic achievement. Risk factors specific to secondary schools included being younger in the school, having poor attendance, having poor academic achievement, being a bystander to bullying and attending a larger school.

Results showed evidence for a cumulative risk effect that increasing numbers of contextual risk factors, regardless of their exact nature, resulted in heightened behaviour difficulties. This relationship was non-linear with increasing risk factors in an individual’s background having a disproportional and detrimental increase in behaviour difficulties displayed. The specific type of risk was however, more important than number of risk factors present in an individual’s background in accounting for behaviour difficulties displayed.

Finally, results revealed significant protective factors at the school level; specifically attending primary schools with high academic achievement and with more children on the SEND register at school action can protect against the display of behaviour difficulties when these children are at risk in terms of having poor positive relationships. Attending urban secondary schools can also protect against the display of behaviour difficulties, when these children are at risk in terms of having poor academic achievement. The implications of these findings are discussed along with directions for future research.

10

DECLARATION

No portion of the work referred to in the thesis has been submitted in support of an application for another degree or qualification of this or any other university or other institute of learning.

COPYRIGHT STATEMENT

The following four notes on copyright and the ownership of intellectual property rights must be included as written below:

i. The author of this thesis (including any appendices and/or schedules to this thesis) owns certain copyright or related rights in it (the “Copyright”) and s/he has given The University of Manchester certain rights to use such Copyright, including for administrative purposes.

ii. Copies of this thesis, either in full or in extracts and whether in hard or electronic copy, may be made only in accordance with the Copyright, Designs and Patents Act 1988 (as amended) and regulations issued under it or, where appropriate, in accordance with licensing agreements which the University has from time to time. This page must form part of any such copies made.

iii. The ownership of certain Copyright, patents, designs, trade marks and other intellectual property (the “Intellectual Property”) and any reproductions of copyright works in the thesis, for example graphs and tables (“Reproductions”), which may be described in this thesis, may not be owned by the author and may be owned by third parties. Such Intellectual Property and Reproductions cannot and must not be made available for use without the prior written permission of the owner(s) of the relevant Intellectual Property and/or Reproductions.

iv. Further information on the conditions under which disclosure, publication and commercialisation of this thesis, the Copyright and any Intellectual Property and/or Reproductions described in it may take place is available in the University IP Policy (see http://documents.manchester.ac.uk/DocuInfo.aspx?DocID=487), in any relevant Thesis restriction declarations deposited in the University Library, The University Library’s regulations (see http://www.manchester.ac.uk/library/aboutus/regulations) and in The University’s policy on Presentation of Theses.

11

ABBREVIATIONS

ADHD: Attention Deficit Hyperactivity Disorder

AfA: Achievement for All

APP: Assessing Pupil Progress

ASD: Autistic Spectrum Disorder

BESD: Behaviour Emotional and Social Difficulty

CD: Conduct Disorder

DfE: Department for Education

DfES: Department for Education and Skills

DCSF: Department for Children Schools and Families

DSM-IV-TR: Diagnostic and Statistical Manual of Mental Disorders IV Text Revision

EAL: English as an Additional Language

EBD: Emotional Behavioural Difficulty

FSM: Free School Meals

HI: Hearing Impairment

ICC: Intraclass Correlation Coefficient

ICD-10: International Classification of Diseases 10

IQ: Intelligence Quotient

LA: Local Authority

SEN: Special Educational Needs

SENCO: Special Educational Needs Co-ordinator

SEND: Special Educational Needs and Disabilities

SA: School Action

SAP: School Action Plus

SEAL: Social Emotional Aspects of Learning

12

SEL: Social Emotional Learning

ST: Statement of SEND

MCAR: Missing Completely At Random

MLD: Moderate Learning Difficulty

MLM: Multi-level Modelling

MSI: Multi-Sensory Impairment

NPD: National Pupil Database

ODD: Oppositional Defiant Disorder

PATHS: Promoting Alternative Thinking Strategies

PD: Physical Disability

PMLD: Profound and Multiple Learning Difficulty

PRU: Pupil Referral Unit

SES: Socio-Economic Status

SpLD: Specific Learning Difficulty

SLCN: Speech, Language and Communication Needs

SLD: Severe Learning Difficulty

VI: Visual Impairment

WOST: Wider Outcome Survey for Teachers

13

ACKNOWLEDGEMENTS

Thanks go to my first and second supervisors; Prof. Neil Humphrey and Dr Michael Wigelsworth, who have supported and advised me throughout this project. Thanks also to Dr Alex Barlow for help with organising the data and Dr Ann Lendrum for her advice in writing a PhD.

I would also like to acknowledge the contribution that my two colleagues have made to my thesis; Judith Hebron and William Bulman. Without them the thesis would be greatly impoverished.

Finally, thanks go to my Mum and Dad, family and friends for their constant encouragement. Also to God for guiding me through these three years, without him I couldn’t have written this thesis and would be completely lost. It’s true I can do all this through Him who gives me strength, (Philippians 4 verse 13).

14

1. Key Concepts: Behaviour Difficulties; Special Educational Needs and Disabilities;

Risk, Promotive, and Protective factors.

1.1 Introduction to chapter

The aim of this chapter is to outline the key concepts within the present study.

These are: section 1.2 behaviour difficulties; section 1.3 risk, promotion and

protection; and section 1.4, special educational needs and disabilities.

Section 1.2.1 begins by discussing the various terminology and definitions that

have been used to describe a child or adolescent with a behaviour difficulty.

This section acknowledges the considerable inconsistency within the field and

how any definition or terminology applied reflects the researcher’s professional

background. A conclusion is then reached on an appropriate terminology and

definition that will be used throughout the remainder of the present study.

Within section 1.2.2, there is a discussion of current prevalence rates and time

trends within these problems and in section 1.2.3, the associated outcomes for

children with behaviour difficulties, both immediate and longer term are

highlighted. 1.2.4 completes this section with summary statements.

Section 1.3.1 begins with a brief overview 1

1 It is beyond the scope of this study to discuss all the complexities and contradictions within this term SEND. It is considerably more complex than this brief overview can give credit for, and so for a more in-depth review of the issues and contentions, the reader is referred to Peer & Reid (2012), Armstrong & Squires (2012), and Warnock, Norwich & Terzi (2010).

of the concept of Special

Educational Needs and Disabilities (SEND). Within this section a brief

description of the study’s population is provided in terms of: a) definitions, b)

historical overview of the concept, c) types or categories of specific need, d)

identification and assessment strategies, e) levels of support or provision, and f)

15

prevalence rates. Section 1.3.2 justifies the use of a SEND population within the

present study and section 1.3.3 provides summary statements for this section.

Section 1.4.1 begins with a discussion of the concept of risk and the inception

of risk factor research. Definitions are provided and discussion is undertaken of

the key theoretical and methodological issues which are essential in its

understanding. Section 1.4.2 discusses the concepts of promotion and

protection, highlighting the important differences between these terms and how

they will be conceptualised within the present study. Finally section 1.4.3

provides summary statements to this section.

16

1.2 Behaviour problems

1.2.1 Definitions and terminologies Psychologists have commonly referred to childhood and adolescent behaviour

problems as ‘acting out’ or ‘externalising behaviours’ that have disruptive and

disturbing affects on others (Papatheodorou, 2005). The types of behaviours

incorporated within this concept can include physical and verbal abuse,

vandalism, lying, fighting and stealing (Goodman, 2001). Specifically, in

schools, these behaviours could involve persistent low-level disruption - such as

students talking when not authorised to do so, avoiding and preventing others

from working, challenging the authority from teachers, making inappropriate

remarks, or being defiant, rowdy and generally disobedient (DES 1989, Cooper

1999a, Long, Wood, Littleton, Passenger & Sheehy 2011).

There has been considerable study of these types of problems, which have

often fallen under the broad umbrella term behaviour problems. Nonetheless,

despite an abundance of research, it remains a vague concept. There have

been substantial variations in how researchers have defined it, and also major

differences in the terms and precise behaviours used within this research

(Ayers, Clarke & Murray 2000). An established and universally agreed upon

definition of what behaviour problems actually means, has not been forthcoming

within the literature, and appears to elude current research (Howarth & Fisher

2005).

A rich tapestry of concepts and definitions has been applied to this field (Hawes

& Dadds 2005), which makes it a particularly difficult area to research.

Conducting even the briefest literature search within this field reveals a number

of distinct terms that appear to be used somewhat interchangeably. It is often

unclear, for example, whether terms such as ‘behaviour problems‘; ‘behaviour,

emotional social difficulties’; ‘externalising problems’; ‘challenging behaviour’;

‘anti-social behaviour’; ‘conduct problems’; and ‘aggression’ to name but a few,

have synonymous meanings, and can therefore be used interchangeably in

17

discussing research on childhood and adolescent behaviour problems. Kavale,

Forness & Alper (1986) reported 25 years ago, in a review of relevant studies,

that approximately fifty terms have been used to describe ‘behaviour problems’.

This number is only likely to have increased in recent years, as different

terminology is introduced to fit with current government and research aims. The

terms appear to share some commonality in the types of behaviours they

encompass; nonetheless, a closer look reveals that each has its own distinct

definition which means they are not completely interchangeable.

The huge variations in the terminology used within this field ultimately reflect the

professional background of the researcher in question. Different disciplines

describe children with behaviour problem using their own distinct terminologies

and definitions (Connor, 2004, Munn & Lloyd 1998). This can be a hindrance for

professionals across the disciplines particularly in terms of how they accurately

communicate to one another regarding a certain child with a particular difficulty

(Papatheodorou, 2005). Visser (2002) argues that it is even questionable

whether professionals within the same field, let alone those across different

disciplines, understand and apply these terms in the same way. The use of the

concept challenging behaviour (Ofsted 2005) is one such example. It was

reported there is a lack of an agreed definition of this concept by schools that

have used and understood this term in various ways across settings. In order to

avoid further confusion concerning terms and definitions used to describe

behaviour problems it is essential to have an explicit definition of behaviour

problems (or whichever other term is chosen for the research), (Visser & Stokes

2003).

The present study will adhere to this principle, and provide both an

acknowledgment of the term chosen to describe behaviour problems, and also

a definition of this term. It should be pointed out here that adopting the term

behaviour problems to describe the types of behaviour of interest has been

dismissed, as although a fairly self-explanatory term, theoretically, it is

problematic. Psychologists have long debated over a viable definition of

‘behaviour’, and there is controversy as to whether ‘behaviour’ refers uniquely to

responses that are overt and objectively measureable, or whether it also

18

includes internal states and mental constructs/representations and even

physiological and neurological processes (Reber & Reber 2001). It is logical to

suggest that adding the word ‘problem’ to this conceptualisation of behaviour

could imply a number of clearly distinct and vastly different types of behaviours

that might not resemble what is assumed to be behaviour problems. As such it

becomes extremely difficult to define it in a meaningful or constructive way, and

is therefore perhaps not the most useful term to employ for the present

research.

Nonetheless, an appropriate term needs to be found and applied within this

study. By reviewing other terms applied to this field and across a range of

different disciplines, greater understanding of behaviour problems, will be

realised (Connor, 2004), thus allowing the most suitable term to be settled

upon. What follows then is a discussion of the various terminologies and

definitions relevant to the field of behaviour problems that have been utilised

within educational, community and clinical settings. A conclusion will then be

drawn as to which terminology and definition is most relevant and whether it is

appropriate to use such a concept within the present study.

Behaviour problems defined within educational settings It is often within schools where behaviour problems are most commonly

witnessed and where the issues are perhaps the most salient. Examining how

professionals within education contexts have identified the problem and

whether terminology applied within these settings could be adopted by the

present study is important to note.

Over the past 75 years, various terminology and definitions have been applied

to children and adolescents within educational settings, who present some

difficulty with their behaviour. The term ‘Maladjustment’ was first used to

describe such problems in Britain around the time of the 1944 Education Act.

This term was considered an official way to describe children with behaviour

problems or those who displayed evidence of some kind of psychological

disturbance or emotional instability, and therefore demanded special

19

educational provision in order for them to become ‘readjusted’ (Papatheodorou

2005, Visser 2002). There was criticism of this term, however, as it included no

exact definition or identifying criteria and only the children most severely

affected were acknowledged, whereas less serious cases, which were in fact

more common, were not. Furthermore, individual circumstances and contexts

were often not taken into account (Papatheodorou 2005), and the term reflected

a commonly held view at the time that these behaviours stemmed mainly from

within the child, and were not influenced by environmental conditions. Around

40 years later, the term lost favour when the Warnock report came out in 1978,

(DES 1978) and the Education Act was published in 1981. These reports

advocated the term ‘emotional or behavioural disorders’ which soon became

Emotional Behavioural Difficulties (EBD), (Frederickson & Cline 2009).

The term EBD has been defined in various ways, The DfE (1994) in Circular

9/94, entitled ‘Pupils with Problems: the Education of Children with Emotional

and Behavioural Difficulties’ proposed an all-encompassing definition that

“children with emotional behavioural difficulties are on a continuum. Their

problems are clearer and greater than sporadic naughtiness or moodiness and

yet not so great as to be classed as mental illness…EBD range from social

maladaption to abnormal emotional stresses. They are persistent (if not

necessarily permanent) and constitute learning difficulties”, (page 7). EBD may

not be seen as a single disorder, but rather as a diverse collection of

characteristics. These may have arisen from either within the child, from a

response to their environment, or result in an interaction between the two

(Cooper 1999b). This definition is problematic however, as it is unclear when

‘sporadic naughtiness or moodiness’ become EBD, or even when a child’s

behaviour is bordering on mental illness (Chaplain 2003). Cooper (1996) also

criticises the definition of EBD describing it as a “crude, ill-defined and not very

useful descriptor. Current definitions of the terms are vague and undeveloped.”

(page 147).

The DfE (1994) state that “Whether the child is judged to have EBD will depend

on the nature, frequency, and persistence, severity, abnormality or cumulative

effect of the behaviour compared with normal expectations of a child of the age

20

concerned. There is no absolute definition”, (page 4). With no conclusive

definition of what EBD actually is, and the fact that it incorporates a wide range

of behaviour problems such as social withdrawal, school refusal, depression,

and anxiety, as well as more externalised behaviours such as being non-

cooperative and oppositional, talking when not authorised to do so, challenging

authority of the teacher, and preventing others from working, (Cooper 1999a), it

is not the most useful term that can be applied to the present study.

The publication of the Special Educational Needs Code of Practice (DfES

2001a) brought with it the term behaviour, emotional and social difficulties

(BESD). Schools within the UK use the Special Education Needs code of

practice (DfES 2001a) in order to define and identify those children and

adolescents with behaviour problems - placing them in the BESD group. BESD

refers to “Children and young people who demonstrate features of emotional

and behavioural difficulties, who are withdrawn or isolated, disruptive and

disturbing, hyperactive and lack concentration; those with immature social skills;

and those presenting challenging behaviours arising from other complex special

needs” (DfES 2001a, page 87). A more recent government report entitled ‘The

Education of Children and Young People with Behaviour, Emotional and Social

Difficulties as a Special Educational Need’ (DCSF 2008a), also suggests that

the term BESD includes children with conduct related disorders, hyperkinetic

disorders (e.g. attention deficit disorder/attention deficit hyperactivity disorder),

as well as those with emotional disorders such as depression, anxiety, school

phobia and those that self-harm. Children within this category have not

necessarily been given a medical diagnosis but those with this diagnosis are

likely to be defined as having BESD by their school. The report also suggests

that children will be considered to have BESD not only due to the nature and

severity of the problems but also frequency and persistence of the difficulties.

The BESD group includes a very heterogeneous sample of children,

incorporating pupils with a broad range of disabilities across a wide continuum

of severity. The BESD label makes no distinction between those individuals who

display aggressive, disruptive, attention-seeking, hyperactive, delinquent and

oppositional behaviours or so-called externalising problems (Achenbach 1991a,

21

1991b, Liu 2004) and those children that display anxious, depressed,

unmotivated, withdrawn behaviours and psychosomatic complaints or so called

internalising problems, (Achenbach 1991a, 1991b, Eisenberg et al. 2001). As

the BESD category comprises at least two distinct groups (externalising versus

internalising problems) it is not an adequate term and its definition cannot be

applied for research that is interested in exclusively the externalising type of

behaviour problems.

Another term commonly found within the research literature which assumes a

synonymous meaning to BESD is ‘social emotional behavioural difficulties’

(SEBD), a term used by the SEBD association (SEBDA). There is some

discussion from this organisation in terms of the exact order of the words

‘social’, ‘emotional’ and ‘behaviour’. Cole (2003) suggests that it is ultimately

social and emotional experiences interacting with environment conditions that

affect behaviour, therefore ‘social’ and ‘emotional’ should come before

‘behaviour’. Although the use of ‘social’ is included in the BESD definition this

has not been broadly used in the research literature (Visser 2003).

Behaviour problems defined within community settings It is within schools that terms such as EBD and BESD have been used. Outside

these settings, within the community, they could be referred to and may have

equivalent meaning as anti-social behaviour (Chaplain 2003). Indeed, Reber &

Reber (2001), in their definition of behaviour problems, argue, for the concept of

anti-social behaviour - defining an individual with behaviour problems as “a

person usually a child or adolescent, whose behaviour is persistently anti-

social” (page 85). According to the Home Office a definition of anti-social

behaviour is “aggressive, intimidating or destructive activity that damages or

destroys another person's quality of life” (Home Office 2011, page 1). This

definition appears to convey a similar meaning to what is understood by

behaviour problems within schools. Specifically antisocial behaviours include

behaviours more frequently, although not exclusively, displayed outside the

school environment e.g. threatening behaviour, verbal abuse, vandalism,

graffiti, and intimidating behaviour, (Millie, Jacobson, McDonald & Hough 2005).

22

There are problems however, with using this term to describe behaviour

problems within the context of the present study. According to Hagell (2007),

anti-social behaviour is socially determined, and if a particular behaviour does

not impact negatively on another person then it is not seen as anti-social. This

idea could be problematic in terms of cheating behaviour, where it does not

necessarily have a direct impact upon others, but nevertheless is a clear

problem behaviour. Furthermore, what one individual interprets as anti-social

behaviour may be considered as appropriate behaviour by a different person,

as what is understood to comprise ‘anti-social behaviour’ is ultimately a product

of contextual factors, the tolerance of the community, location and the quality of

life expectations (Home Office 2004). There is a great deal of subjectivity

therefore in how it is defined and what specific behaviours are included within it.

The term antisocial behaviour also appears to relate more specifically to adults

than children and young people, who are the focus of the present study.

Although anti-social behaviour may be a term used in community settings and is

at least partially consistent with EBD or BESD in schools, it is not an

appropriate term for the present study that is interested in school behaviour

problems.

Delinquent behaviour is another descriptor that has been used to account for

children and adolescents with behaviour problems, (Visser 2002). There is

however, uncertainty as to when behaviour is considered delinquent or when it

is called anti-social. There seems to be some cross-over between these

constructs. There is further confusion, as highlighted by Campbell, Shaw &

Gilliom (2000), who argue that the term delinquency is flawed as it is often used

interchangeably for children who have been arrested for anti-social acts, as well

as being a clinical label used to describe certain behaviours. Many children

given the label delinquent have never been arrested, and those children that

have been arrested for anti-social acts and called delinquent in the legal sense

do not always meet the clinical criteria to gain the label in this domain.

Definitions of delinquency therefore relate to illegal acts and, sometimes, clinical

labels for behaviour problems. They are not suitable terms to use for the

present study that is investigating behaviour problems in schools, and aims to

23

incorporate a range of problematic behaviours from the less severe to the more

extreme.

Behaviour problems defined within clinical settings Professionals from clinical approaches often use the term mental disorders to

categorise children with behaviour difficulties into different groups. These

categorisations often depend upon the severity of their disabilities displayed.

The mental disorders that are relevant to the field of behaviour problems in

childhood include so called disruptive behaviour disorders such as oppositional

defiant disorder (ODD), conduct disorder (CD) and attention deficit hyperactivity

disorder (ADHD). Clinical professionals either use the Diagnostic and Statistical

Manual of Mental Disorders IV Text Revision (DSM–IV-TR, American

Psychiatric Association 2000) or the International Classification of Diseases 10

(ICD 10, World Health Organisation 1992) to diagnose these problems. The

manuals provide standardised frameworks for identifying and categorising

mental disorders, based upon specific behaviours displayed.

The DSM-IV-TR therefore defines ODD as “a recurrent pattern of negativistic,

defiant, disobedient, and hostile behaviour toward authority figures that persists

for at least 6 months” (page 100). CD (considered more severe than ODD) is

defined as “repetitive and persistent pattern of behaviour in which the basic

rights of others or major age-appropriate societal norms or rules are violated”

(page 93) and ADHD is defined as “persistent pattern of inattention and/or

hyperactivity-impulsivity that is more frequently displayed and is more severe

than is typically observed in individuals at comparable level of development”

(page 85).

Of particular relevance to the present study are behaviours associated with CD

and ODD. The specific types of behaviours included in the CD classification fall

in to 4 broad areas; these have been termed aggression to people and animals;

destruction of property; deceitfulness or theft; and serious violations of rules

(DSM-IV-TR 2000). Individuals who display these problem behaviours, although

to a lesser degree, may meet the criteria for ODD, or if they do not meet the

24

diagnostic criteria for either of these disorders, but display some of the

behaviours listed, they could be described as having or displaying conduct

problems. Although these terms have been clearly defined and contain details

of specific behaviour that are similar to those of interest within the present

study; using them is not a feasible option for an educational focussed study that

does not have the expertise in using the DSM-IV-TR or ICD-10 correctly.

A further alternative label often used within clinical settings is externalising

behavioural disorders (King, Iacono, & McGue 2004). This phrase is often used

to refer to those who display oppositional, aggressive, and destructive

behaviours (Calkins, Blandon, Wiliford & Keane 2007). This label maybe a

useful means of explaining behaviour problems within the present study, as the

term refers solely to those children who have problems in ‘acting out’

behaviours and not those with emotional difficulties such as depression and

anxiety (i.e. internalising problems). This concept therefore overcomes the

inherent problems with the definitions such as BESD, as it is able to distinguish

externalising from internalising behaviours. Nonetheless, some researchers

have argued that these constructs may not be mutually exclusive categories

and are often comorbid (Gilliom & Shaw 2004).

There are concerns with using the term externalising problems for the present

study; as Fergusson, Horwood & Ridder (2007) argue that this term can be

broken down or subdivided into conduct problems such as aggressive

behaviours (which would include CD and ODD) and attentional problems, such

as overactivity (which would include ADHD). The focus of the present study is

exclusively on behaviours making up the conduct component, therefore the

broader term externalising problems is not appropriate here.

The conduct construct remains the primary interest of the present study;

however, this term can also be further subdivided to include overt aggressive

behaviours, such as fighting and cruelty, and covert delinquent behaviours such

as lying, cheating or stealing (Barnow, Lucht & Freyberger 2005). Both of these

sub-domains are of interest to the present study. However, using such a term

as conduct within the present study was dismissed due to its inherent

25

association with clinical settings whereas the focus of the present study remains

an educational based piece of research.

Conclusions The amount of terminologies and definitions applied to this field of behaviour

problems is enormous, and “the diversity in this language reflects not only the

historical trends, but also conceptual distinctions between and within

disciplines” (Hawes & Dadds, 2005, page 74). Although there has been some

consistency between those from educational, community and clinical

backgrounds in their interest in, and how, they define behaviour problems

(Visser 2003), the exact terminology often varies significantly.

The lack of consistency in terminology used across studies will have a

detrimental effect on the search for new knowledge in this area. New research

should therefore clearly and coherently state the terminology and definition

applied at the outset. In making these explicit, results from these studies will be

interpreted more accurately allowing knowledge to advance further in this area.

Having reviewed the various terminologies and definitions found within the

literature, the term ‘behaviour difficulties’ will be adopted by the present study,

and will be used throughout the remainder of this report. Other terms signifying

similar meaning to behaviour difficulties will only be used when discussing

previous research that has defined the problem using that terminology.

Using the term behaviour difficulties was selected to bridge the gap between

different perspectives and incorporate terminology from educational concepts

such as BESD, clinical ones such as CD, and community ones such as anti-

social behaviour. Focusing upon the common features between these

approaches will allow the present study to have a broader focus, incorporating a

larger evidence base and be of greater benefit to children experiencing these

problems, both within and outside of education.

26

The choice of the word behaviour was selected, as discussed above it appears

to reflect a broadly understood construct. Although it could be interpreted in a

number of different ways, alongside a clear definition it will be a constructive

word to use. Furthermore, it also appears in a number of terms already used

within this area, such as emotional behaviour difficulties, and behaviour,

emotional and social difficulties, as well as anti-social behaviour. The word

difficulty was chosen over words such as problem and disorder, not only as

these words are not commonly used within the UK government’s educational

guidelines (Visser 2003), but because of their underlying inherent meaning.

According to Delfos (2004), using the word disorder signifies that the researcher

sees the source of a child’s behaviour coming to a greater extent from within the

person and is therefore predispositional. Whereas the word problem carries a

different value judgment - that the child’s behaviour stems more from

environmental factors outside of the child. The present study is not attempting,

as yet, to make any inferences as to where the main source of the behaviour in

question lies – whether it is predispositional, or environmental - but by utilising

the term difficulty it adopts an interaction perspective acknowledging the

importance of both aspects.

A definition of behaviour difficulties for the present study will therefore be:

An individual’s behaviour that is persistently aggressive, destructive,

dishonest or disobedient, and is deemed inappropriate for their age or

cultural background. These behaviours will have a negative affect upon

themselves, others or property, and may occur solely within a unique or

across multiple contexts.

Behaviour difficulties is used here as a kind of umbrella concept to describe a

number of behaviours that refer more generally to externalised behaviour

problems, rather than internalising ones, and within the externalising framework

refer more specifically to behaviours that are conduct-related such as physical

abuse and cheating (therefore including both covert and overt behaviours)

rather than inattentional problems such as hyperactivity.

27

This definition highlights that children will therefore vary in the degree to which

they are displaying behaviour difficulties, and does not categorise children as

always displaying a problem. This is because all children at some point will

display behaviour difficulties. They become more severe problems when they

are persistent and fairly common patterns in that individual’s behaviour, and not

appropriate for their age and cultural background.

This definition is attempting to provide a general overview of what behaviour

difficulties look like and the types of problem behaviours they encompass.

Highlighting the key behaviours that are associated with this term, will aid

researchers in adopting a more consistent approach, which is what Ayers, et al.

(2000) state that when defining behaviour difficulties “specific, descriptive

phrases should be used instead of vague colloquialisms” (page vii). Therefore

in relation to the present study and the above definition, the behaviours that are

of particular interest are disobedience, cheating and lying, stealing, physical

and verbal aggression and vandalism, all of which are measured within the

WOST survey, (see chapter 5). These types of behaviours are termed

behaviour difficulties as they have a significant and detrimental effect on the

individual child as well as other people’s general well-being.

Providing a single definition for any construct within the broad area of childhood

and adolescent behaviour problems has been argued against by some

researchers (Visser 2003, Munn & Lloyd 1998). They acknowledge that distinct

terms will inevitably be used in the literature and be understood in diverse ways

by different professionals (Daniels, Visser, Cole & DeReybekill 1999). If new

descriptive labels are sought and applied, researchers may begin to investigate

interventions for these new terms, when essentially they are referring to the

same children as before but under different labels. Important knowledge from

previous research may then be lost (Visser 2002). Furthermore, criticism has

been levelled against adding a new terminology to the literature base. Connors

(2004) argues that it is unlikely that any term will adequately and sufficiently

incorporate the wide range of behaviours generally included within this concept.

28

Nonetheless, despite these criticisms it is argued here that researchers need to

define the problem for their own research, so that there can be some

meaningful comparison between studies that is not based upon assumptions

and subjective interpretations of the various terminologies applied. If definitions

for behaviour difficulties have clarity of meaning, and a shared understanding,

by professionals working within the same field, as well as those across

disciplines, this will ultimately aid reliability and accuracy in reporting on, and

then supporting, children with these difficulties. Without a clear definition, there

is a danger that all children categorised as having some kind of behaviour

difficulty are assumed to be somewhat similar and therefore requiring

comparable interventions whereas in reality their problems could be quite

distinct, (Cooper 1996).

29

1.2.2 Prevalence rates and time trends in behaviour difficulties

Prevalence Establishing the current prevalence rates of childhood and adolescent

behaviour difficulties is a valuable exercise as they can help elucidate the

aetiological and potential protective factors involved (Achenbach, Dumenci &

Rescorla, 2003). In a report carried out on behalf of the Department of Health,

Green, McGinnity, Meltzer, Ford & Goodman (2005) showed the current

prevalence rates of childhood and adolescent mental health problems within the

UK to be around 10%. That is one in ten children and young people aged 5-16

had a clinically diagnosed mental disorder. Specifically, in terms of conduct

disorders they found a 6% diagnosis rate. The sample was split by age group

and it was found that in children (5-10 year olds) the rate of conduct disorders

was 4.9% whereas it was even higher, 6.6%, for adolescents (11-16 year olds).

These findings reflect similar figures within schools, where the percentage of

children categorised as BESD (and therefore likely to be displaying some

behaviour difficulties) is around 4.7% of the school population (DfE 2011a).

There is however, some confusion about the extent of childhood behaviour

difficulties. In a review of 52 studies investigating child and adolescent mental

disorders the prevalence rates ranged from 1-50% (Roberts, Attkisson &

Rosenblatt 1998). These authors suggested that the reason for this variation in

results was the studies in question used different methods and a range of

different definitions. It is important, therefore, that further prevalence studies (or

indeed any research around this topic) to have standardised assessment

measures with clear definitions, using representative samples across different

time points, to assess the reliability of whether behaviour disorders are in fact

increasing (Collishaw, Goodman, Pickles & Maughan, 2007).

30

Time trends Whether the prevalence of behaviour difficulties have increased in recent years

has attracted considerable attention. Indeed investigating the potentially

changing extent of childhood and adolescent behaviour difficulties across years

may help to establish causal factors or highlight clear associations between

them and aspects of a changing society.

Smith & Rutter (1995) conducted a review of the literature spanning 50 years

and argued that there had been a considerable increase in the number of young

people affected by psychosocial disorders. This evidence has more recently

been supported by Collishaw, Maughan, Goodman & Pickles (2004) who

collated data spanning 25 years, from 1974-1999 within the UK, and showed

that mental health problems in adolescence, especially conduct problems, had

substantially increased, and this effect was evident across both genders and

was not related to family type or social class. Further support for this view

comes from outside the UK, where Tick, van de Ende, & Verhulst (2007), who

investigated trends in emotional and behavioural problems in Dutch children

across 20 years, found small increases in children’s emotional and behavioural

problems across the years.

Not all evidence is in support of this view, as Green et al.’s (2005) study looked

at the changes between 1999 and 2004, finding little difference in prevalence of

conduct disorders between those years. This evidence has been further

supported by Sourander et al. (2004), who conducted a 10-year comparison

study investigating time trends in mental health problems among children by

taking a representative sample of 8-9 year olds from Finland in 1989 and then

again in 1999. A significant finding to emerge was that among boys, the 1989

sample showed higher levels of conduct problems than the 1999 sample (for

both the teacher and parent report). However, teacher and parent report for

girls’ conduct problems over this period remained constant with no significant

change over time.

Achenbach, Dumenci & Rescorla (2002, 2003), showed that in samples of 11-

18 year olds a small decline in psychological problems was evident between

31

1989 and 1999. Specifically, they demonstrated that in samples of three cohorts

of 7-16 year olds, in 1976, 1989 and 1999, problem behaviour scores had

increased from 1976 to 1989 but decreased again in the 1999 sample (although

not to the same levels as in 1976). These studies suggest that externalising

problems had risen between the late 1970s and 1980s but had fallen again by

late the 1990s. Finally, Maughan, Collishaw, Meltzer & Goodman (2008), using

data collected from two nationally representative samples in 1999 (Ford et al.

2003) and 2004 (Green et al. 2005) found that the differences were marginal

between the year groups. Teachers reported that conduct problems remained

at the same level in both cohorts, whereas parents reported a slight drop from

the 1999 to the 2004 sample, which was a similar finding for the adolescent

self-report. The authors of this study suggested that the initial increase in

conduct problems seen before and throughout the 1980s may have plateaued

and possibly reversed.

Conclusions In concluding this research, it is important to note that there are difficulties in

comparing prevalence studies of childhood and adolescent mental health

problems, as there is often a wide variation in sampling, measurements used,

diagnostic criteria/definitions applied, how studies have combined informant

reports, and distinct types of disorder investigated, (Ford, Goodman & Meltzer,

2003). Furthermore, the changing levels of behaviour difficulties may be due to

an increased awareness or people being more willing to report on these

problems. These ideas should be taken into account before coming to a firm

conclusion.

Overall, although there is some evidence that conduct-related problems appear

to be decreasing or at least plateauing, concern remains that the number of

children and adolescences with behaviour difficulties remains worryingly high.

Further research, which can attempt to explain why this may be the case -

especially in terms of the prominent risk, promotive and protective factors

relevant to this current time period is therefore justified.

32

1.2.3 Outcomes associated with childhood behaviour difficulties

Immediate term Childhood and adolescent behaviour difficulties remain a particular concern due

to the various negative outcomes associated with them. These behaviours have

immediate and profound influences in schools, especially on the learning

environment, achievement and children’s social development, (Calkins, et al.

2007). It has been reported that children with behaviour difficulties are more

likely to be involved in bullying both as a victim (Humphrey et al. 2011) and

perpetrator (Laukkanen et al. 2002), have poorer positive relationships

(Humphrey et al. 2011) and perform less well academically (Humphrey et al.

2011). Childhood and adolescent behaviour difficulties also causes significant

stress to teachers (Chaplain 2003, Cooper 1999b) who may then perceive

these children more negatively resulting in the child acting in line with their

teachers negative expectations (Soles, Bloom, Heath & Karagiannakis 2008).

The effect of these types of behaviour also causes more stress and conflict with

parents (Hastings 2002), and can have significant detrimental affects when

displayed in the wider community, particularly if expressed as vandalism and

arson (Delfos 2004).

Long term These difficulties not only have immediate impacts on the individual displaying

them, their family, school and wider community but also often lead to various

deleterious outcomes later in life, (Fergusson et al. 2007). This is because

childhood behaviour difficulties are generally considered to remain relatively

stable and persist into adulthood, displaying long-term continuity at least up to

middle adulthood (Reef, Diamantopoulou, van Meurs, Verhulst & Ende, 2010).

Research has therefore begun to explore the long-term outcomes that are

associated with behaviour difficulties in childhood.

• Cost to society: Scott, Knapp, Henderson & Maughan (2001) argue that

behaviour difficulties in childhood are a powerful predictor of how much an

individual will cost society. They found that those individuals at aged ten who

33

had been rated as having the most severe behaviour problems had cost

society 10 times more by the time they were twenty-eight, compared against

those rated as having no behaviour difficulties in childhood.

• Unemployment: Healey, Knapp & Farrington, (2004), have suggested

that those children who were reported to have behaviour difficulties (and then

engaged in criminal activity throughout their adolescence), were more likely

to suffer long term unemployment prior to age 32. Unemployment may

however, have been mediated through low academic attainment at

secondary school, and being convicted of criminal acts in early adulthood.

• Adult mental health problems: Research has shown that behaviour

difficulties in childhood are a strong predictor of an increased risk of suicide,

(Darke, Ross & Lynskey, 2003) and increased risk of experiencing a

psychiatric disorder in early adulthood (Sourander et al. 2005).

• Crime: Fergusson, Horwood & Ridder (2005) investigated the

relationship between behaviour difficulties in middle childhood and later

psychosocial outcomes into adulthood across a 25 year period. They found

associations between behaviour difficulties in childhood and later criminal

behaviour, (individual, family or social factors did not confound these

negative outcomes in adulthood). Other studies have found an association

with childhood behaviour difficulties and substance abuse, (Flory, Milich,

Lynam, Leukefeld & Clayton 2003).

• Poor relationships: Associations between childhood behaviour difficulties

and poor adult relationships such as more negative romantic relationships in

adulthood have been established. These relationships are often

characterised by violence, both in terms of an increase in being the victim as

well as the perpetrator of the violent acts, (Woodward, Fergusson &

Horwood, 2002).

34

From the research highlighted above there has been criticism that a number of

these studies have used clinical and ‘high risk’ samples from more

disadvantaged populations, which lead to a bias in the results. Nonetheless,

Colman et al. (2009), using a sample with a range of participants from no to

severe conduct problems maintain that adolescents who had been rated by

their teacher as having more severe externalising problems were more likely to

experience family and relationship problems, leaving school without

qualifications, and mental health difficulties than those with less severe or no

behaviour problems.

The evidence remains fairly consistent that behaviour difficulties in childhood

and adolescence are associated with a number of detrimental outcomes in

adulthood. A clear need therefore exists for research to further investigate

behaviour difficulties in order to attempt to combat these problems before they

outwork their influence into adulthood. With a broader and more in-depth

knowledge of this phenomenon, particularly in terms of risk, promotive and

protective factors, effective interventions can be sort that will overcome these

negative effects.

35

1.2.4 Summary statements

• Definitions: A vast breadth of terminology is used to describe a child or

adolescent with behaviour difficulties. The terminology selected by any

individual is likely to reflect their educational, community or clinical

background. The term ‘behaviour difficulties’ has been selected for use

within the present study, and a justification and definition for this concept

has been provided

• Prevalence: There exists a high prevalence rate of these problems

within the UK, which may be up to 6% of children (Green et al. 2005).

The exact figures however, are dependent upon the methodology utilised

in prevalence studies.

• Time trends: There is evidence that behaviour difficulties displayed by

children and adolescents have changed over time, with this effect

appearing to have plateaued in recent years, although at relatively high

levels.

• Outcomes: Children with behaviour difficulties not only suffer with

immediate negative consequences such as poor academic achievements

and negative outcomes within their schools, families and wider

communities. There are also a number of long term detrimental

consequences to their behaviour difficulties in childhood that have

significant effects into adulthood.

36

1.3 Special Educational Needs and Disabilities (SEND)

1.3.1 Overview of the concept of SEND

What is SEND2

According to the Education Act (1996) and also included in the SEN code of

practice (DfE 2001a) “A child has ‘special educational needs’ … if he has a

learning difficulty which calls for special educational provision to be made for

him” (section 312).

?

Children are considered to have a learning difficulty if they: “a) have a significantly greater difficulty in learning than the majority of children of the same age; or (b) have a disability which prevents or hinders them from making use of educational facilities of a kind generally provided for children of the same age in schools within the area of the local education authority (c) are under compulsory school age and fall within the definition at (a) or (b) above or would so do if special educational provision was not made for them”. Children must not be regarded as having a learning difficulty solely because the language or form of language of their home is different from the language in which they will be taught.

Special educational provision means: (a) for children of two or over, educational provision which is additional to, or otherwise different from, the educational provision made generally for children of their age in schools maintained by the LEA, other than special schools, in the area (b) for children under two, educational provision of any kind.

(DfES 2001a, page 6 section 1.3, Education Act 1996, section 312)

2 The term SEND (special educational needs and disabilities) is a synonymous construct to SEN (special educational needs). The difference is in name only, with the former term being utilised more recently by the DfE.

37

This definition is included within the Education Act (1996), SEN code of practice

(DfES 2001a) and the Special Education Needs and Disabilities Act (DfES

2001b). Despite its use within these documents, it remains a contentious

definition. It has been criticised for being open ended, and lacking an absolute

characterization. It therefore includes a particularly large and diverse group of

individuals (Halliwell 2003), who are affected by their SEND to various degrees

(O’Regan 2005), with considerable variation found across schools (Ofsted

2005).

The history of SEND SEND has a detailed and complex history, and has been named, defined and

conceptualised in distinct ways throughout modern educational history. The

purpose of this section is to give a very brief overview of its development, noting

where this term originated and therefore providing some background context.

For a more detailed review, see Squires (2012).

Compulsory education was introduced for all children in 1870 with the

introduction of the Forster Act (Spooner 2006). It was not until 1880 however,

that compulsory attendance was required for all children between 5 and 10,

(Squires 2012). Children with more severe SEND where largely ignored during

this time and deemed not able to be educated. It was not until the education act

of 1944 that a shift in this focus emerged; highlighting the requirement of

meeting the needs of ‘handicapped children (Hodkinson & Vickerman 2009).

Within this act it was asserted that children considered maladjusted, subnormal

or physically handicapped were entitled to receive education (Squires 2012).

Nonetheless, as the focus remained upon a medical model of disability, the act

recommend children to be placed into one of eleven categories of handicap.

This enabled educationists to establish what ‘treatment’ these children required.

The emphasis was upon educating these pupils in special schools rather than

within mainstream schools (House of Common report 2006).

The next major shift in educational policy for children with SEND emerged with

the publication of the Warnock report (DES 1978) and its subsequent

38

recommendations for the 1981 Education Act. There was now a growing

emphasis towards the individual child and an acknowledgment of a continuum

of need rather than discreet categories. The assertion was made that children

should be educated in mainstream schools where possible (House of Common

report 2006). This report coined the term special educational needs in a bid to

remove the deficit focus terminology. A more needs focus approach was

adopted where children were supported according to their individual needs

(Warnock 2010). The Warnock report was particularly influential throughout the

1980’s and 1990’s as the intake of special schools reduced and the number of

children identified and educated within mainstream schools increased (House of

Commons report 2006). Indeed Warnock (2010) notes that the framework of the

1981 education act remains in particular force even today more than 30 years

since its implementation.

In 1996, with the publication of The Education Act a definition of SEN was

offered. This definition was also applied to the SEN code of practice (DfES

2001a), and is still used to date. The code of practice was established as a

guide for assessing children with SEND, it does not make any rigid rules about

the category of need in which a child may fall, (but it does acknowledge four

different broad areas of need where a child may have a particular need). The

focus is upon teaching and learning and individual development, a move away

from any medical model’s emphasis on disability, (Squires 2012).

Categories of SEND Within the SEN code of practice (DfES 2001a), it is suggested that there are no

rigid categories of SEND, but rather a broad range of needs which are often

interrelated (Hodkinson & Vickerman 2009). Nonetheless, need should be

considered to fall within at least one of four main domains, these are termed, a)

cognition and learning, b) behaviour, emotional and social development c)

communication and interaction d) sensory and /or physical needs. Despite the

move away from a medical model of categorisation of disability, these 4 broad

domains have been further broken down in to eleven more specific categories

denoting primary need. The reason for this emphasis on categorisation stems

39

from some who want specific disorders to be recognised e.g. autism spectrum

disorders, as well as the government’s desire to collect more detailed

information on pupils and their educational needs. It has also been argued that

a closer monitoring of SEND will allow assessments of initiatives and

interventions for specific types of SEND to be carried out more effectively

(Riddell, Weedon & Harris 2012, DfES 2003).

Space limits a detailed discussion of these terms, although each of the 12

categories will be briefly described to give the reader a flavour of the types of

needs that children placed within these groups require. Within each category

there is a range of need from mild to severe. For more detail on this categories

refer to the SEN code of practice, (DfES 2001a).

Cognition and Learning Needs Specific Learning Difficulty (SpLD): Children with SpLD may have difficulties

with reading, writing, spelling and/or number work that are not related to their

genuine level of ability. Dyslexia, dyscalculia, and dyspraxia would be

considered under this title.

Moderate Learning Difficulty (MLD): Children with MLD have some kind of

developmental delay across multiple areas, and are therefore below the

expected curriculum level. They often have particular difficulty with basic literacy

and numeracy; and may have a number of other needs i.e. language, self-

esteem or concentration problems.

Severe Learning Difficulty (SLD): Children with SLD, have considerable global

delay, often involving communication difficulties and mobility or coordination

problems. They will typically be working on the upper P scales (below level 1 on

the national curriculum), and will need significant help in order to access the

curriculum.

40

Profound and Multiple Learning Difficulty (PMLD): Children with PMLD have

particularly complex needs, combining significant learning difficulty alongside

severe physical difficulty or sensory impairments. They will also require a high

level of adult support and are typically working on the lower P scales.

Behaviour, Emotional and Social Development Behaviour, Emotional and Social Difficulty (BESD): Children with BESD have

particular difficulty with social interactions, they may have concentration

problems, display temper tantrums, be aggressive, disruptive and defiant.

Children who are also withdrawn and depressed would be included in this

category. Disorders such as Attention Deficit Hyperactivity Disorder, Conduct

Disorder and, Oppositional Defiant Disorder would be included here.

Communication and Interaction Needs Speech, Language and Communication Needs (SLCN): Children with SLCN

have problems in communicating through language, either as being understood

or in understanding others. They may have particular difficulties with certain

sounds or pitches and make vocabulary and grammatical errors. They will

generally be behind their peers in oral language skills.

Autistic Spectrum Disorder (ASD): Children with ASD can be assumed to have

a triad of impairments of various degrees. These include difficulty in

understanding non-verbal communication, particularly body language; difficulty

in interacting with others and using appropriate social behaviour; and difficulty

in being flexible and imaginative, preferring routine and involving logical and

repetitive activities. Asperger’s syndrome would be included here.

Sensory and/or Physical Needs Visual Impairment (VI): Children with VI require alterations to be made to their

learning environment on the basis of visual aspects. Although children with

glasses are not included within this category, it does consist of children who are

partially sighted to blind.

41

Hearing Impairments (HI): Children with HI need alterations to be made to their

learning environment and may require particular teaching strategies or may

wear hearing aids. Hearing loss ranges in ability from having slight to profound.

Multi-Sensory Impairment (MSI): Children with MSI have impairments with

hearing and sight and may be both deaf and blind. Particular difficulties for

these children are in, communicating and acquiring information.

Physical Disability (PD): Children with PD have mobility problems that may, or

may not, involve using a wheelchair. Conditions such as Spina Bifida, and

Cerebral Palsy would be included under this heading.

Other Children within the other category have unusual special educational needs that

are significantly different from the categories defined above. (DfES 2003, O’Regan 2005, Spooner 2006, Hodkinson & Vickerman 2009)

Identification and assessment of SEND Identification of SEND often starts with observations from the class teacher.

These include noting whether the child is learning or progressing in a similar

manner to their peers, whether their behaviour can be managed by the normal

behaviour policy, whether they have any problems in communication and

interaction with other students and whether they have any sensory needs i.e.

hearing or vision that do not improve with glasses or hearing aids (Halliwell

2003).

If problems are noted within any of these areas teachers will become aware that

a child may have a SEND. More formal techniques for identification however,

have been outlined within the SEN code of practice (DfES 2001a). Within this

document strategies that teacher and schools can use include teacher

observations, standardised tests and tools, progress on National Literacy and

Numeracy Strategy Frameworks, performance on National Curriculum levels or

baseline screening tests (DfES 2001a). According to the SEN code of practice

42

(2001a) assessment should be carried out as a continuous process and not

based on a single event or time point. Furthermore, taking into account the

severity, frequency and context in which needs occur, as well as the age of the

child are important when identifying a child with SEND.

The Department for Education has provided guidelines in terms of identification

and categorisation of SEN (DfES 2001a, DfES 2003). However, whether a child

is deemed to have SEND, or indeed categorised within a certain SEND group,

is an arbitrary process, often depending upon the school and their ability to

provide for the child’s needs. Furthermore, as many children have multiple

needs that may overlap in a number of categories, it is the schools decision

which they feel is the child’s primary need and again this could be seen as an

arbitrary judgment. Primary need is only required to be recorded in the SEND

register at the school action plus level of provision, and those at school action

may not be categorized at this stage (DfES 2003).

Levels of SEND support Pupils, who have been defined as having SEND, are offered provision at one of

three levels, termed School Action, School Action Plus or Statement of Special

Educational Need (DfES 2001a, Frederickson & Cline 2009). This is a

graduated response to their needs with more help being offered to the child as

they move through the levels.

• The school action (SA) level of support asserts that children should

receive additional and different help, which goes beyond the normal

differentiated curriculum (DfES 2001a). The basic assumption behind this

level is that a child is not making sufficient progress despite the

differentiated teaching in place. It is the school’s responsibility to provide an

intervention or plan in order to provide for their additional needs, setting this

out within an IEP (individual education plan), (Halliwell 2003). Children who

continue to make little or no progress despite being at school action, would

be placed at the next level school action plus.

43

• The school action plus (SAP) level of provision is received from outside

agencies, such as educational psychologists or speech and language

therapists. This specialist support may take the form of advising the

teacher/SENCO or supporting the child directly. Interventions need to be

different and additional to those being supported at school action alone

(DfES 2001a, Farrell 2004).

• A statement of special educational needs (ST) is given for children with

the most severe and complex needs, and when the additional help from

school action plus is having a limited effect. The school or parent(s) notifies

the local authority and an assessment is made and a formal document

written which highlights the additional help they require (DfES 2001a). This

type of help may involve interventions such as direct teaching from a

specialist teacher, or therapy from a specialised professional (Farrell 2004).

The level of provision that a child receives does not necessarily reflect their

severity of need, and comparisons across schools of children at the same level

of provision may not be equivalent. This is because the level of provision

inherently reflects the school’s ability to meet a particular child’s needs. A child

may be at the school action level at one school, however, if moved to another

school, would be placed at the school action plus level as this school needs

more support from outside agencies to meet the child’s needs.

Prevalence rates In a recent government report (DfE 2011a), prevalence rates of the different

SEND groups were recorded. It has been reported that approximately 21% of all

school pupils in 2011 had been recorded as having SEND (this included those

at school action, school action plus and statement) specifically this equates to

1.67 million children. There has been an increase in the number of children with

SEND as in 2006; this number was significantly lower, 1.53 million, (equating to

19% of school population).

44

The percentage of primary need group of SEND are recorded in table 1.13

Table 1.1: Percentage of SEND by Primary need, (DfE 2011a)

Primary Need Percentage of SEN pupils (school action plus and statement)

SpLD 11.1%

MLD 22.9%

SLD 4.2%

PMLD 1.4%

BESD 22.5%

SLCN 17.2%

ASD 8.8%

VI 1.3%

HI 2.3%

MSI 0.1%

PD 3.8%

Other 4.3%

As can be seen from the above table the two most common primary needs are

MLD (22.9%) and BESD (22.5%). The lowest two groups are composed of MSI

(0.1%) and VI (1.3%).

This information was taken from the report Children with Special Educational

Needs: an analysis – 2011 (DfE 2011a), other key findings to emerge revealed

that boys had a 2 ½ times greater chance of having a statement for SEND at

primary school than girls. During secondary school boys were almost three

times more likely than girls to have a statement for SEND. Ethnic differences

were also found with Chinese pupils having the least probability of having

SEND in primary schools and Black pupils having the highest probability. In

3 These percentages include only school action plus and statement levels as teachers are not required to assert a primary level of need for children at the school action level of support, (DfES 2003).

45

secondary schools, Chinese pupils again had the least likelihood of having

SEND, and Black pupils the greatest chance of having SEND without a

statement and White, Black and Mixed race the highest likelihood of having

SEND with a statement. Children who were entitled to free school meals were

also at an increased likelihood of having SEND compared with those not

entitled to free school meals (DfE 2011a).

46

1.3.2 Justification of SEND population within the study

The current section attempts to highlight the key arguments as to why this

population was chosen for use within the present study that is attempting to

identify the key risk, promotive and protective factors for behaviour difficulties

for this population.

Large number: The number of children within England who have been identified

as having SEND is estimated to be approximately 1.67 million (DfE 2011a). This

population makes up approximately one fifth (21%) of the school population in

England, including those children at school action, school action plus and those

with a statement for SEND. This is not a small or insignificant population as one

fifth represents a considerable number of school pupils. They all have a least

one thing in common - special education provision is being made for them. Any

research adopting an appropriate sampling method from this population will be

able to generalise its findings to a large number of children, making it a

particularly beneficial study.

Little research: To my knowledge there has been no study that has specifically

utilised a SEND population to investigate risk, promotive and protective factors

for behaviour difficulties within England. Therefore, research using the whole

range of children with SEND to investigate risk, promotive and protective factors

for behaviour difficulties is warranted and will add to the literature base.

Overlap: There is overlap between the risk factors for children with SEND and

the risk factors for those developing behaviour difficulties. Evidence shows that

being male (Ehrensaft 2005), having lower economic status (Schonberg &

Shaw 2007), speaking English as an additional language (Brown & Schoon

2008), and lower academic achievement (McIntosh, et al. (2008) are key risk

factors for developing behaviour difficulties and also salient characteristics of

the SEND population (DfE 2011a). Furthermore, a study investigating the

mental health of children and young people within the UK found clear overlaps

between those individuals identified as having conduct problems and those

identified as having SEND e.g. 52% of children who were reported to have

47

conduct problems were also identified at their school as having SEND (Green et

al. 2005). The same study showed that of the children identified as having

conduct problems, 66% also had developmental problems, and 59% were

behind their peers in school (both key issues within the SEND population). It

would therefore be beneficial for further research to add to the literature base

for this group of ‘at risk’ individuals.

Outcomes: It is important that future enquiries are able to identify key risk,

promotive and protective factors for behavioural difficulties within the SEND

population as current research shows that these children not only have an

increased likelihood to being excluded from school (DfE, 2011a) but also in

developing mental health problems (Rose, Howley, Fergusson, & Jament,

2009). These negative outcomes could be abated once the aetiology of

behaviour difficulties is established (from risk, promotive and protective factor

identification). These findings will help in the development of effective

interventions that will combat or alleviate the potential negative outcomes.

Group variations: Evidence has suggested that risk, promotive and protective

factors for behaviour difficulties vary as a function of gender (Storvoll &

Wichstrom 2002) and socio-economic status (SES) (Schonberg & Shaw 2007)

i.e. these factors have differing effects for, males versus females and those

from high versus low SES backgrounds. It is likely therefore that a different

population such as those within school identified as having SEND may be

affected by distinct risk, promotive and protective factors, which may be distinct

from the general population. Research therefore is required to investigate this

pertinent issue.

All SEND vs. Just BESD: Children with SEND are potentially the most ‘at risk’

students within school in terms of displaying behaviour difficulties, (Murray &

Greenberg 2006). Indeed any child who is suffering from the most severe

behaviour difficulties will be likely to have been identified as having SEND and

specifically placed within the BESD category. This research will therefore target

those children who are most likely to display behaviour difficulties. The entire

population of SEND was selected, however, over those solely identified as

48

having BESD, as other groups such as those with ASD also often display

significant problems with their behaviour. Furthermore, children are recorded as

having SEND and placed into a category based on their primary need, it is often

the case that needs extend across categories with children having needs in

multiple areas, e.g. a child with MLD may also be experiencing BESD. What is

ultimately decided as the child’s primary need maybe an arbitrary judgment.

Consequently it cannot be assumed that only children with BESD will have

behaviour difficulties, these problems are also likely to occur in the other

groups. Identification of SEND is a problematic matter and despite the DfE

providing guidelines on the issues of SEND (DfES 2001a, DfES 2003), whether

a child is identified and categorised depends more upon the schools ability to

provide for their needs (Ofsted 2005). Therefore the same child with the same

difficulty may not be recorded as having SEND if placed in a different school.

An individual is placed into a category on the SEND register in a fairly arbitrary

fashion, because of the interrelationship of these categorises, and as definitions

are based upon schools capacity to provide for the needs. It could be argued

therefore, that basing any findings on a specific group of SEND children is

flawed, and a more appropriate method would be to focus on the whole

population of SEND children.

49


• This section has provided a brief overview of the concept of special educational

needs and disabilities. Important issues have been acknowledged including:

definitions of the concept, the historical background, categories of need,

identification and assessment, levels of support and prevalence.

• A justification for why this population has been included within the present study

has been provided. This includes children with SEND representing a large

under researched population, with considerable overlap between the risk

factors for behaviour difficulties generally and for having a SEND. Children with

SEND are also likely to be the most ‘at risk’ students within their school in terms

of displaying behaviour difficulties.

50

1.4 Risk, promotion and protection

1.4.1 Risk

Overview Risk is a popular term in the psychological literature (Keyes 2004), although it is

also a potentially confusing concept that has numerous definitions and can be

operationalized in many different ways (Kraemer, Lowe & Kupfer 2005). It is

generally defined as the “probability of an outcome within a population”

(Kraemer et al. 2005, page 5). Of particular relevance to the present study are

the factors that contribute to enhancing the probability of the outcome in a

specific population, i.e. the focus is on so called risk factors.

Risk factor research stems from the field of epidemiology that attempts to

distinguish certain factors that heighten the chances of disease in an individual.

This type of research aimed to indentify a broad range of possible factors that

individuals are exposed to that ultimately leads to disease. It was from this

research that the early risk factor models emerged within the psychological

literature; and they were made up of lists of predictor variables for certain

outcomes, such as mental illness (Garmezy 1996).

Risk factors Risk factors are defined as “any event, condition or experience that increases

the probability that a problem will be formed, maintained or exacerbated”

(Fraser & Terzian 2005, page 56). Wright & Masten (2005) emphasise the

importance of measurement defining a risk factor as “a measureable

characteristic in a group of individuals or their situation that predicts negative

outcome on a specific outcome criteria” (Wright & Masten 2005 page 9). These

definitions convey the generally accepted conceptualisation that anyone

exposed to a risk factor has a higher probability of suffering with a negative

developmental outcome (Schoon 2006, Deater-Deckard, Dodge, Bates & Pettit

1998, Atzaba-Poria, Pike & Deater-Deckard 2004, Gerard & Buehler 2004a). It

has also been argued that risk factors are not only associated with a higher

51

probability of developing a poor outcome, but also with increased severity and

longevity of that outcome as well (Coie et al. 1993). Key to risk factor related

research is that these factors ultimately reflect probabilities and not certainties,

they influence the outcome but they do not necessarily guarantee it will occur,

(Kraemer et al. 1997).

More specifically, risk factors are comprised of traits, experiences, situations or

relationships that are ultimately measurable (Keyes 2004). They impinge on

development contributing to negative trajectories, (Murray, 2003). These factors

will be present and operate across different ecological domains including

individual, familial, school, and neighbourhood levels (Raviv, Taussig, Culhane

& Garrido 2010, Jenson & Fraser 2011). They may stem from genetic,

biological, psychological, environmental, socio-economic and social-cultural

influences (Schoon 2006) or indeed combinations of these (Richman & Fraser

2001).

Correlates, Predictors, Causes Although the term risk factor is a particularly prolific one found within the

psychological literature (Kraemer et al. 1997), there has been considerable

imprecision and inconsistency in its conceptualisation. Some researchers have

used this term to reflect basic correlates of a specific outcome, whereas others

use this to indicate predictors or even causes of the outcome (Ribeaud & Eisner

2010, Murray et al. 2009). The result of this confusion has led to a general

consensus in the literature that some distinction should be made between

correlates, predictive risk factors and causal risk factors, (Murray, Farrington &

Eisner 2009).

Correlates are established when a significant relationship (either positive or

negative) is determined with the outcome in question, and at a certain point in

time, (Offord & Kraemer 2000). No judgment however, can be made as to which

(the factor or outcome) preceded the other, and therefore these factors cannot

be predictive of the outcome or assumed causal in anyway, (Murray et al.

52

2009). They should not therefore be referred to as risk factors, but correlates

(Kraemer et al. 1997).

In order for the term risk factor to be applied to any variable it is required not

only to be significantly related to the outcome, but importantly to also precede it

(Offord & Kraemer 2000). Risk factors can then be defined as predicting the

negative outcome in question. In order to apply such terminology to any

investigation the use of longitudinal data is therefore a prerequisite (Murray et

al. 2009). According to Kraemer et al. (2005), a common error of investigations

are their use of the term ‘risk factor’ when they are actually measuring

correlates as no temporal issues of when risk factor and outcome occurred are

taken into account.

This conceptualisation of risk factors is fairly simplistic and in fact there are two

different types can be distinguished; fixed markers and variable markers,

(Kraemer et al. 1997). Fixed markers are variables such as gender or ethnicity

that are related to outcome and precede it although cannot change or be

changed, whereas variable markers are related to the outcome and precede it

although can change or be changed, (Kraemer et al. 1997).

There is a final type of risk factor, which has been termed a causal risk factor.

These are so called when they are manipulated in some way cause a change in

the outcome (Murray et al. 2009). The use of the term causal risk factor is

favoured over cause to prevent a deterministic effect creeping in to

psychological research that is ultimately based on probabilities, (Murray et al.

2009). Using the term cause however, in causal risk factors does not indicate

that it is the only or unique cause to the outcome in question (Offord & Kraemer

2000).

Conclusion The conceptualisation of risk factor terminology has experienced a number of

changes and modifications from its inception (Wright & Masten 2005). The

original researchers within this field often investigated broad correlates of the

53

outcome in question and paid little attention to the temporal relationships

between risk factors and outcome (Jenson & Fraser 2011). In order for any

variable to be accurately conceptualised as a risk factor it needs to not only be

related to the outcome but also precede it. Defining the concept in this way is

not an inconsequential matter, (Schoon 2006), and it remains particularly

important that a distinction can be made between correlates, risk factors and

causal risk factors (Wright & Masten 2005). Researchers need to use these

terms precisely and consistently (Offord & Kraemer 2000), and explicitly state

the type of risk factors that are present in their research, i.e. whether they are,

fixed/variable markers, or causal risk factors (or even correlates) (Kraemer et al.

1997).

If there continues to be imprecision in terminology used to explain risk factors,

distinct differences in this research may emerge, that will not only fail to give an

accurate account of the very outcome under investigation, but also fall short of

providing effective interventions to help those suffering with the negative

outcome (Kraemer et al. 1997). In the context of the present study the term risk

factors will be utilised where variables predict a heightened probability of later

behaviour difficulties, some being fixed markers and others variable markers.

54

1.4.2 Promotion and protection

Overview Any study solely acknowledging the impact of risk factors on a developmental

outcome will provide an insufficient account of this behaviour. An additional

requirement is for research to account for the variables and processes that

enhance positive outcomes (Leshner 2002, Murray 2003). Such variables have

been traditionally termed protective factors as they are related in a positive way

to the outcome in question (Schoon 2006).

The inception of protective factor research was focused upon searching for

personal characteristics of individuals that would help to overcome their risk

experience, this narrow focus however, soon shifted to incorporate features of

the social environment as well as within person factors (Schoon 2006).

Protective factors have now been acknowledged to originate from three broad

domains; at the individual level (e.g. positive temperament or intellectual skills)

at the family level, (e.g. parental involvement or positive family relationships)

and wider community level (e.g. school belonging and community resources).

The factors referring to the individuals personal characteristics such as coping

skills or self-efficacy have also be termed assets whereas resources are factors

external to the child such as positive parenting, and community support

organisations (Fergus & Zimmerman 2005, Windle 2011).

The definitions provided for the concept ‘protective factors’ have been

considerably more controversial than those used for risk factors, (Dekovic

1999). Indeed protective factors have been conceptualised in distinct ways by

various researchers who disagree on the exact processes involved,

(Stouthamer-Loeber et al. 2002, Vanderbilt-Adriance & Shaw 2008). The two

contentious questions surrounding the conceptualisation of these factors are

firstly whether they are qualitatively different from risk factors or whether they

are in fact the same concept but at different poles of the continuum; and

secondly, do protective factors promote competence for all individuals

regardless of risk status, or are they are more specific and only operate in

context of high risk (Colman & Hagell 2007).

55

Turning initially to the first question addressed, there is considerable evidence

that risk and protective factors are the two polar opposites on a given variable

(Luthar & Zelzo 2003, Stouthamer-Loeber, et al. 1993). A variable is termed risk

or protective on the basis of the end that is emphasised. Using the example of

intelligence, above average levels are often conceptualised as a protective

factor for behaviour difficulties whereas low intelligence is seen as a risk factor

(McIntosh, et al. 2008). There are also a number of dichotomised variables

where the two categories represent risk and protection respectively e.g. gender,

evidence shows being male is a risk factor for behaviour difficulties where as

being female is a protective factor (Brown & School 2008). Protective factors

have also been identified by some as purely protective, offering enhanced

adjustment when present but no detrimental effect when absent, i.e. a musical

talent, or religious involvement (Masten 2001). This may be the case when

measuring categorical variables, with certain ‘groups’ being related to

decreases in the favourable outputs and therefore being noted as risk, and

certain ‘groups’ being related to increases in the favourable outcome and

therefore being coded as protective, and other groups having no relationship

with the outcome.

There are cases in the literature where variables evidence a non-linear pattern,

and are not bipolar. Therefore these factors offer protection in middle values but

risk at either extreme, e.g. parental control, too controlling or too lax parenting

may be risks whereas a healthy balance offers protection. Risk and protective

factors are not always opposite ends of the same variable, with some being

uniquely risk related or uniquely protective related.

The second pertinent question is whether protective factors promote positive

outcomes for all individuals or if they only offer protection in the context of risk.

Originally the term protective was only used in the context where interaction

effects occurred between protective variables and risk in influencing outcome,

(Garmezy, Masten & Tellegen 1984) although other researchers (e.g. Werner &

Smith 1992) referred to protective factors as those that had direct positive

effects on outcome regardless of risk experienced. Distinguishing between

56

these two is important in highlighting most accurately how protective factors

function.

There are at least two different processes that can explain how protective

factors work, (Fergusson & Horwood 2003). They can either be protective

processes; where protective factors assist those in high risk situations to

develop positive adaption but offer limited help to those in low risk situation, or

promotive processes; where the protective factor is beneficial to all individuals

regardless of their risk status. Protective processes aim to establish interaction

effects between protective factor and risk in influencing outcome, whereas

promotive process aim to establish main direct effects of protective factor on

outcome. The difference between the two will be drawn within the present

study.

Protective factors The term protective factors should be applied to variables that have particular

salience, though only in contexts of high risk, in moderating the impact of risk

exposure on outcomes, (Sameroff, Bartko, Baldwin, Baldwin & Seifer 1998). A

definition will be applied to the present study which states a protective factor is

a “quality of a person or context or their interaction that predicts better

outcomes, particularly in situations of risk or adversity, (Wright & Masten 2005,

page 19). Protective factors are therefore features of the individual or their

environment that are associated with overall better than expected outcomes for

those considered at high risk (Jenson & Fraser 2011). It has been suggested

that in order to have some meaning, protective factors must be defined by the

interaction process of modifying risk situations to influence the outcome in a

positive direction (Stouthamer-Loeber et al. 2002). Statistical models including

these types of factors investigate them through interaction terms rather than

main effects terms.

In low risk contexts the effect of the protective factor is either non-existent or

offers limited extra advantage, (Fergusson & Horwood 2003). It may even be

undetectable in the absence of risk as the factor modifies risk experience rather

than enhancing typical development generally (Rutter 1985). When risk is

57

present however, the protective factor interacts with it modifying or buffering its

potential impact on the outcome. Protective factors therefore provide a shield or

barrier that impedes or halts the effect of risk (Schoon 2006), and offsets the

negative trajectory (Hanewald 2011). The impact that the protective factor has

depends on whether risk is present or absent, offering protection largely in

situations of risk (Herrenkohl et al. 2003, Rutter 2003, Van der Lann, Veenstra,

Bogaerts, Verhulst & Ormel 2010). An example of this effect has been showed

by Criss, Pettit, Bates, Dodge & Lapp (2002) who report that high levels of peer

acceptance (a protective factor) is particularly beneficial for children exposed to

significant family adversity (i.e. at high risk). These factors when present reduce

the likelihood of negative outcomes as they are offering some kind of protection

over risk (Luthar & Cicchetti 2000, Greenberg 2006, Windle 2011, Murray 2003,

Coleman & Hagell 2007, Schoon 2006).

Promotive factors The debate whether protective factors are concerned with main or interaction

effects has led some to argue for the use of different terminology to explain

these different effects and clear up some of the confusion (Sameroff et al.

1998). Factors that promote positive outcomes regardless of any risk

experienced have therefore been called promotive factors, (Sameroff et al.

1998) and should be used when discussing main effects. Promotive factors can

be defined as “a measureable characteristics in a group of individuals or their

situation that predicts general or specific positive outcomes” (Wright & Masten

2005, page 19). These types of factors when present decrease chances of

negative outcomes for all individuals regardless of their different risk exposure

and foster positive adaption. e.g. high self-esteem (Donnellan, Trzeniewski,

robins, Moffitt & Caspi 2005). Statistical models including these types of factors

look for direct main effects and not interaction effects (Luthar & Cicchetti 2000).

Promotive factors rely upon main effects findings, as does risk related research

(albeit in the opposite direction). There is an argument that if promotive and risk

factors are just opposite ends of the same variable then both terms do not need

to be explicitly acknowledged, and either one would suffice. In the majority of

58

cases these variables are bipolar in nature (Masten 2001) with the positive end

of variable being promotive and the negative end being risk related. The same

variable can therefore signify risk to one individual (i.e. low academic ability),

whereas to another act as a promotive factor (high academic ability), and a third

a neutral factor (i.e. medium level of academic ability) (Van der Laan et al.

2010, Stouthamer-Loeber et al. 2002). Nonetheless this does not however

mean risk and promotive factors are direct opposites or mirror images of one

another (Stouthamer-Loeber et al. 1993, 2002). The intensity of effect at each

pole of the continuum (on continuous variables) may be different suggesting

variability in magnitude (Loeber, Slot & Stouthamer-Loeber 2008). Emphasizing

this effect can more accurately account for when variables are not normally

distributed as well as when the variable and outcome are not related in a linear

pattern (Van der Laan, et al. 2010). Finally, some variables are uniquely

promotive i.e. musical talent which may offer a positive outcome when present

but no detriment when absent.

Conclusion Controversy exists in the literature as to whether protective factors are

independent from risk factors or whether they depend on them in order to be

determined as protective (Jenson & Fraser 2011). This debate stems around

the mechanisms in how these variables function, (Van der Laan et al. 2010), as

protective factors moderate effects of risk, whereas promotive factors decrease

chances of negative outcomes regardless of risk. Studies may choose to

investigate either main or interaction effects however, Luthar, Cicchetti &

Becker (2000b) suggests that both types are important and beneficial, and they

equally may yield important findings. Furthermore, some factors could be both

promotive and protective depending on the context and are not necessarily

mutually exclusive (Wright & Masten 2005).

To conclude distinctions will be made between promotive and protective factors

within the context of the present study, which will attempt to measure main as

well as interaction effects on the outcome behaviour difficulties with a sample of

children with SEND. An example of a promotive factor with the present study

59

may be academic achievement which promotes positive behaviour regardless

of other risks experienced. An example of a protective factor with the present

study could be small school size which may be particularly beneficial for

children at high risk but not provide any additional advantage for those in low

risk situations.

The focus of the current study is upon the key risk, promotive and protective

factors for the display of behaviour difficulties in a population with children with

SEND. Chapter 2 contains a discussion of various risk and promotive factors for

these behaviours as these terms are conceptualised as either opposite ends of

the same variable, or variables that are uniquely risk related or uniquely

promotive. Variables that may be considered protective, (i.e. that have a

greater impact on outcomes in contexts of high risk) are discussed in Chapter 4.

60


• Risk factors: A definition of a ‘risk factor’ has been provided, which

broadly acknowledges that in their presence they are related to more

negative outcomes. Importantly, to be classified as a true risk factor, they

must not only be significantly related to the outcome in question but also

precede it.

• Promotive factors: A definition of a ‘promotive factor’ has been provided

which broadly conceptualises it as the opposite to a risk factor and in its

presence related to more positive outcomes.

• Protective factors: A definition of a ‘protective factor’ has been provided

which acknowledges the interaction or moderation effects that these

have on reducing risk and therefore improving outcomes.

61

1.5 Aim of study and research questions

1.5.1 Aim

The broad aim of this study is to investigate risk, promotive and protective

factors for behaviour difficulties in children with SEND. The purpose was to gain

a better understanding of how to reduce such problems, and therefore enable

better outcomes in this large but under researched group of children.

The justification for investigating the concept of behaviour difficulties was

discussed within section 1.2, where it was highlighted there is a serious concern

that these behaviours displayed in childhood have pervasive and long-term

effects into adulthood and across various domains of functioning. More in-depth

investigations into behaviour difficulties particularly in terms of risk, promotive

and protective factors will allow the most effective intervention programmes to

emerge. These will ultimately be able to combat problem behaviours before

their influences become entrenched into adulthood. Despite some evidence

suggesting these behaviours in childhood might have decreased in recent years

or at least plateaued, there still remains large numbers of children who

experience these difficulties, and therefore research should continue to attempt

to combat these problems. Including a SEND population was justified in section

1.3, as little research has explicitly investigated behaviour difficulties within this

large group of children who are considered particularly at risk of displaying

behaviour difficulties. Unsurprisingly there is also considerable overlap in the

risk factors for those who develop behaviour difficulties generally and those who

develop a SEND, making them a population of considerable interest.

The more specific aims are: firstly to investigate a number of potential risk and

promotive factors for behaviour difficulties across the school and individual

ecological levels. Chapter 2 provides an overview of these factors and a

justification for their inclusion within the present study. Secondly, to investigate

the cumulative effects of contextual risk factors on behaviour difficulties,

whether number of risks or specific types of risk is a more salient issue, and

62

assessing the functional form of the relationships between increasing risk and

behaviour displayed. Chapter 3 discusses these issues justifying their presence

within the study. Thirdly and finally; to investigate school level protective factors

and whether these are particularly beneficial for children who are considered at

high risk in reducing their behaviour difficulties. Chapter 4 provides a discussion

of the pertinent issues in this regard.

The research questions that have come from these three chapters are

presented in the summary section at the end of each chapter respectively. For

the purposes of clarity, however, all the research questions included within the

study are presented within this section as a complete whole. Although these

questions have emerged from the literature a salient point is that they have not

been previously acknowledged with a special needs population within England.

This provides a justification that the present study is unique and offering a

significant contribution to knowledge.

63

1.5.2 Research questions4

Research question 1: Risk and Promotive Factors

a) What proportions of variance in behaviour difficulties are attributable to

the school and individual levels?

b) Which school and individual level predictors explain a statistically

significant proportion of variance in behaviour difficulties?

c) Of the variance initially attributed to the individual and school levels, how

much can be accounted for by the predictors used in the study?

Research question 2: Cumulative Risk

a) Is there a cumulative effect of contextual risk factors on behaviour

difficulties, where higher numbers of risk factors present are associated

with increased levels of behaviour difficulties?

b) What is the nature of the relationship between exposure of cumulative

risk and behaviour difficulties?

c) Is the number of risks present within an individual’s background more

important than the specific types of risks in accounting for behaviour

difficulties?

Research Question 3: School Protective Factors

a) Which school level predictor variables have a statistically significant

interaction effect with contextual risk?

b) Do the significant interactions display evidence of protective-

stabilising or protective-reactive effects?

4 The research questions are included here as a whole, however, each of the broad questions are justified for use in the present study after discussing the literature from which they emerged in chapter 2, 3 and 4.

64

2. Risk and Promotive factors

2.1 Introduction to chapter The aim of this chapter is to provide an overview of some of the most salient

risk and promotive factors for behaviour difficulties among children and young

people. As discussed in section 1.4, risk and promotion are often

conceptualised as the opposite ends of the same construct, and this approach

has been adopted within the present study. As such when any factor is

discussed in terms of being a risk factor for behaviour difficulties i.e. being male

or low academic achievement, the opposite end of the construct is assumed to

be a promotive factor i.e. being female or high academic achievement. These

factors are noted in terms of whether they are fixed (i.e. cannot change or be

changed) or variable (i.e. can change of be changed).

As this field is vast it is important to categorise these factors in some meaningful

way in order to aid understanding. The present study has clustered these

factors to different ecological levels in which they operate. Commonly within the

literature they have been referred to as individual level factors, (discussed in

section 2.2) family level factors, (discussed in section 2.3) school level factors

(discussed in section 2.4) and neighbourhood level factors, (discussed in

section 2.4) (Jenson & Fraser 2011, Coleman & Hagell 2007, Murray 2003). It

has been argued that accounting for various influences across multiple

ecological levels is essential in order to capture some of the complexity behind

the display of behaviour difficulties (Ribeaud & Eisner 2010).

The present study does not aspire to conduct an exhaustive review of all the

relevant risk and promotive factors for behaviour difficulties; the field is too vast

65

and the space too limited. The current review provides a synopsis of the most

salient risk and promotive factors at each ecological level, in order to justify the

inclusion of some of these variables into the current study. To some extent the

investigation may appear speculative due to the breadth of variables included,

however, the ends ultimately justify the means; the current study is looking to

provide evidence around risk and promotive factors for behaviour difficulties

exclusively for an under-researched population; children with SEND. There has

been a lack of research specifically utilising a sample of these children. As such

the present study will be able to provide a significant contribution to knowledge

in this area.

A particular focus of the present study is upon the individual and school

ecological levels; although other influences such as the neighbourhood and

family levels are acknowledged, these have been paid less attention in favour of

a more in-depth analysis of individual and school influences. The justification for

a focus on the individual level is that studies have consistently found variables

within this level are more powerful predictors of behaviour difficulties than in

other ecological levels (Reis, Trockel & Mulhall 2007, Maes & Lievens 2003,

Gottfredson & DiPietro 2011, Aveyard, Wolgang, Markham & Cheng 2004). The

justification for the school level variables is that the population of interest

(children with SEND) is identified within the school context, so it is likely these

institutions have some effect on behaviour. Furthermore, studies specifically

investigating school level influences on behaviour problems have not been

forthcoming (Reinke & Herman 2002), with a stronger focus on family over

school level influences (Bronfenbrenner 1994). Children spend a considerable

amount of their time within a school context, with levels of behaviour difficulties

varying remarkably between schools, (McEvoy & Welker 2000). It is not

surprising therefore that these institutions will exert their unique and powerful

influence on childhood and adolescent outcomes, (DeWitt et al. 2000). Section

2.6 within this chapter therefore discusses the relative importance of the

individual and school ecological levels in accounting for behaviour difficulties.

The final section 2.7 concludes the chapter by providing summary statements,

and related research questions.

66

2.2 Individual level risk and promotive factors

2.2.1 Age

There has been considerable research that has explicitly investigated whether

an individual’s age is a fixed risk factor for behaviour difficulties. Studies have

often suggested that older children are more likely to display behaviour

difficulties than younger children, (Green et al. 2005, Maughan, Rowe, Messer,

Goodman & Meltzer 2004). Nonetheless, there is accumulating evidence to

suggest that age exerts its effects to a differing degree depending on the

behaviour in question (Bongers, Koot, van der Ende, & Verhulst 2004). It

appears that more aggressive behaviour difficulties decline with age as non-

aggressive problems increase. For example Bongers, et al. (2004) argue that

aggression, oppositional behaviours and property violations all decline with

increasing age whereas status violations (such as truancy, running away from

home and alcohol and drug use) increase with age. This finding is fairly

consistent as Tremblay (2000) reports as a child becomes older they are less

likely to engage in physically aggressive acts although more likely to show other

behaviour difficulties such as drug use and truancy. Not all the evidence

however, is in total agreement with this view, as Lahey et al. (2000) showed that

although younger children received higher scores on parent report of

oppositional behaviours, no age differences were found for aggressive

behaviours, and the middle ages of a 9-17 year old sample had the most

aggressive behaviours.

Lahey et al. (2000) have argued that attempting to understand how age effects

the development of behaviour difficulties is futile if not considered from a

developmental perspective. The types of behaviour difficulties displayed change

across the course of development so researchers need to be aware there are

multiple developmental pathways for these problems, with at least two groups,

which have been termed early onset (i.e. within childhood) and late onset

(within adolescence). The causes or risk factors associated with them are likely

to be distinct (Lahey & Waldman 2003). In acknowledging this stance behaviour

67

difficulties will be assessed using two separate analyses, a primary school

model and a secondary school model within the present study (see section

5.9.2).

Whether age is a fixed risk factor for behaviour difficulties is a complicated

issue. As the measure of behaviour difficulties used within the present study

includes overt aggressive acts such as fighting and vandalism as well as covert

acts such as lying and stealing, it is not possible to establish whether overt acts

decrease with age and covert ones increase as the majority of evidence

suggests. Nonetheless, it is anticipated that age will have some effect on global

behaviour difficulties displayed, with older children being more at risk than their

younger counterparts.

2.2.2 Relative age in the school year

Research has shown that the month in which a child is born affects academic

and behavioural outcomes (Martin, Foels, Clanton, & Moon 2004, Goodman,

Gledhill & Ford 2003). Specifically, children born at the latter part of the school

year (who are the youngest in their classes) have a significant academic

disadvantage (Martin, et al. 2004), with the effects appearing to be long-term

even influencing university entrance (Russel & Startup 1986).

In terms of behavioural outcomes, Goodman, et al. (2003) showed that in a

large sample of 5-15 year olds, the younger children in the year were more

likely to be at an increased risk of psychiatric problems, including conduct

difficulties. This evidence is fairly consistent with Menet, Eakin, Stuart & Rafferty

(2000), who found that children younger in the school year were not only rated

by their teachers as having more behaviour problems, but are also referred to

Educational Psychology Services to a greater extent. Another study suggesting

that younger relative age in a school year is a significant risk factor in the

display of behaviour difficulties, found that children with emotional and

behaviour difficulties, were over-represented within the later part of the school

year (Polizzi, Martin & Dombrowski 2007).

68

Despite this fairly consistent evidence, Lien, Tambs, Oppedal, Heyerdahl &

Bjertness (2005) demonstrated that although relative age in the school year had

some effect on academic outcomes, it exerted no influence on behaviour

outcomes. While these results may appear contradictory, their study used a

somewhat limited sample of only 15-16 year olds and not a representative one

across all year groups within the school system. There could be an argument

here that relative age in a school year has more pronounced effects on younger

children rather than older children. Menet et al.’s (2000) study, which found the

differences between the youngest and oldest children in each school year, was

more marked for children in year 1 compared with those in year 3 and 5. They

suggested this could reflect maturity levels; as the difference between the

oldest and youngest children in year 1 is proportionally greater (in terms of their

overall age) than the differences found in older school years. This is a plausible

explanation, although in Lien et al.’s (2005) study, a physical measure of

maturity was taken, which showed no differences between younger and older

children in the same school year, suggesting that maturity was not the key

issue. Nonetheless their sample did include older adolescents rather than

younger children, which may explain the lack of a significant finding.

A slightly alternative explanation has been offered which suggests that younger

children within a school year are more likely to display behaviour difficulties as

they are less socially and physically skilled and therefore have a social

disadvantage particularly in terms of peer relationships (Martin et al. 2004).

These influences on their peer relationships could in turn have an influence on

them displaying more inappropriate behaviours. This argument has been

supported by Lien et al. (2005), as although they did not find behaviour

differences between younger and older children in the school year, differences

between problems with peers were noted, with older children having fewer

problems.

In concluding this debate Goodman et al. (2003) suggests that relative age is

an independent fixed risk factor of behaviour difficulties, although the effects are

fairly small in comparison to other established risk factors. The present study

takes Goodman et al.’s (ibid.) view and predicts that younger children in the

69

school year will be at an increased likelihood of displaying behaviour difficulties

compared with their older peers.

2.2.3 Gender

Gender is a commonly reported predictor or fixed risk factor of behaviour

difficulties. Research has overwhelmingly showed that boys are more at risk of

displaying problem behaviours compared with girls (Brown & School 2008,

Green et al. 2005). Evidence has demonstrated that gender differences are

evident in the display of aggressive behaviour for children as young as 1 and a

half years of age, (Baillargeon et al. 2007). This study showed that 5% of boys

compared with only 1% of girls were regularly displaying serious aggressive

behaviours. Other research is in agreement, finding that boys are more likely to

display aggressive behaviours and partake in vandalism (Lahey et al. 2000),

and are more likely to be exposed and vulnerable to specific risks for behaviour

difficulties (Storvoll & Wichman 2002).

Within the UK, in a study investigating the prevalence of mental health problems

in children and adolescence, Green et al. (2005), showed that the rates of those

identified with a conduct disorder were significantly higher in males compared

with females. Indeed of those identified with this disorder 63% were boys. More

recently in the fourth update from the millennium cohort study, of those children

at age 7 who had a serious behaviour problem, twice as many were boys

compared with girls (Brown & Schoon 2008).

Researchers in this field have suggested a number of reasons why males may

be at an increased risk of displaying a behaviour difficulty compared with

females. It has been argued this could relate to biological and hormonal

differences, (Book, Starzyk & Quinsey 2001), as well as variations in parenting

practices that may reflect gender stereotypes (Crick & Zahn-Waxler 2003).

There is a suggestion that girls display higher levels of prosocial behaviour,

which could potentially protect them against the display of behaviour difficulties

(Messer, Goodman, Rowe, Meltzer & Maughan 2006). One study has also

70

argued that boys and girls may be exposed to other risk factors for behaviour

difficulties to different degrees, which could explain the differences in some of

these findings (Storvoll & Wichstrom 2002). As the evidence presented above is

fairy unequivocal, the present study predicts that males will experience more

significant behaviour difficulties than females.

2.2.4 Ethnicity

A number of researchers have investigated the extent to which a child’s

ethnicity impacts upon their behaviour difficulties displayed, and whether it can

be noted as a fixed risk factor. Some studies have argued that clear ethnic

differences exist in the display of behaviour difficulties (Zwirs et al. 2011).

However, this is not a consistently held view, with others finding limited effects

of this variable in influencing problem behaviours displayed, (Dekovic, Wissink,

& Meijer 2004).

Within the UK, evidence from the millennium cohort study (Brown & Schoon

2008) found significant differences between children from different ethnic

backgrounds in their prevalence of behaviour difficulties. These children were

assessed at age 7 by being rated on the Strengths and Difficulties

Questionnaire (SDQ) (Goodman 2001), by their mothers. Results revealed that

Black African children had the least behaviour problems; White children also

had low levels of behaviour difficulties with a mean score below the overall

mean. However, the mean scores for Pakistani, Bangladeshi and Black

Caribbean children were considerably higher than the overall population mean.

As this study used a large representative sample, it provides strong evidence

that ethnic difference in the prevalence of problem behaviours in childhood do

exist. Nonetheless, Greenberger, Chen, Beam, Whang & Dong (2000)

suggested that differences in problem behaviour between ethnic groups may

not emerge until later on in development, and in fact well into adolescence.

A number of reasons have been proposed for the potential differences found. It

has been argued that definitions and measurements of behaviour difficulties

71

may vary by ethnic groups. Parents from different ethnicities may rate their

children in regard to their own cultural expectations of what is acceptable and

developmentally appropriate behaviour, and these may not be consistent or

comparable across all ethnic groups (Guttmannova, Szanyi & Cali 2007).

Furthermore, ethnic differences in behaviour difficulties could actually reflect

and be a response to parenting practices. There is evidence to suggest that at

least one aspect of parenting; physical discipline varied according to ethnicity

(McLoyd & Smith 2002), and this could account for the behaviours displayed.

With regard to the present study, ethnic difference may have some influence on

behaviour difficulties displayed, although whether or not significant findings are

discovered may depend upon whether the sample sizes for each ethnic group

are large enough to be included within any analysis.

2.2.5 Academic achievement

A fairly prominent finding from previous research is that academic achievement

and behaviour problems are related to one another, (Hinshaw 1992, Reid,

Gonzalez, Nordness, Trout, & Epstein 2004). Evidence suggests that those

children who have a reading problem (Morgan, Farkas, Tufis & Sperling 2008),

or generally a lower level of academic performance (McIntosh, Flannery, Sugai,

Braun & Cochrane 2008) will be more likely to display behavioural difficulties

than those children who experience academic success at school. These effects

have also been investigated longitudinally, with findings showing that their

influence is significant and long-term. Specifically Morgan et al. (2008)

investigated whether children in their first year at school with a reading problem

would be more likely to display a behaviour problem two years later. After

controlling for a number of possible confounding variables they found this to be

the case.

Other studies, however, have suggested that it is the display of behaviour

problems that actually predicts poorer academic achievement. Breslau, Breslau,

Miller & Raykov (2011) conducted a longitudinal study aiming to assess the

effects of behaviour problems at ages 6 and 11 on later academic achievement.

72

They demonstrated that problems at each of these ages were independent risk

factors for academic achievement even after controlling for a number of

potentially confounding variables. They concluded that children who display

behaviour problems early in their developmental history will experience long

lasting effects on their academic achievement. Support for this argument comes

from McLoed & Kaiser (2004) who demonstrated that behaviour problems at

ages 6-8 predicted a decrease in the probability of gaining a high school

degree, as well as enrolment in to further education.

The relationships between academic outcomes and behaviour difficulties are

clearly evident in the literature, what is currently being debated are whether

academic problems are a cause for behaviour difficulties or whether it is the

display of behaviour difficulties that impinges on academic outcomes. It has

been suggested that a child with behaviour problems will display behaviours

that prevent them from making and sustaining social relationships with peers

and teachers, and it is this that untimely impinges on their learning process and

the gaining of academic skills, (Reid et al. 2004, Breslau, et al. 2011). Morgan

et al. (2008), on the other hand, highlights that academic difficulty could be the

cause of behaviour problems, as children begin to ‘act out’ as they become

increasingly frustrated with their perceived failure on a task. A third plausible

explanation is that poorer levels of attention could be a cause for both

behaviour difficulties and academic problems. From the discussed literature

presented above, there appears to be a fairly strong relationship between these

variables and therefore academic failure or underachievement is predicted to be

a variable risk factor for behaviour difficulties within the present study.

2.2.6 Attendance

Research has suggested that children and adolescents with poorer levels of

school attendance and consequently higher truancy levels or unauthorised

absence have a negative impact on pupil outcomes, (Wilson, Malcom, Edward

& Davidson 2008). Specifically previous research has showed a relationship

between poor attendance and academic underachievement, difficulty making

73

friends, loss of confidence, stress and low self-esteem, as well as some kinds of

behaviour problems, (Malcolm, Wilson, Davidson & Kirk 2003).

Some research has shown that children who are poor attendees at school

(having higher unauthorised absences) experience more significant behaviour

problems (Miller & Plant 1999). Indeed, higher levels of truancy have also been

considered an early warning sign for delinquency and substance abuse, (Yeide

& Kobrin 2009). A great deal of this research, however, has focused specifically

on smoking, alcohol and drug abuse, (McAra 2004, Chou, Ho, Chen, Chen

2006, Henry 2007, Henry & Huizinga 2007, Henry & Thornberry 2010), rather

than behaviour difficulties in schools. Nonetheless all these studies showed that

poorer attendance was associated with higher levels of drug abuse, smoking

and underage drinking, all considered problem behaviours.

Although the evidence appears to suggest that children missing school through

truancy are at an increased risk of displaying problem behaviours such as drug

abuse, the relationship between these two variables maybe more complex. It

could be argued that those with behaviour problems truant more and truancy is

a consequence and not a cause of these problems. Nonetheless, others have

argued that children who are truanting will have more time unsupervised by

adults and therefore have an increased likelihood of engagement in such

problem behaviours (McAra 2004).

In evaluating this research note should be taken of the difference between

absences that parents are aware of (although not preventing) and absences

that parents are unaware of (Wilson et al. 2008). These different ‘types’ of

truancy could affect whether it emerges as a significant risk factor for behaviour

difficulties. With a plethora of research showing that higher levels of absence

from schools are associated with delinquent and substance abuse behaviours,

it is conceivable that despite limited research children who truant more often will

be at an increased likelihood of displaying behaviour difficulties in school. Poor

attendance can therefore be seen as a variable risk factor.

74

2.2.7 Positive relationships

Research has suggested that a child’s relationship with their teacher has a

significant impact on their behavioural outcomes, and can be conceived as a

variable risk factor. Hamre & Pianta (2001) established that teacher child

relationships are unique predictors of later behaviour outcomes in children.

Furthermore, Silver, Measelle, Armstrong & Essex (2005) have showed that

more conflict between teachers and children, specifically during school

transition, increased the likelihood of earlier development of externalizing

behaviour problems. It has been argued that certain qualities of teacher-student

relationships are able to predict adjustment outcomes for children in school,

(Baker, Grant & Morlock 2008). Specifically relationships that comprise of

higher degrees of trust, warmth and less conflict are related to more positive

outcomes for children. Indeed positive teacher classroom relationships are

characterized by closeness between the child and teacher, whereas negative

teacher classroom relationships are characterized by conflict and dependency

(Buyse, Verschueren, Doumen, Damme & Maes 2008).

Although some studies have found a relationship between behaviour difficulties

and child-teacher relationships, (Buyse et al. 2008, Ladd & Burgess 1999), it is

not clear in which direction the effect lies, and whether poor teacher-pupil

relationships are a cause or consequence of behaviour difficulties. Some

researchers have demonstrated that this relationship is in fact bidirectional with

both concepts influencing one another. Doumen et al. (2008) showed that

aggressive behaviour at the beginning of the school year resulted in an

increase in child-teacher conflict that in turn lead to further increases in

aggressive behaviour at the end of the school year.

Researchers have also been interested in explaining how and why child-teacher

relationships can influences behaviour difficulties. One suggestion is that

children with more secure attachments with their primary care giver will be able

to form more secure relationship with other adults such as teachers, which in

turn leads them to more positive behavioural outcomes, (Hamre & Pianta 2001).

75

There is also evidence to suggest that poor relationships with peers can be a

variable risk factor for the display of behaviour difficulties. Silver, et al. (2010)

showed that negative peer relationships increased the likelihood that an

individual would display higher levels of behavioural difficulties. In a review of

the literature specifically regarding children with learning difficulties, it was noted

that positive peer relationships may act as a promotive factor for behavioural

competence, and therefore reduce the risks of displaying behaviour difficulties

(Wiener 2004).

Further evidence for this relationship comes from Laird, Jordan, Dodge, Pettit, &

Bates (2001), who conducted a longitudinal study to assess the influence of

rejection by peers in middle childhood and being involved with anti-social peers

in adolescence on the development of behaviour difficulties. They showed that

being rejected by peers in middle childhood was a more important factor than

involvement with anti-social peers in predicting display of behaviour difficulties.

It may be that children who are rejected by their peers and have more negative

classroom peer relationship are at a heightened risk of developing behaviour

difficulties. In support of this view, Dodge et al. (2003), have also found that

peer rejection was a predictor of increased aggression.

Evidence has been presented as to why positive peer relationships are of

particular importance in reducing the risk of behaviour difficulties. It is

suggested that they are able to moderate some of the risks associated with

behaviour difficulties particularly family adversity (Criss, et al. 2002). From the

evidence discussed above within the present study it is predicted that poor

teacher and peer relationships will be a significant variable risk factor for

displaying behaviour difficulties.

2.2.8 Bullying

Research has suggested that children involved in bullying as either victims or

bullies are at risk of displaying behaviour difficulties (Gini 2008, Kim, Leventhal,

Koh, Hubbard & Boyce 2006). Bullying is typically defined as the display of

76

aggressive acts that are intended to harm another person. They are typically

repeated over time and occur in interpersonal relationships where there is an

imbalance of power (Olweus 1993, Smith 2004). Bullying may involve both

physical and verbal behaviours as well as socially excluding others (Kim et al.

2006). It is not surprising that those who are rated as bullies have the most

severe behaviour difficulties, as bullying behaviour at least to some extent, is

comprised of aggressive acts that are also a measure of behaviour difficulties.

Wolke, Woods, Bloomfield, & Karstadt (2000) showed that from a sample of 6-9

years olds those children that were involved in direct bullying had significantly

higher scores on conduct problems, hyperactivity, peer problems, and the total

behaviour problems scale, compared with those children who were not involved

in bullying. Further support comes from Hampel, Manhal & Hayer (2009) who

showed children characterized as bullies also showed a higher likelihood of

problem behaviours. In a study by Kim et al. (2006), children who had either

been bullies or victim-bullies at a baseline measure, were significantly more

likely to display more aggressive behaviour ten months later. In their analysis

they were able to provide evidence that bullying behaviour was not a just a

correlate to behaviour problems but a cause of these problems.

In terms of being a victim to bullying incidents evidence seems fairly consistent

that these children are at a significantly higher risk of displaying behaviour

difficulties, than those not involved in bullying incidents. Gini (2008) conducted

an investigation where primary school children completed a self-report about

their experiences of bullying whilst teachers completed the SDQ on their behalf.

Findings showed that victims, compared with those not being bullied, had a

higher likelihood of displaying problem behaviours. Further evidence has

suggested that being a victim of bullying is associated with mental health

problems including aggressive and violent behaviour (Arseneault, Bowes &

Shakoor 2010). It has been suggested that victims are at risk of displaying

behaviour difficulties as they may suffer from maladaptive coping responses

which lead to increases in emotional and behavioural problems (Hampel et al.

2009). Specifically within a SEND population, Hebron (2012) has noted that

victims of bullying are most likely to be children classified with either BESD or

77

ASD, and it is within these groups of SEND where children display the most

severe behaviour difficulties.

Evidence suggests that those children who are rated as ‘bully-victims’ (i.e.

sometimes the perpetrators of aggressive acts and sometimes the victims of

bullying incidents) are the most at risk group in terms of adjustment problems

(Gini 2008). In relation to the present study it is therefore predicted that children

who are bullies as well as children who are victims of bullying and bully-victims

will be more likely to display behaviour difficulties than those children not

involved in these incidents. Involvement in bullying is therefore seen as a

variable risk factor.

2.2.9 Special Educational Needs and Disabilities

The category in which a child with SEND is placed is likely to be related to their

behaviour difficulties displayed. Children are placed on the SEND register,

within one of four broad areas of need, and these can be broken down in to

eleven more specific categories of need (DfE 2001a) (see section 1.3.1 for an

overview). One of these categories has been termed Behavioural, Emotional

and Social Difficulties (BESD). Children who have any difficulties with their

behaviour, such as displaying temper tantrums, being aggressive, disruptive or

defiant are placed into this category (DfES 2003).

It is probable therefore that this group of individuals will be the most likely to

display behaviour difficulties, and a child with severe behaviour difficulties as

measured in the present study would likely be placed in the BESD group on the

SEN register. Behaviour difficulties as measured within the present study may

be a synonymous term with an SEND measure, and as such it would be

inappropriate to use the SEND category variable as a predictor of behaviour

difficulties. Nonetheless, there is some concern, that the BESD group will not be

a reliable predictor of behaviour difficulties, as children could be included within

this group on the basis that they are displaying internalising behaviours such as

anxiety and depression, as well as, or in isolation to, behaviour difficulties

78

displayed (DfES 2003). Children displaying these types of behaviours are not

measured within the present study’s construct of behaviour difficulties, which

may cloud the findings as whether SEND category (and specifically being the in

BESD group) is a useful predictor of behaviour difficulties displayed.

An interesting investigation will be carried out here; by assessing whether being

in the BESD group as defined by the SEN code of practice is related to

behaviour difficulties measured in the present study. A strong relationship would

reflect these may be synonymous constructs. If such a relationship is

established between BESD and behaviour difficulties measured within the

present study, it may be appropriate to remove this variable, and rerun the

models comparing and contrasting the findings. If the BESD group fails to have

a strong relationship with behaviour difficulties in the present study, this could

be a more worrying finding, suggesting that BESD is not a good indicator of

whether a child has behaviour difficulties.

The level of support that a child with SEND receives could also be considered a

related factor with behaviour difficulties. For an overview of level of support for

SEND, see section 1.3.1. It could be conceived that those at the higher level of

support (i.e. those statemented), are more severely affected by their SEND than

those at low levels (i.e. those at school action plus or school action), as pupils

move through the levels of support as their needs become more severe (DfES

2001a, Frederickson & Cline 2009). Therefore those with the most severe

behaviour difficulties are likely to be statemented. Nonetheless, it has been

noted that level of support may not be an accurate reflection of a child’s severity

of need but in fact a schools ability to cope with the need that they present. As

such there is considerable variation across schools in how they define and

assess a child’s level of SEND (Ofsted 2005). There could be an argument that

as the most severely affected children are receiving the most amount of

support, this could act as a protective factor and reduce their likelihood of

displaying behaviour difficulties. In such a case those children at school action

may be the most at risk of displaying behaviour difficulties, as they are receiving

less support. The present study predicts that this variable will have some

79

influence on behaviours displayed but exactly which group of children are most

at risk remains to be established.

2.2.10 Other individual level risk factors

This review of risk and promotive factors around behaviour difficulties is not an

exhaustive one. The factors highlighted above are some of the most salient;

however, there are a plethora of other significant predictors of behaviour

difficulties. Although these go beyond the scope of the study, they are

acknowledged nonetheless to highlight the fact that considerable diversity in

risk and promotive factors for behaviour difficulties exist within the literature. For

example low birth weight, (Hille et al. 2001), certain personality traits (Viding,

Frick, & Plomin, 2007), low self-esteem, (Schonberg & Shaw 2007)

hyperactivity (Thorell & Rydell 2008), lack of involvement in school activities

(Mahoney 2000) and association with delinquent peers (Laird et al. 2001) all

pose risks for behaviour difficulties. There are also genetic effects (Arseneault

et al. 2003) hormonal influences (Book et al. 2001) and neurological influences

(Dodge & Petit 2003) that predict behaviour difficulties at the individual level.

80

2.3 Family level risk and promotive factors

2.3.1 Socio-economic status

There is evidence to suggest that family socio-economic status (SES) has a

substantial influence on a number of childhood outcomes. In a comprehensive

review on SES and its effects on child development, Bradley & Corwyn (2002)

demonstrate the importance of this construct in predicting health, cognitive and

socio-emotional outcomes in children. Specifically in terms of behaviour

outcomes research has been fairly consistent in showing that children who are

living in families of lower SES are at a greater risk of developing behaviour

problems than those living in more affluent families (Propper & Rigg 2007,

Brooks-Gunn & Duncan 1997).

Brown & Schoon (2008) have also showed that children who were classified as

living in households of poverty had significantly higher mean total difficulty

scores, (as measured on the SDQ), compared with children not classified as

living in such circumstances. Characteristics of these families included parents

who had no formal qualifications and who were not working. In another large-

scale study, Green et al. (2005) also demonstrated that children identified as

having a conduct disorder were more likely to live in families that had lower

income, with parents who had no qualifications, and who were also not working.

Further support for this view comes from a non-correlational study using a

quasi-experimental design which demonstrated that children living in lower

income families were at an increased risk of developing a behaviour difficulty

(D’Onofrio, et al. 2009). Finally, in a comprehensive review of studies

investigating SES effects on child behaviour, Qi & Kaiser (2003) concluded that

around 30% of pre-school children who are living in low SES backgrounds

experience some behaviour difficulty compared with 3-6% of children in the

general population. Family income can have substantial effects on childhood

well-being, and those children living in poverty for longer amounts of time

appear to be at the greatest risk of these negative outcomes (Brooks-Gunn &

Duncan 1997). Poverty has also been demonstrated to have differential effects

81

depending on when it occurs, with onset during early school years particularly

detrimental (Brooks-Gunn & Duncan 1997).

Evidence appears fairly consistent that lower socio-economic status is

associated with behaviour problems. Research has therefore begun to

investigate why this is the case and how SES exerts its effects on behaviour

difficulties. Poorer children are often exposed to more negative influences in

their immediate environment than their more affluent peers, including more

familial stress, chaotic/unstable households, poorer parenting behaviours, and

lack of cognitive stimulation. It may be the accumulation of these risks that

results in behaviour difficulties (Evans 2004). There may be a mediating role of

some of these variables whereby lower SES and poverty causes chronic stress

that impacts on parenting skills that ultimately affect child behaviour (Jensen

2009).

Within the school context, socio-economic status is measured by whether a

child is eligible for receiving free school meals (FSM). This is assumed to be a

family level factor as FSM is calculated on parental income; nonetheless this

variable will be included at the individual level to allow for a more coherent

analysis. Within the context of the present study SES is noted as a variable risk

factor and it is predicted that children of FSM status will be at a higher risk of

displaying a behaviour difficulties than their peers.

2.3.2 Other familial level risk factors

There are numerous additional familial level risk and promotive factors for

behaviour difficulties which are beyond the scope of this study. Important

familial level factors include poor child-parental attachment (Moss, et al. 2006),

lack of parental monitoring (Pettit, Laird, Dodge, Bates & Criss 2001), parental

mental health problems (Kim-Cohen, Moffitt, Taylor, Pawlby, & Caspi, 2005)

inconsistent or harsh discipline (Chang, Schwartz, Dodge, & McBride-Change

2003), and lack of parental involvement in their child’s education (Hill et al.

2004).

82

2.4 School level risk and promotive factors

2.4.1 School urbanicity

School location, and whether it is found with an urban or rural setting, has been

investigated as a risk factor for behaviour difficulties. It has been suggested that

within urban schools children will be more exposed to anti-social and

aggressive role models. The result of such an effect is that a larger number of

individuals within the schools will be influenced by these problem behaviours,

leading to more behaviours difficulties displayed within the school, (Hope &

Bierman 1998). Support for this evidence comes from a number of studies,

reporting that schools situated in urban areas are better predictors of more

severe behaviour difficulties, than schools in rural areas, (Stewart 2003,

Larsson & Frisk, 1999, Hope & Bierman 1998). Urban schools with higher rates

of behaviour difficulties have similarly been suggested to reflect the higher rates

of violence already displayed in urban community settings (Warner, Weist &

Krulak 1999).

Despite this evidence, not all research is in support of the view that urban

schools are associated with poorer outcomes for students. Sellstom &

Bremberg (2006), in a review of a number of studies investigating school effects

on pupil outcomes, suggested that schools within urban locations actually had

more positive effects on pupils than rural schools. This effect may result from

more resources being made available to inner city and urban schools. This

evidence could also reflect the fact that studies included within this review

measured a broad range of outcomes for children rather than specifically

behavioural ones. Nonetheless, this evidence that rural school students are not

better off than their urban counterparts is consistent with another group of

studies reporting no differences between urban and rural students, in terms of

alcohol use (Wilcox-Rountree & Clayton 1999), and fighting behaviour, (Swahn,

Gaylor & Bossarte 2010). The evidence discussed above is fairly mixed and no

clear pattern has emerged in terms of which school type will have the most

influence on pupil behaviour difficulties, as such within the present study it is

83

predicted no significant difference will emerge between rural and urban schools

in their behaviour difficulties displayed.

2.4.2 School size

School size (in terms of numbers of pupils on roll) is potentially a significant

predictor for individual behaviour difficulties. There is some evidence suggesting

that larger schools are a risk factor for behaviour problems and violence

(Stewart 2003, Haller 1992, George & Thomas 2000), as well as victimisation

(Bowes et al. 2009). Although victimisation is not a ‘type’ of behaviour difficulty,

it could be considered associated with behaviour problems, (i.e. the children

displaying behaviour difficulties cause others to experience victimization).

Furthermore, victimization has been showed to be interrelated with a deviant

lifestyle (George & Thomas 2000).

Other studies however, have either found no effect of school size on student

victimisation, (Khoury-Kassabri, Benbenishty, Astor & Zeira 2004), or found a

relationship between smaller schools and student victimisation (Gottfredson &

DiPietro 2011). In terms of other types of behaviour difficulties; no relationship

has been found with aggression and school size (Wilson 2004) or smoking and

drinking behaviour in adolescents and school size (Maes & Lievens 2003).

The inconsistency in these findings could reflect the fact that larger schools

often have more comprehensive policies and interventions for students with

behaviour difficulties, and therefore reducing the risks. However, smaller

schools will be able to develop more positive relationships between teachers

and students, hence also countering some of the potential risks (Gottfredson &

DiPietro 2011). The evidence presented here remains fairly inconclusive,

therefore it is predicted in the present study that school size will have a limited

effect on behaviour difficulties displayed.

84

2.4.3 School level SES

As stated earlier, socio-economic status (SES) is often calculated within

educational research by assessing whether the family has eligibility to, and is

claiming free school meals for their child (Hobbs & Vignoles 2007). At a school

level, the percentage of the total number of students claiming free school meals

(FSM) is calculated and used as a school level variable of SES. A higher

percentage would indicate that a greater number of students are claiming FSM

and the overall level of SES in the school is low.

High average SES within schools is generally associated with positive

outcomes for students (Sellstom & Bremberg 2006), and there is considerable

evidence to suggest that low average school SES is a significant risk factor for

childhood and adolescent behaviour difficulties. Barnes, Belsky, Broomfield &

Melhuish (2006) showed that a measure of school disorder, (including

measures of bullying, exclusions and aggressive behaviour displayed by

pupils), was linked to schools that had higher percentages of children claiming

free school meals. Other studies have found that higher proportions of students

whose families are claiming FSM in the school are associated with increased

levels of delinquency (Felson, Liska, South & McNulty 1994), higher levels of

victimisation (George & Thomas 2000), (which could be considered a proxy for

behaviour difficulties), and increases in weapon carrying into school (Wilcox &

Clayton 2001), (although not a direct measure of behaviour difficulties, likely to

be linked with anti-social behaviour). Specifically within the UK context a recent

government report found a positive relationship between the number of

exclusions in schools (presumably for some kind of behaviour problem) and the

percentage of students claiming FSM within the school, (DfE 2012). Not all the

evidence is consistent, however as Stewart (2003) showed that although a

number of school level variables can account for school misbehaviour, poverty

levels were not significant in predicting behaviour problems at the individual

level.

85

Some discussion needs to take place as to why this variable may predict the

display of behaviour difficulties at the individual level. It could be quite obvious

that schools with a higher overall level of students claiming FSM, will comprise

of more individuals of lower SES, and this is a known risk factor for behaviour

difficulties at the individual level (Propper & Rigg 2007). Alternatively, a

speculative argument suggests schools with a higher percentage of students

claiming FSM may create a culture or climate within the school of dependency.

This could have a negative influence on other children, who are not eligible or

claiming FSM, resulting in more behaviour difficulties displayed. Although the

exact mechanisms by which this variable has an influence on behaviour

outcomes is unknown, the evidence discussed above generally shows lower

school SES is associated with an increase in behaviour difficulties and this is

what is predicted for the present study.

2.4.4 School level EAL composition

Evidence has been presented to suggest that the percentage of children within

a school speaking English as an additional language (EAL) may pose a

significant risk factor for behaviour difficulties. One study showed that the

number of children in a class with EAL was of particular importance at the

school level in accounting for aggression in young children starting school. This

study found that schools with a higher proportion of students learning English

as an additional language had lower individual aggressive scores, (Kohen,

Oliver & Pierre 2009). However, another study investigating individual, school

and neighbourhood effects on student aggressive behaviour failed to find any

impact from a school level variable of percentage of EAL in the school (Barnes

et al. 2006). Other evidence has shown that schools with 50% or more students

from ethnic minority backgrounds (and presumably reflecting higher levels of

EAL) had significantly more students with behaviour problems compared with

school with less than 50% ethnic minority students, (Ma, Truong & Strurm

2007).

86

A discussion needs to take place to decipher whether it is the aggregation of

individual children with EAL at the school level which maybe a risk or promotive

factor for behaviour difficulties, or whether schools with a higher percentage of

EAL have a distinct culture or climate which affects other non EAL students

behaviour difficulties. Some studies have shown individuals with EAL are more

at risk of behavioural problems (Brown & Schoon 2008). If these individual

effects were aggregated to a school level it could explain why percentage of

EAL students is a potential predictor for behaviour difficulties. Nonetheless

there is an argument that schools with higher EAL students may adopt a more

inclusive ethos and celebrate difference and diversity. This may have a positive

effect on all students and reduce the risk of displaying behaviour difficulties.

From the evidence discussed above the present study predicts the percentage

of students with EAL within a school is unlikely to have a significant effect on

behaviour difficulties.

2.4.5 Proportion of SEND pupils

Another potential risk factor at the school level could be the proportion of pupils

with special education needs or disabilities (SEND) within the school. One study

that has investigated SEND proportion at a school level, (Barnes et al. 2006)

showed a positive relationship between a measure of school disorder, including

aggressive behaviour displayed by pupils and more children on the SEND

register.

It would be interesting to note whether percentage of SEND children in the

school is a predictor for behaviour difficulties due to the individual-level risk

being aggregated up to a school level, or whether there is something

qualitatively different about this variable when assessed at a school level. As

already mentioned in section 2.2.9 at the individual level children with a SEND

are more at risk of displaying a behaviour problem than those children without,

(Green et al. 2005). It could therefore be conceived that if these influences are

aggregated up to a school level, more children with SEND will result in more

behaviour difficulties. However, the question is also asked whether the impact

87

of having more children with SEND influences other students in displaying

behaviour problems.

A number of inclusion based studies have assessed the impact on the ‘other’

students who are in classes with more SEND children. Some studies have

suggested that the inclusion of more pupils with SEND into the classroom has

no effect on behaviour outcomes of the other students (Sharpe, York & Knight

1994, Tapasak & Walther-Thomas 1999). Other studies however, have noted

significant negative effects on pupils with higher proportions of SEND pupils

included within their classrooms (Daniel & King 1997, Brown 1982). Although

these studies were investigating the impact of behaviour problems on children

specifically without SEND they do go some way to explain how this school level

factor may operate in being a predictor for behaviour difficulties. In a review

investigating the effects of inclusion on other pupils and incorporating some of

the studies above, Kalambouka, Farrell, Dyson & Kaplan (2005) suggest that

the impact of including children with SEND will not have a major impact on the

majority of pupils in the school without SEND, however this report does

acknowledge that the impact on behaviour outcomes is more pronounced than

that of academic outcomes.

It is perhaps logical to suggest therefore that a higher number of SEND pupils in

the school will lead to an increase in behaviour difficulties. This may reflect an

aggregation effect of the individual level but also that a higher proportion of

SEND children in a school can affect other students. This could be in a negative

direction, placing an extra burden on teachers who have less time to spend

supporting other students. Alternatively schools with more SEND pupils may

adopt more inclusive policies and have access to resources within them that

creates a more constructive climate which affects other pupils in a positive

direction. The present study predicts that a school’s percentage of SEND pupils

will be a significant predictor of behaviour difficulties although the direction of

this relationship awaits further research.

88

2.4.6 School level achievement

The average academic achievement level within a school has been assessed

as an important influence on an individual’s behaviour difficulties. It has long

been argued that an ethos within a school that promotes academic

achievement is associated with lower levels of problematic behaviour in schools

(Rutter, Maughan, Mortimore, Ouston & Smith 1979). Furthermore, studies that

have used a ‘value added 5

’ measure have suggested that value added

education is associated with lower levels of adolescent substance use (Bisset,

Markham & Aveyard 2007). Mooij (1998) has also suggested that attending a

lower achieving secondary school is a particularly important school level risk

factor for aggressive behaviour. A further study has shown that measures of

school disorder including aggressive behaviour and bullying displayed by pupils

was related to lower academic levels in key stage 1 and key stage 2 (Barnes et

al. 2006). As this study was correlational, causation cannot be inferred and it is

entirely possible that lower academic standards in the school were the

predictors of more school level disorder. Nonetheless this evidence taken

together does suggest that an individual’s behaviour difficulties are negatively

affected by attending low-achieving schools. In higher achieving schools pupils

with SEND may benefit by having peers around them that are more able to help

them with their academic work, as such any frustration they have with their work

would be reduced leading away from the display of behaviour difficulties.

This evidence has however, been disputed by others researching in this field.

Wilson (2004) has argued that school performance measures are not related to

aggression scores. A further alternative view is that attending a higher

achieving school can actually have a detrimental effect on pupil behaviour.

Felson et al. (1994) found that students would be more likely to engage in

delinquent behaviour when the school had a culture of high academic

achievement. The authors suggest that in such competitive climates when a 5 Schools show evidence of ‘value added’ education when the average overall student performance in a school is better than would be expected from their particular socio-demographic composition.

89

student does not have the capability to compete with others they ultimately

suffer in comparison to their classmates. This could lead them to display more

behaviour difficulties as they are suffering from increased pressure and a sense

of failure. It is likely that school level academic achievement will have a

significant effect upon an individual’s behaviour difficulties although it is unclear

in which direction the effect will have.

2.4.7 School level attendance

Overall attendance rates are a concern for all schools. Improving these rates

will not only allow schools greater standing in any league tables, but also

improve the academic and behavioural outcomes for their pupils. Evidence

clearly shows that children who are not regular attendees to school are more at

risk of developing behaviour problems than children who are regular attendees

(Yeide & Kobrin 2009). A pertinent question is whether or not children who are

in classes and schools with poorer overall rates of attendance are at a greater

risk of developing a behaviour problem than they would be if in a school with

higher overall attendance.

A number of studies have investigated school level attendance rates and their

influence upon pupil outcomes. These studies have utilised the concept of

truancy; generally defined as “any unexcused or undocumented absence from

school” (Claes, Hooghe & Reeskens 2009, page 124). Truancy rates or

unauthorised absence rates at a school level have been found to a have

significant effects on other pupils within the school. Wilson et al. (2008)

investigated effects of truancy on other non-truanting pupils finding that

although some pupils said they were not affected by those truanting, a

predominant emotion to emerge was irritation at these pupils and disruption in

class when they came back and needed additional help with their work. There

was also concern that the poor attendees could become negative role models

for the better attendees. Furthermore, these schools effects have also been

noted in terms of influencing school performance (Claes et al. 2009). Although

these studies are not directly measuring behaviour problems displayed they do

90

suggest that school-level truancy impacts upon individual students within the

school.

Some studies have begun to address the issue of school level attendance on

individual behavioural outcomes. Bisset et al. (2007), investigated adolescent

substance use (an aspect considered to be related to behaviour problems), and

whether this was influenced by what they termed ‘value added’. A school would

achieve this status when their students achieve higher academic outcomes and

truant less than would be expected from their particular socio-demographic

composition. They found evidence that value added education was associated

with lower levels of alcohol consumption and illegal drug use. In support of this

evidence that school truancy levels can impact upon pupils behaviour

difficulties, Maes & Lievens (2003) argue that truancy levels between different

schools could account for differences in smoking and drinking behaviour. They

suggested that truancy is not solely an individual characteristic, but a school-

related one as well, and therefore smoking and drinking behaviour is influenced

by school truancy levels which perhaps reflects school climate. Within the

present study it is predicted that higher levels of school-unauthorised absences

will be related to an increase in behaviour difficulties.

2.4.8 School level behaviour difficulties

Whether an aggregated level of behaviour difficulties at a school level has any

influence on individual behaviour has generated some exposure in the

psychological literature (Hoglund & Leadbeater 2004, Kellam, Ling, Merisca,

Brown & Ialongo 1998). It has been argued that schools are vastly different in

their overall levels of behaviour difficulties displayed and these effects are likely

to have influence upon the pupils attending them (Reid 2010).

Studies have aimed to assess whether a school or indeed a classroom with a

higher overall rating of more severe behaviour problems has a negative

influence upon the individuals within them. Mooij (1998) provided some

evidence for such a suggestion that being in a class with a larger number of

91

disruptive pupils will have negative effects on individual behaviour. DeWit et al.

(2000) conducted a study to assess the effects of school culture and mediating

mechanisms on behaviour problems in adolescence. They found that school

culture (comprised of a number of concepts including school and classmate

behaviour problems) had direct effects as well as indirect effects on a number of

their measures of problem behaviour. They suggested that perceptions of a

more negative school culture are associated with more behaviour problems

displayed. On a slightly different yet related track, Theriot, Craun & Dupper

(2010), investigating school exclusion using a multilevel study, showed that

school suspension rate (calculated by dividing total exclusion by number of

pupils) is a significant predictor of individual level exclusion. Although school

exclusion is not a synonymous term with behaviour difficulties, the relationship

is likely to be very strong.

Other studies have provided compelling evidence that more aggressive

classrooms and schools have influences on individual aggression. Barth,

Dunlap, Dane, Lochman & Wells (2004) found evidence that aggregated

classroom level behaviour problems had a significant impact on students within

them over time. Specifically, classrooms rated as poorer environments with

more behaviour problems were associated with more individual level problem

behaviours. The authors suggest that peers within these contexts may become

effective role models, and reinforce the display of problem behaviours. Thomas

& Bierman (2006) provided evidence that being in a more aggressive classroom

for a greater number of years held stronger associations with individual

aggression than being in an aggressive classroom for a shorter period. Mercer,

McMillen & DeRosier (2009) showed that children within classrooms with a

higher initial aggregated score for behaviour problems reported more significant

increases in levels of behaviour problems over the year. The evidence is fairly

compelling, and therefore in the present study - children in schools with more

behaviour difficulties are at a greater risk of developing a behaviour difficulty

themselves.

92

2.4.9 Other school level risk factors

There are a number of other school level variables that have been linked to

pupils’ behaviour difficulties. A good school climate (often including quality

interactions between students and teachers) has a positive influence on pupils

outcomes, and results in fewer externalising problems (Kuperminc, Leadbeater,

& Blatt (2001). Perceptions of school fairness and student involvement has

been related to reductions in problem behaviour, (Dewit et al. 2000) and also

predicted lower levels of offending and misconduct behaviours (Welsh 2000).

Measures of school connectedness where students feel a sense of belonging to

the school have been associated with lower levels of violence and substance

use (Bond et al. 2007), and fewer risk taking and problem behaviours (Denny et

al. 2011). Finally, schools that have effective policies on behaviour and anti-

smoking and drugs appear to experience fewer school level problem

behaviours than schools without effective policies in these areas (Evans-Whipp

et al. 2004).

93

2.5 Neighbourhood level risk and promotive factors

2.5.1 Neighbourhood disadvantage

Risk factors associated with behaviour difficulties have also been

acknowledged at a neighbourhood level. Neighbourhoods are geographical

areas in which groups of people reside, where people experience shared social

environments (Drukker, Kaplan, Feron & van Os, 2003). It is therefore probable

that these shared social environments will have some influence upon the

individuals living within them. In an extensive review of the literature within this

area Leventhal & Brooks-Gunn (2000) have argued that of all the

neighbourhood factors, neighbourhood disadvantage or neighbourhood socio-

economic status as it is sometimes referred to holds the strongest

neighbourhood influence over behavioural outcomes of children and

adolescents. Indeed, Edwards & Bromfield (2010) showed that significant

differences emerged between children living in more deprived neighbourhoods

compared with those from more affluent neighbourhoods across a number of

domains including, peer problems, emotional problems and hyperactivity.

Evidence is fairly consistent that children living in more disadvantaged

neighbourhoods are more at risk of developing behaviour problems (Schonberg

& Shaw 2007, Drukker et al. 2003, Stewart, Simons & Conger 2002). It has

been suggested that living in poorer communities increases the risk of

behaviour difficulties as within such areas there is a lack of institutional

resources available, poorer quality housing and greater exposure to anti-social

peers. Others have suggested that effects such as poverty, crime and violence

often found in lower SES neighbourhoods can be mediated through another

variable such as parental stress (Schonberg & Shaw 2007).

94

2.5.2 Other neighbourhood level risk factors

Other important neighbourhood factors have included residential instability, with

those living in communities with higher levels of instability (i.e. with large

amounts of changes in house occupancy within short periods of time) are more

at risk of problem behaviours, (Brooks-Gunn, Duncan, Leventhal & Aber 1997).

Furthermore, exposure to violence is an important issue, those children more

exposed to neighbourhood violence have an increased likelihood of engaging in

violent behaviour themselves (Patchin, Huebner, McCluskey, Varano, & Bynum

2006). Finally, children and adolescents who have greater exposure to anti-

social peers within their neighbourhood are more likely to engage in anti-social

behaviour (Ingoldsby et al. 2006).

95

2.6 The relative importance of the individual versus the school ecological level in accounting for behaviour difficulties

There is a fairly unanimous view amongst researchers that the most salient

predictors of behaviour difficulties are found within more proximal ecological

levels such as those within the individual, rather than more distal contextual

influences such as the school influences (Andersson, et al. 2010). Nonetheless,

it has been suggested that attempting to explain these difficulties by accounting

solely for factors considered proximal to the child and excluding the influence of

the contextual factors is a poor oversight (Aveyard et al. 2004). Contextual

influences have significant influences on childhood developmental outcomes

(Theriot et al. 2010) and particularly in the development of behaviour difficulties

(Mooij 1998).

Over thirty years ago it was suggested that one such contextual influence i.e.

the school environment had a powerful influence on school behaviour (Rutter et

al. 1979). It is only fairly recently, however, that the effects of the school context

on childhood behaviour difficulties has gained greater attention within the

academic literature (Sellstrom & Bremberg 2006). It still appears that relatively

little is known in terms of exactly how school environments impact on childhood

developmental outcomes, (Maes & Lievens 2003). Furthermore, the relative

strength of school level factors compared with individual level factors in

predicting the display of behaviour difficulties has not been extensively

investigated.

It has only been fairly recently that appropriate statistical measures have been

applied to this field, which have used hierarchical linear modelling techniques

i.e. multi-level modelling (Twisk 2006). These techniques are able to

acknowledge that data is organised hierarchically, with pupils being found

nested within schools, (and indeed schools being nested within

neighbourhoods). The hierarchical structure of this data needs to be

acknowledged in order to tease apart the relative influences of each level and

assess how important each level is in accounting for childhood and adolescent


96

Studies that have used these techniques have been fairly consistent in their

findings, suggesting that school level variables can be significant risk factors for

the display of behaviour difficulties, although once individual level variables are

accounted for their relative influence is fairly small, (Reis et al. 2007, Maes &

Lievens 2003, Gottfredson & DiPietro 2011, Aveyard, et al. 2004). Research

that acknowledges the importance of multiple ecological levels in influencing

behaviour difficulties consistently finds the individual level accounts for the

largest proportion of this variance (Andersson, et al. 2010).

Although small in comparison to the weight of the individual level, schools are

important contextual influences in accounting for behaviour difficulties

(Gottfredson 2001). School level studies have provided evidence that this

ecological level accounts for variance on a number of behavioural outcomes.

Typically it has been suggested that the amount of variance accounted for at

the school level for individual behaviour is around 8-15% (Gottfredson 2001).

This figure is consistent with Felson et al. (1994) who argue contextual

variables usually only account for between 5-10% of the outcome in question,

and Khoury-Kassabri et al. (2004) who found between 9-15% of variance was

accounted for at the school level when the outcome was student victimization.

Kuperminc, et al. (2001), assessed the effects of children’s psychological

vulnerabilities and school climate perceptions to their externalising behaviour.

They found that school climate was able to add 5% to the model explaining

externalising problems, with more positive perceptions of school climate related

to fewer behaviour problems. Kohen et al. (2009) investigated the effects of a

number of school and neighbourhood variables of different outcomes for

kindergarten children. They found that although 95% of the variance in physical

aggression was attributed to the individual/family level, 3% was accounted for at

the school level. Specifically schools that had a higher proportion of students

with English as an additional language had lower aggressive scores. The

remaining 2% was attributable to the neighbourhood level. Other studies have

found a significant but small effect at the school level. Resis et al. (2007) found

97

the school ICC 6

was about 2% with the outcome of aggressive behaviour.

Payne (2008) found an ICC of around 6.3% in delinquency, while Andersson et

al. (2010) suggested that school level factors account for 6-8% of variance in

mental health difficulties, which in fact dropped to 2-3% when confounding

factors were taken into account.

In an investigation attempting to account for variance in aggressive behaviour

by the individual, class and school level factors, (Mooij 1998), it was shown that

although the school level was important, the individual level had the strongest

effect. Furthermore the classroom level was more powerful than the school level

but not as powerful as the individual level. This evidence suggests that as the

influences become more distal on the individual, the effects become less

pronounced, as influences within the individual level have a direct effect

whereas contextual effects are often more indirect being mediated through a

third variable. Finally, Sellstom & Bremberg’s (2006) extensive review of

multilevel studies investigating the school effects on behaviour outcomes, report

that the schools effect on problem behaviour did not exceed 8% across 4

studies. Only in one study, investigating weapon carrying in school, did the

school effect reach 25%.

It appears from the evidence discussed above that schools do exert a powerful

influence on individual behaviour difficulties, although their effects are

considerably weaker than those found within the individual level. A key

objective for the present study is to calculate the amount of variance in

individual display of behaviour difficulties that is attributable at the individual and

school levels respectively, within a population of children identified as having

SEND. It is predicted within the present study that the individual level will

account for more variance in behaviour difficulties than the school level.

6 Inter Cluster Correlation (see section 5.9.4 for an explanation)

98

2.7 Summary statements and research questions

• Behaviour difficulties cannot be fully accounted for by a single ecological

level, the influence of multiple levels; including factors at the individual,

familial, school and neighbourhood level are required

• A justification has been presented for a particular focus on the individual

and school ecological levels and the relative importance of these in

accounting for behaviour difficulties.

• At the individual level variables that have been associated with behaviour

difficulties and may be particularly salient for a SEND population include,

age, relative age within a school year, gender, ethnicity, academic

achievement, attendance, positive relationships and bullying. Socio-

economic status of the individual (although technically a familial level

factor) is also an important influence.

• At a school level, school urbanicity, size, numbers of children from lower

socio-economic backgrounds, percentages of students with English as

an additional language, as well as school level overall academic,

attendance and behaviour rates are also important factors to consider

when accounting for behaviour difficulties.

• The research questions to emerge from this chapter (labelled research

questions 1a, 1b, and 1c) are presented below:







99

3. Cumulative Risk


The overall aim of this chapter is to introduce the concept of cumulative risk and

discuss the pertinent issues surrounding it. The chapter begins with section 3.2

explaining the importance of acknowledging multiple rather than single risks in

analyses. Section 3.3 describes what the cumulative risk score is and identifies

its principle feature. Section 3.4 evaluates the cumulative risk model in terms of

its evidence and theoretical concerns. Section 3.5 gives an overview of how

cumulative risk is measured. Section 3.6 discusses the functional relationship

between cumulative risk and behaviour difficulties and whether this conforms to

a linear or non-linear pattern. Section 3.7 examines cumulative risk across

ecological levels, and whether the same amount of risk in one level has the

equivalent effect when spread across multiple levels. Section 3.8 concludes

these issues and section 3.9 provides summary statements and research

questions that have emerged from the chapter.

3.2 Multiple risk factors

Research has been fairly prolific in investigating the causes and correlates of

childhood and adolescent behaviour difficulties (for an overview see chapter 2).

A criticism of some of this research however, is that these factors have often

been studied in isolation whereas in reality they are not independent of one

another and often cluster together within the same individual (Flouri & Kallis

2007). As young people often experience multiple risks within their backgrounds

and across distinct contexts which impinge on their lives (Sameroff, Gutman &

Peck 2003), acknowledging the unique influence of a single factor will provide

an insufficient account of these problems. Kraemer, Stice, Kazdin, Offord &

Kupfer (2001) argue that it is not possible to fully understand the effects of any

100

single risk factor outside of the context of all the others, and failing to account

for them increases the chances of over-emphasising the importance of any one

risk factor (Sameroff et al. 2003).

Investigating single risk factors for behaviour difficulties in isolation suffers from

another drawback in that no one factor is deemed necessary or indeed

sufficient to cause these problems (Sameroff et al. 2003). Sameroff and

colleagues cite evidence from the Framingham Study of heart disease (Dawber

1980) to explain the functioning of risk factors for childhood developmental

outcomes. In the Framingham study a number of risks such as smoking, lack of

exercise, hypertension, and obesity all contributed to heart disease, although,

no single factor in isolation was sufficient to account for the disease. It was the

combination of multiple risks which was the salient issue, as every individual

with this disease had a different array of risk factors within their background.

Other researchers within the field of developmental psychopathology, also

argue that no single factor when present can completely account for the display

of behaviour difficulties, and that these problems are the result of numerous

influences acting on the individual (Dodge & Pettit 2003, Ackerman, Izard,

Schoff, Youngstrom & Kogos 1999). The general consensus is that to further

our understanding of risk, and how these factors impact on behavioural

outcomes research needs to move beyond identifying single risk factors and

include multiple factors simultaneously within the same model, (MacKenzie et

al. 2011, Gutman. Sameroff & Cole 2003).

A number of analytical methodologies have been proposed in the literature in

order to assess the effects of multiple risk, (Jones, Forehand, Brody &

Armistead 2002, Farris, Smith & Weed 2007). One approach is to use a

regression analysis where multiple variables are put independently into the

model (Gutman et al. 2003). Placing risk variables into a multiple regression

analyses allows the unique relationships between contextual risks and problem

behaviours to be observed. This method is particularly useful for establishing

the specific risk factors that are the strongest predictors of a certain outcome as

101

well as acknowledging the combination of factors that can account for the most

variance: such models have been termed independent additive models.

This approach of inputting multiple variables independently into a regression

analysis does however, give rise to a number of shortcomings. Firstly, although

statistically significant relationships are often found between risk variables and

the outcome in question in such analyses, in isolation they are only able to

account for small amounts of variance (Sameroff, et al. 1998, Forehand, Biggar

& Kotchick 1998, Dodge & Pettit 2003, Fergusson & Horwood 2003). Secondly,

there may also be power issues which limit the number of factors that can be

included in a model at the same time (Ackerman et al. 1999). Finally, as this

method assumes variables are independent from one another, this neglects the

fact that risks are often found to cluster within individuals and are therefore

likely to be related in some way (Flouri & Kallis 2007). This type of model is

utilised within the present study to acknowledge the significant risk factors for

behaviour difficulties. Nonetheless, an alternative method is needed to

overcome some of these limitations. One such approach has been termed the

cumulative risk model.

3.3 What is cumulative risk?

The basic premise of cumulative models (Gerard & Buehler 2004a) is that by

assessing contextual risk variables in combination with one another, specifically

by adding them up to produce a cumulative risk score, will result in a better

predictive model than could be achieved if their influences were assessed

independently (Appleyard, Egeland, Van Dulmen & Sroufe 2005). There are two

basic underlying assumptions of cumulative risk models, firstly that the total

number of risk factors within an individual’s background holds a greater

influence over development than any specific risk factor or particular

combination of risk factors. Number is therefore seen as more important than

type or kind of risk, (Morales & Guerra 2006). This idea is rooted in the principle

of equifinality (Dodge & Pettit 2003), that proposes a negative behavioural

102

outcome can occur via a number of distinct pathways. It is not the specific type

of risk that is important but the number of increasing risks that an individual is

exposed to, because as stresses build up in a child’s background they

overwhelm coping mechanisms which leads to disorder, (Flouri & Kallis 2007).

The second assumption is that those individuals with backgrounds comprising

more risks for behaviour difficulties are at an increased likelihood of suffering

problems than those with fewer risks in their background; i.e. the larger the

number of risks the greater the prevalence of problem behaviour (Trentacosta,

et al. 2008). Multiple risk factors therefore influence behaviour problems by

working in a cumulative manner where the addition of each extra risk factor

results in an increase in the problem behaviour, (Atzaba-Poria, et al. 2004).

3.4 Measurement of cumulative risk

Cumulative risk approaches maintain that the contextual risk factors within an

individual’s background can be assessed, measured then summed together to

form a cumulative risk score which is used as a predictor variable (Gerard &

Buelher 2004a). These risks are not weighted and each factor is acknowledged

as having no greater importance than any other (Flouri & Kallis 2007). Although

this could be seen as a criticism, it is in fact consistent with what the model

suggests; that number rather than any single risk, regardless of its intensity, is

what best predicts an outcome, (Flouri 2008, Evans 2003). Cumulative indices

do not comprise of an exhaustive lists of risks, and there is no consistently

identified set of risks which could be considered the ‘gold standard’ and be

included within any cumulative risk model (Lima et al. 2010). Indeed the number

of risk variables that have been included within any model has ranged from 2 up

in excess of 15 (Lima et al. 2010), with larger models not necessarily better at

predicting outcomes than smaller ones, as the addition of extra variables can

sometimes create ‘noise’ that ultimately distracts from the findings (Farris et al.

2007). The indices are made up of potentially important variables that

accumulate to better predict an outcome, they may therefore be

103

interchangeable, as what is of interest is not the individual factors themselves

but their cumulative effect on the outcome.

In order to measure contextual cumulative risk, each factor or variable within the

study needs to be defined as a ‘risk variable’. Usually risk variables are defined

as such when the correlation between them and the dependent variable is

significant (Lima et al. 2010) or equal to or exceeds .25 (Atzaba-Poria et al.

2004). On binary risk variables risk is coded as 1 if present, and 0 if absent,

whereas on continuous variables scores that are in the worst 25% of the sample

are deemed to be ‘risk’ and are coded as 1, with the remainder of the sample

coded as 0. Where risks can be defined using clinical cut offs such as when an

individual’s IQ falls below 85, (Raviv et al. 2010) or on conceptual means such

as being below the poverty line, or being in a single parent household (Flouri

2008), risk is defined using these methods. These total scores are then

summed for each individual to produce their overall cumulative risk score.

Although alternative methods to generate a cumulative score exist within the

literature (Atzaba-Poria et al. 2004), the method described above is by far the

most common (Raviv et al. 2010, Whipple et al. 2010, Lima, et al. 2010, Gerard

& Buelher 2004a, 2000b, Forehand et al. 1998), and in order to remain

consistent with previous studies it was used within the present study.

3.5 Evaluation of cumulative risk

Cumulative approaches are often overlooked within the literature, despite

evidence in their favour suggesting these models can be powerful predictor of

developmental outcomes (Appleyard et al. 2005). These models show that as

the number of risks increase within a child’s background the severity of

behaviour difficulties displayed also augments, and this is regardless of the

specific risks which make up the cumulative score (Raviv et al. 2010, Lima,

Caughy, Nettles & O’Campo 2010, McCrae & Barth 2008, Appleyard et al.

2005, Atzaba-Poria et al. 2004, Kerr, Black & Krishnakumar 2000, Forehand et

al. 1994). The present study aims to test these two assumptions that increasing

risk is associated with increased behaviour difficulties and whether number i.e.

104

a cumulative model is a more powerful predictor than type i.e. an independent

additive model.

There is a growing body of evidence arguing that studies incorporating multiple

risk factors using a cumulative metric are better able to account for negative

behavioural outcomes than studies acknowledging solely single factors

(Whipple, Evans, Barry, & Maxwell 2010, Evans, Kim, Ting, Tesser & Shanis

2007, Evans 2003, Forehand et al. 1998). The cumulative risk index can also

effectively predict other developmental outcomes including academic outcomes

(Gutman et al. 2003) IQ (Sameroff, Seifer, Baldwin & Baldwin 1993) and early

sexual experience (Price & Hyde 2009).

Cumulative approaches are also particularly beneficial as they acknowledge

that risk variables naturally co-vary, (Flouri 2008) e.g. in the case of behaviour

problems, risk factors such as having reading problems and poorer peer

relationships are often related (Messer et al. 2006). These models are able to

simultaneously utilise large numbers of risk factors, (Sameroff et al. 2003,

Evans 2003), even when small samples are used. Independent additive models

hoping to include a large number of variables with a small sample size would

struggle not to violate sample size assumptions of multiple regression analyses

(Gutman et al. 2003).

Cumulative risk scales have been praised for having high face validity and

being more likely to predict problem behaviours than any single risk factor, as

scales that incorporate more items by and large generate more reliable scores,

(Luthar 1993). As cumulative scores will be summed across a number of

variables they have the advantage of not only being a more stable measure, but

being better able to detect any effects, as measurement errors are reduced as

scores are added together (Flouri, Tzavidis & Kallis 2010). Nonetheless, it has

been argued that critical information can be sacrificed when more powerful

continuous variables are dichotomized in order to generate a cumulative risk

score (Raviv et al. 2010). Furthermore, it has been argued that questionable

statistical relationships can emerge when utilising this method (MacCallum,

Zhang, Preacher, & Rucker 2002). Cumulative risk scores that use the worst

105

25% as a cut off to signify risk are limited to the sample in which the data were

collected and may not reflect the entire population, limiting generalizability

(Raviv et al. 2010).

Cumulative models have also been criticised for not taking into account the

relative importance or power of each individual risk factor. This could be a

flawed assumption by the cumulative model that type of risk is not an important

issue as there is some evidence that different risks do pose different effects on

outcomes (Hooper, Burchinal, Erwick-Roberts, Zeisel & Neebe 1998, Ackerman

et al. 1999, Hall et al. 2010, Rousse & Fantuzzo 2009), and that cumulative risk

models are not always the most powerful predictive models (Hall et al. 2010).

This evidence at least to a degree disputes the claim that risks within

cumulative models are interchangeable and that number of risks is more

important than type. Kraemer et al. (2001) argue against the use of cumulative

risk models suggesting that these approaches, which ultimately involve scoring

and summing risk factors, do not help in understanding the aetiology of any

problem behaviour, or indeed are effective at explaining how the process of risk

are interrelated and operate to influence outcomes (Loukas, Prelow, Suizzo &

Allua 2008, Olson, Ceballo & Park 2002, Masten & Wright 1998).

Within a cumulative score all variables are assumed equal, which could result in

a problem when proximal and distal variables are both included within the risk

indices and assumed to have equivalent impact on outcomes (Hall et al. 2010).

In acknowledgement of this issue, within the present study cumulative risk

scores were generated as domain specific scores, to tease apart the more

proximal and distal influences on development. This approach is consistent with

Deater-Deckard et al. (1998), Atzaba-Poria et al. (2004), and Ribeaud & Eisner

(2010).

Flouri & Kallis (2007) maintain the importance of investigating risk accumulation

alongside risk specificity as evidence is fairly consistent in showing risk models

of accumulation provide a better fit to outcomes compared with specificity

models. As such both models will be acknowledged within the present study.

106

3.6 The functional form of cumulative risk models

Perhaps the most interesting, although often neglected aspect of cumulative

risk research is establishing the function of the relationship between contextual

risk and behaviour difficulties (Gerard & Buehler 1999, Jones et al. 2002). Tests

can be carried out to find this functional form (Flouri et al. 2010), assessing

whether the relationship is linear or non-linear, and if a non-linear pattern is

observed establishing at what point a ‘threshold’ may occur. Whether a linear or

non-linear model best describes the data is an important issue that remains a

disputed aspect within the literature (Gerard & Buelher 2004a, Appleyard et al.

2005).

If the relationship between cumulative risk and behaviour difficulties was linear

then the proportion of cumulative risk would be equal to that of the problem

behaviour; any increase in number of risks an individual is exposed to would be

mirrored in the extent of problem behaviour displayed (Appleyard et al. 2005).

Cumulative risk researchers have found evidence for such linear effects, that as

the number of risks increases within an individual’s background there is a

similar additive effect on problem behaviour, (Raviv et al. 2010, Flouri & Kallis

2007, Appleyard et al. 2005, Gerard & Buehler 1999, 2004a, 2004b, Dekovic

1999).

A viable alternative to a linear relationship between cumulative risk and

behaviour difficulties is a non-linear (or curvilinear) relationship (Rutter 1979,

Bierderman, et al. 1995, Forehand et al. 1998, Jones et al. 2002). Although

there are numerous patterns of non-linear relationships which could explain the

data, a quadratic type relationship, is often discussed in the context of these

variables. This relationship suggests that there is a disproportionate increase in

problem behaviour displayed after a certain number of risks or level of

cumulative risk is reached, i.e. it occurs only after a certain threshold is attained,

(Raviv et al. 2010, Morales & Guerra 2006). The combined effect of risk on a

certain outcome has been termed mass accumulation (Gerard & Buelher

2004a) signifying that the total effect of cumulative risk exerts influences on

107

problem behaviour greater than the sum of their individual parts (Lima et al.

2010, Flouri & Kallis 2007). This model therefore suggests that risk variables

potentiate one another to produce a more serious effect on behaviour.

Numerous studies have found evidence for this view, with Rutter (1979) being

one of the first to note this relationship. Rutter initially identified 6 family

correlates of childhood psychiatric disorders, then made comparisons on rates

of these problems between groups of participants holding differing numbers of

risks. Findings showed that children with one risk factor were no more likely that

those with zero to have any psychiatric disorder. However, when any two risks

were present there was a fourfold increase in rates of psychiatric disorders and

the presence of any four risks resulted in a tenfold increase. According to Rutter

risks potentiated one another so in combination their effects were more

powerful than the sum of the individual parts. This study provided evidence that

a threshold effect exists and that after 4 risks are reached there is a dramatic

increase in risk of adjustment problems.

Bierderman et al. (1995), utilising the same risk factors as Rutter showed that

the odds of having ADHD was 9.5 time higher for children with 2 risk factors

compared with those with none, but 34.6 times higher for those with 3 risk

factors compared with those with none. This has been supported by Forehand

et al. (1998) who showed that as the number of risks increased from 3 to 4

there was a dramatic increase in the amount of problem behaviour displayed.

These authors suggest that young people may have the resources to counter

three risks, however once four risks are reached, this could be the trigger point

when their coping mechanisms fail to overcome the risk exposure. Greenberg,

Speltz, Deklyen & Jones (2001) also found evidence of a non-linear relationship

between risk and problem behaviour (specifically in this case oppositional

defiant disorder). The study showed that when three or more risk factors were

present there was a large increase in problem behaviours. Finally Jones et al.

(2002) investigated the number of risks that children can tolerate before their

adjustment is significantly compromised. This study found that although the

cumulative risk score was not significantly related to externalising problems

there was nonetheless a dramatic increase in problem behaviour after 4 risks

108

were encountered. The above studies, despite utilising different individual

factors to make up the cumulative score and investigating distinct outcomes,

are remarkably consistent in arguing that dramatic increases in problem

behaviour occur when a trigger or threshold point is reached around 3 to 4

risks.

An alternative non-linear relationship between cumulative risk and behaviour

difficulties has also been discussed in the literature. This way of explaining the

data has been termed a saturation model; that after a certain amount of

cumulative risk there is not a sudden increase in problem behaviour but a

levelling off or plateau effect, (Evans 2003). In the first incidence a linear

relationship would describe the pattern between cumulative risk and problem

behaviour, however at a certain point a threshold is reached whereby any

additional risk has no further detrimental effect on problem behaviour (Gerard &

Buelher 2004a). Some evidence has been provided for this relationship as

Morales & Guerra (2006) found that initially there was a linear relationship

between cumulative risk and adjustment problems but once three risks in a

child’s background was reached this was a threshold, after which there was a

levelling off or plateau effect where additional risks (at least up to 5) had no

further negative effect on adjustment. Evans (2003) also showed that as the

number of risks increased in the sample of children so did mother report of

psychological distress in the child - this was true up to about 4 risk factors when

reports of psychological distress began to level off.

The present study aims to bring more lucidity to this discussion and assess

whether the relationship between cumulative risk and behaviour difficulties is

linear or conforms to a non-linear pattern.

109

3.7 Cumulative risk across different ecological levels

Addressing differences in cumulative risk exposure across ecological levels and

ultimately how problem behaviour is influenced differently by each cumulative

score at each ecological level has not received much attention within the

literature (Gerard & Buehler 2004a, 2004b). A few studies have measured a

cumulative score within each ecological level, for example Ribeaud & Eisner

(2010) generated a cumulative risk scores across different domains, finding that

the presence of more risks across the domains and a higher cumulative risk

index is associated with more aggressive behaviours. Deater-Deckard et al.

(1998) also found evidence that each of the cumulative scores at the 4

ecological levels, (individual, parenting, peer and socio-cultural) all made up of

differing number of risks nonetheless all contributed uniquely to the variance in

explaining behaviour problems. These authors suggest that risk factors at all

these levels are important in understanding the aetiology of this problem.

Atzaba-Poria et al. (2004) aimed to assess whether cumulative risk at different

ecological levels has different effects on behaviour problems. They found

evidence that cumulative risk at the different ecological levels was related

differently to the various outcome measures; specifically the microsystem level

(Bronfenbrenner 1979) was the best predictor for externalising problems. Other

researchers show that problems are particularly evident for youth who have

accumulated risks across multiple domains (Gerard & Buelher 2004b, Whipple

et al. 2010).

It is an interesting idea that cumulative risk varies in its impact at each

ecological level, although perhaps a more salient issue has been acknowledge

by Morales & Guerra (2006) who highlighted the importance of discovering

whether the same number of risks across multiple contexts has the same effect

as when observed in a single context. For example an individual may be

exposed to 2 risks within their school and 2 within their neighbourhood,

whereas another individual is not exposed to any risks at the school although 4

in the neighbourhood. An interesting investigation would be to assess how

these two compare, acknowledging the spread of risk across contexts.

110

Individuals may need a ‘safe haven’ or an ecological context where no risks are

present in order to protect against the risks experienced in other contexts (Call

& Mortimer 2001). The present study aimed to assess this idea of cumulative

risk within different ecological levels, it is anticipated a cumulative risk score

would be generated at the individual and school levels and compared with one

another to highlight the most important level.

3.8 Conclusions

This chapter presented and discussed the concept of cumulative risk identifying

it as a salient construct as a means of organising multiple factors into a

framework to account for behaviour difficulties. From the literature reviewed,

there appears a strong argument that cumulative risk is an important predictor

of developmental outcomes, with compelling evidence in its favour. This makes

it a particularly effective model for accounting for behaviour difficulties.

A cumulative risk model in isolation however, may not always provide the most

comprehensive account of how multiple factors influence behaviour difficulties,

and evidence has shown number is not always more important than type of risk.

Schoon (2006) therefore suggests that assessing risk most accurately needs to

be undertaken by utilising multiple models, (i.e. a cumulative model alongside

an independent additive model) that can account not only for the intensity but

also the number of factors that best predict the outcome.

Discussion has also taken place on a number of other issues which remain

pertinent with a cumulative framework; these issues include the nature of the

relationship between cumulative risk and behaviour difficulties and whether it is

linear or non-linear, and whether a certain number of risks in one ecological

domain are equivalent to the same number spread across multiple domains.

Of particular interest for the present study is whether cumulative risk is

predictive of behaviour difficulties displayed in children with SEND. Within the

111

present study a cumulative risk score will be generated by using the principles

highlighted above. In analysing the data using cumulative risk models, tests can

be carried out to assess whether increasing risk is related to increases in

behaviour difficulties displayed. An analysis can be undertaken to assess the

relational pattern between cumulative risk and behaviour difficulties, noting

whether a linear or non-linear model best describes the data. Finally, comparing

these models with independent additive models allows comment to be made on

whether it is the number of factors or the type of factors that are most powerful

in predicting behaviour difficulties.

Incorporating Bronfenbrenner’s ecological systems theory into the cumulative

model, shows how the different ecological levels of cumulative risk account for

behaviour difficulties, and whether the spread of factors across different

ecological levels is a significant issue in explaining behaviour difficulties.

However, it was only possible to generate an individual cumulative risk score

(see section 6.5.2). Therefore, whether the same number of risks concentrated

within a single domain, is equivalent to those spread across multiple domains,

in terms of their effect on behaviour difficulties could not be assessed, and

awaits further research.

112


• There is a need to acknowledge multiple risks in influencing behaviour

difficulties within the same study.

• There is a requirement to test the assumptions of a cumulative risk

approach and specifically for the present study that increasing risk

results in heightened behaviour difficulties.

• It is important to acknowledge the issue of the functional form of the

relationship between cumulative risk and behaviour difficulties and

whether this is linear or non-linear.

• A comparison between the cumulative risk model and the independent

additive model will test whether the number of risks is more important

than the type of kind of risk.

• Future research should investigate the cumulative risk and its relative

importance across different ecological levels.

• The research questions that have emerged from this chapter (labelled

research questions 2a, 2b and 2c), are presented below:








difficulties?

113

4. Protective Factors

4.1. Introduction to chapter

The aim of this chapter is to investigate the role of protective factors that

moderate risk experience to enhance positive adaption and therefore reduce

behaviour difficulties. The chapter begins with section 4.2 explaining the

importance of resilience as a construct and how studies in this field have

searched for salient protective factors that promote it. Section 4.3 and 4.4

describe the key issues of risk measurement and appropriate outcome when

investigating protective factors within the context of resilience. Section 4.5

discusses the variable and person focused approaches that have been taken to

account for protective factors. Section 4.6 discusses the salient protective

factors for behaviour difficulties. Section 4.7 adopts a more specific focus for the

present study discussing the school related protective factors. Finally section

4.8 provides a conclusion and 4.9 summary statements and related research

questions.

4.2 Protective factors and resilience

Protective factors have already been defined in chapter 1 as a “quality of a

person or context or their interaction that predicts better outcomes, particularly

in situations of risk or adversity”, (Wright & Masten 2005, page 19). The search

for these factors often falls within resilience research and so briefly discussing

the issues surrounding resilience and how it relates to protective factors is an

important issue.

There is no universally accepted definition of resilience (Naglieri & LeBuffe

2005), however, for the purpose of the present study resilience will be defined

as “a dynamic process encompassing positive adaption within the context of

114

significant adversity” (Luthar, Cicchetti & Becker 2000a, page 543). Resilience

cannot be directly measured; it is inferred on the basis of the how protective

factors interact with risk to influence positive adaption (Naglieri & LeBuffe 2005).

An acknowledgment also needs to be made of the context in which resilience is

occurring, as well as the population of interest, as some individuals will be

resilient only in regards to certain outcomes and to only to certain risks

experienced (Fergus & Zimmerman 2005, Luthar & Cicchetti 2000).

Acknowledging the interaction therefore of risk variables that (promote

vulnerability) and protective factors that (moderate this risk and promote

competence) is key to resilience research (Werner 2000). The broad aim of

research in this field is to establish how these interactions occur, and what

protective factors are present for children who maintain typical development

(Wright & Masten 2005).

The inception of resilience research began around 40 years ago, when

researchers noted that some children despite experiencing significant adversity

in their lives were able to achieve positive outcomes and not suffer with the

predicted psychopathology, (Werner 1993, Wright & Masten 2005, Schoon

2006). Resilience research therefore asks the question why these children who

are experiencing significant risk or adversity in their backgrounds achieve

positive developmental outcomes (Masten 2001). These children have not

avoided risk, but rather in some way responded to it in a positive way and

overcome the odds (Rutter 1999). The variables that promote resilience have

been termed protective factors, and may be composed of factors within the

individual, (i.e. IQ, Tiet et al. 2001) within the family (i.e. parental involvement,

Domina 2005), at school (i.e. participation in extracurricular activities, Mahoney

2000) or within the community (i.e. high SES, Masten 2006).

The present study will focus on searching for protective factors that promote

resilience as according to Masten (2001) this is a common phenomena and

“does not come from rare and specific qualities, but from everyday magic of

ordinary, normative human resources in the minds, brains, and bodies of

children, in their families and relationships, and in their communities” (page

235). Therefore factors within different ecological levels are worthy of

115

investigation as potential protective factors for behaviour difficulties. This type of

research is particularly appealing as it focuses upon healthy development and

the investigation of strengths rather than the deficits within individuals, (Fergus

& Zimmerman 2005, Evans & Pinnock 2007).

4.3 Measuring risk contexts for protective factor research

If any factor is to be labelled a protective factor it must be assessed in relation

to risk. How these constructs interact with one another and influence an

outcome in a positive direction equates to resilience. Measuring this risk,

however, has not been a simple exercise. Risk situations have been variously

defined; as being exposed to a single significant adverse situation, e.g. poverty;

aggregating a score of multiple risks, e.g. from a check list of negative life

events; or a cumulative risk score from various socio-demographic risks

(Masten 2001, Luthar & Cushing 2002). These differences in risk measurement

could pose a problem for research, as without consistency across studies

interpretation and generalisation of findings is problematic, (Olsson, Bond,

Burns, Vella-Brodrick & Sawyer 2003). Luthar et al. (2000a) however, argue

that with diversity in risk measurement that nonetheless highlights similar

protective factors, the stability of the resilience construct will only be enhanced.

In measuring risk an important issue is acknowledged; not all children assumed

to have similar experiences of adversity will necessarily have been subjected to

the same degree of risk, (Luthar & Zelazo 2003). Risk exposure has been

defined in various ways by researchers within this field, some have used

measures of ‘actual’ whereas others use ‘statistical’ measures (Luthar et al.

2000a). Measures of ‘actual’ risk are open to subjective interpretation (Luthar et

al. 2000a) and could be interpreted differently e.g. parental divorce; the majority

of children see this as a negative event, whereas for a child who was being

abused by that parent this could in fact be interpreted positively. The salient

issue remains that despite similar ‘risk’ experiences actual risk may be

somewhat different between individuals (Fergus & Zimmerman 2005). Statistical

116

risk on the other hand, is defined when significant correlations are found

between the measured variable and the outcome in question across the whole

population. Within this group however, an individual’s exposure to ‘actual’ risk

may vary considerably (Luthar et al. 2000a), with no guarantee any risk was in

fact experienced (Kaplan 2002). As risk factors are probabilities and not

certainties, (Schoon 2006), there is likely to be considerable heterogeneity in

risk experience for a population that has a particular risk factor in their

background.

Within the present study risk is measured using the various factors that

displayed a significant relationship to behaviour difficulties in research question

1. In research question 2 a comparison will be made between the cumulative

risk score (comprised of the number of contextual risk factors added together

e.g. FSM, poor relationships and experiences of bullying) and the additive

model where risks are acknowledged independently in their original form.

Whichever is the most parsimonious model will be used as a measure of risk

within the present study.

117

4.4 Measuring outcomes for protective factor research

If any factor is to be labelled a protective factor there must not only be a

significant interaction with risk context but also a better than expected or

positive outcome. For example, those with high levels of cumulative risk do

better than expected in terms of behaviour difficulties displayed in the presence

of the protective factor i.e. higher school SES. The focus therefore is not upon

those who display solely positive adaption but those who display it despite

experiencing significant risk (Luthar & Cicchetti 2000). Within the context of the

present study a multiple item measure of behaviour difficulties is used as the

measures of adjustment. This measurement is a rating scale and provides a

continuation from adjustment to maladjustment in the context of behaviour

difficulties.

Utilising a single outcome i.e. behaviour difficulties could be criticised, as some

researchers have argued that multiple domains of functioning are required in

order to label a protective factor as such (Luthar & Zelazo 2003). Nonetheless

others argue that resilience and the protective factors that support it are context

specific and will affect only a single domain of functioning (Vanderbilt-Adriance

& Shaw 2008). Therefore, terms such as behavioural resilience or academic

resilience should be used and protective factors only acknowledged within

these specific contexts (Vanderbilt-Adriance & Shaw 2008, Windle 2002). It is

important therefore that findings are only generalised across conceptually

similar domains of functioning and not across distinct ones as this could lead to

misleading findings, (Windle 2011, Luthar et al. 2000a, Rutter 1993).

118

4.5 Approaches to protective factor research

There are two broad approaches that can be taken when designing resilience

studies, in order to highlight the key protective factors and processes involved.

These have been termed person-focused and variable-focused approaches

(Masten 2001).

4.5.1 Variable focused approaches

Variable based approaches aim to establish statistical links between risks,

outcome and protective factors that may either compensate or protect the

individual from the negative outcomes associated with risk (Masten 2001). This

approach takes account of the whole sample of ‘at risk’ individuals and is

predominantly beneficial for investigating particular protective factors for

particular outcomes (Masten & Reed 2002). Within variable based approaches

there are various models which attempt to account for how factors function to

modify the trajectory from risk experience to negative outcome (Garmezy et al.

1984, Masten et al. 1988, Luthar et al. 2000a, Masten & Powell 2003, Schoon

2006). These include:, compensatory; protective; challenge; and inoculation

models. The two most dominant, however, which will be discussed further here

are the compensatory and protective models, (Zimmerman & Arunkumar 1994,

Fergus & Zimmerman 2005, Windle 2011). These models are not necessarily

mutually exclusive and “They may operate simultaneously or successively in the

adaptive repertoire of a resilient individual, depending on his or her stage of

development”, (Werner 2000, page 116).

Compensatory model The compensatory model assumes that protective factors have direct and

independent effects on the outcome variable, and they do not interact with risk,

(Fergus & Zimmerman 2005). These factors have been termed promotive

factors within the present study and are beneficial in both high and low

adversity situations (Windle 2011). By increasing the number or intensity of

these factors within a child’s background the effects of risk can be averted or

119

neutralised in high risk situations, and to the same degree improve outcomes in

general for those in low risk situations (Schoon 2006).

Compensatory models are often measured using independent direct effects in

multiple regressions (Fergus & Zimmerman 2005). These models take into

account the independent contributions of three types of predictors, pure risk i.e.

premature birth, (a risk when present, not when absent), pure protection i.e. a

musical talent (protective when present, no risk when absent) and bipolar

variables that denote risk at one pole and protection at the other (Zimmerman &

Arunkumar 1994). These variables function equally across high and low risk

situations, (Masten 2001, Masten 2002), therefore on the graph below the lines

(representing protective factor being present or absent) run parallel and have

equivalent slopes. For the purposes of clarity these types of effects are

discussed in chapter 2. See figure 4.1 for the compensatory model

Figure 4.1: An example of a compensatory model

Low Risk High Risk

Beha

viou

r Diff

icul

ties

The Compensatory Model

Attribute Present

Attribute Absent

120

Protective Model The protective model explains a process whereby a protective factor interacts

and moderates the influence of risk upon the outcome (Zimmerman &

Arunkumar 1994, Fergus & Zimmerman 2005). This model argues protective

factors are generally more effective in moderating the impact of risk in high-risk

situations compared with low-risk situations. The presence of the risk factor

could be said to potentiate the impact that the protective factor has upon the

outcome. These interaction effects are seen as the heart of protective factor

research (Luthar 2006, Luthar et al. 2000b) as certain factors are context

specific often operating more profoundly in high compared with low risk

contexts.

Within this broad model, there are various ways in which protective factors work

to influence outcomes, these have been termed protective-stabilizing, and

protective-reactive (Luthar et al. 2000a). Protective-stabilizing models show that

increasing levels of risk are linked with increases in problem outcomes although

only when the protective factor is absent. In its presence this relationship

between risk factor and outcome is neutralised (Fergus & Zimmerman 2005).

There is stability therefore in outcome despite increases in risk as protective

factors have overcome this relationship, (Windle 2011).

A protective-stabilising effect displays little variation in outcome between

different levels of the protective factor within low risk situations; however, large

differences in outcomes between different levels of the protective factor in high

risk situations. An example of this model is anti-social peers (risk) is related to

behaviour difficulties (outcome), however, in presence of quality parenting

(protective factor) the relationship between risk and outcome is neutralised.

See figure 4.2. In low risk situations the two lines are almost touching, with a

larger difference between them in high risk situations.

121

Figure 4.2: An example of a Protective-Stabilising model

Protective-reactive models on the other hand show that protective factors can

decrease the influence of risk factors on outcome although the relationship is

only weakened in presence of the protective factor and not completely

overcome (Fergus & Zimmerman 2005). Protective factors serve some

advantages although the effects are weaker under increased risk, (Windle

2011).

Protective-reactive effects show that although the protective factor when

present confers advantages for those at high risk, it has a greater influence on

outcomes in low risk situations. For example there will be a larger difference

between the two lines in low compared with high risk situations. An example of

this model could be anti-social peers (risk factor) being related to behaviour

difficulties (outcome), although this relationship is weaker for children who have

been part of an intervention programme (protective factor). See figure 4.3.

Low Risk High Risk

Beha

viou

r Diff

icul

ties

Protective-Stabilising

Attribute Present

Attribute Absent

122

Figure 4.3: An example of a Protective-Reactive model

Protective models are the most extensively studied model of resilience

(Zimmerman & Arunkumar 1994), and are measured using interaction terms in

regression analyses (Windle 2011). A over reliance however on interaction

effects can be problematic as they are not only difficult to detect (Rutter 2000,

Luthar et al. 2000b), but are often unstable associated with small effect sizes

and can conceal main effects findings, (Luthar 2006). Debate persists whether

protective factors should be defined uniquely by interaction effects of whether

they should incorporate main effects as well (Vanderbilt-Adriance & Shaw

2008). In terms of the present study a distinction has been made between the

two, protective factors that only display main effects have been termed

promotive factors, whereas the term protective will be utilised for interaction

effects.

Low Risk High Risk

Beha

viou

r Diff

icul

ties

Protective-Reactive

Attribute Present

Attribute Absent

123

4.5.2. Person focused approaches

Person focused approaches involve isolating a group of resilient individuals

(deemed as having high risk statuses but high competence levels) and

comparing them with so-called competent individuals (low risk, high

competence) and maladaptive individuals (high risk, low competence), (Masten

& Obradovic 2006, Jaffee et al. 2007). These analyses are conducted in order

to highlight the protective factors that lead away from poor outcomes (Windle

2011). These findings often show salient similarities between resilient and

competent individuals even with differences in risk exposure, and significant

differences between resilient and maladaptive children despite having similar

risk exposure (Masten & Obradovic 2006). This approach is often utilised when

assessing the percentage of individuals within the study termed resilient (Windle

2011), and is particularly good at looking at multiple outcomes across long time

periods (Masten & Reed 2002). Defining whether someone is at high or low risk,

and whether they are displaying high or low competence is often defined fairly

arbitrarily. With no conclusive means of conceptualising either of these

constructs as high or low within the present study, this approach will not be

investigated.

4.6 Protective factors and behaviour difficulties

The variation in outcomes experienced by individuals at risk was a particularly

salient finding in research investigating effects of adversity on developmental

(Rutter 2007). This led to the search for so called protective factors that could

potentially mitigate against the effects of adversity and lead to resilience.

From the vast array of studies on resilience conducted, a fairly consistent

pattern has emerged to the kinds of factors that aid positive adaption for

children exposed to risk (Masten 2006). This list of factors termed the ‘short list’

by Masten (2001), is a relatively small number of protective factors that have

been corroborated across distinct studies, using heterogeneous designs,

124

samples and outcomes (Wright & Masten 2005). These factors are however,

correlates of positive outcomes and although associated with resilience they

cannot necessarily be assumed to be causal agents (Masten & Coatsworth

1998). Table 4.1 shows the short list of protective factors that aid the

development of positive adaption for children at risk, (Masten 2006, Wright &

Masten 2005, Masten & Coatsworth 1998)

Table 4.1: Examples of protective factors that foster resilience for children at

risk

Individual characteristics:

• Problem solving abilities, good intellectual capabilities • Prosocial relationships • Positive view of self (high levels of self-esteem, self-

confidence and self-efficacy) • Positive attitude on life (faith, hopefulness, having in

sense of meaning in life) • Appealing qualities (talents, intelligence, athletic

abilities, attractiveness, sociable personality) Family characteristics:

• Socio-economic advantages • Parents involved in their child’s education • Supportive and consistent parenting • Close relationships with family members • Extended family support

School characteristics:

• Effective schools • After school clubs • Competent teachers • Prosocial relationships with teachers

Neighbourhood characteristics:

• Socio-economic advantage • Neighbourhood quality (safe, affordable housing,

access to facilities) • Involved in prosocial activities in the community • Support from positive adult role models in the

community (Adapted from Masten 2006, Wright & Masten 2005, Masten & Reed 2002 and Masten & Coatsworth 1998).

Although much of the conceptualisation and operationisation of resilience has

focused upon positive adaption in the general sense, suggesting that protective

factors are universal for resilience functioning (Luthar & Zelazo 2003) others

have argued that there may be specific factors relevant to particular outcomes

(Vanderbilt-Adriance & Shaw 2008).

125

The present study aimed to investigate the protective factors that foster

resilience with the specific outcome behaviour difficulties, and with an under

researched group, children with SEND. These factors may or may not be

consistent with the short list which argues these are general protective factors

important for multiple outcomes. A number of these factors within the short list

have been previously acknowledged (see chapter 2), as having main effects

(i.e. promotive effects) on behaviour difficulties, and were termed promotive

factors. What is yet to be discussed is the degree to which they can also offer

protective effects whereby they are particularly beneficial for children at high

levels of risk in reducing behaviour difficulties. Although the same factor can be

both promotive and protective (Wright & Masten 2005) depending on the

context, sample or outcome chosen for a study, in the review that follows only

their influence as protective factors will be acknowledged (i.e. interaction

effects).

Intelligence: Intelligence is probably the most cited protective factor in the

literature (Luthar 2006). Specifically research has, provided evidence that

higher IQ is related to the absence of any psychiatric disorder particularly for

children at high levels of risk (Tiet et al. 2001, Rutter 2000, Garmezy et al. 1984,

Masten et al. 1988, Tiet et al. 1998). However, others have argued that

intelligence levels may not able to remain protective in context of particularly

high risks (Vanderbilt-Adriance & Shaw 2008).

Relationships: Werner (2000) cites evidence that peer relationships marked by

closeness and competence, acceptance and likeability are associated with

resilient children. This has been further supported with evidence showing

positive peer relationships were able to moderate the negative effects of family

adversity on behaviour problems, i.e. children who had greater levels of

acceptance and friendships were not as affected by family adversity as those

children who were less accepted and had poorer friendships (Criss et al. 2002).

However, others have found no evidence of a protective effect of either peer

acceptance or peer attachment on moderating the influence of risk (Dekovic

1999).

126

Socio-economic status: SES can also be conceptualised as a protective factor.

In an extensive review of the literature Eriksson, Carter, Andershed &

Andershed (2011) highlighted high familial socio-economic status as a key

protective variable for children at risk. Other research however, has failed to

find a strong effect of SES as a protective factor (Tiet et al. 2001, Maggs,

Frome, Eccles & Barber 1997).

Parental Involvement: It has been suggested that effective parenting

(composing of parental support, monitoring and involvement) is a protective

factor for anti-social behaviour (Masten 2001), being especially effective in

promoting competence in children from higher risk backgrounds (Domina 2005).

When the risk measured is community violence, Kliewer, Ramirez, Obando,

Sandi & Karenkeris (2006) found that although exposure to this was associated

with maladaptive behaviour, parental monitoring could weaken this effect.

Others have suggested that when risk (in this case community violence

exposure) reached particularly high levels, the protective effects of parental

monitoring was attenuated, (Ceballo, Ramirez, Hearn & Maltese 2003, Sullivan,

Kung, Farrell 2004).

Extracurricular activities: In terms of school influences as protective effects a

particular focus has been upon participation in extracurricular activities. These

activities may be primarily beneficial for ‘at risk’ children, not only providing

them with a sense of belonging to a particular group, but also teaching them a

range of intellectual, social and physical skills which might be lacking, and could

be applied in other settings (Eccles, Barber, Stone & Hunt 2003). There is fairly

consistent evidence for this view that high risk children will benefit more from

extracurricular activities than low risk children (Luthar & Cicchetti 2000), and

particularly in terms of reduction in later involvement in criminal behaviour as

young adults (Mahoney 2000).

127

4.7 School level protective factors

The majority of the evidence specifically investigating protective factors for

behaviour difficulties appears to have focused upon the individual and family

levels. There has been a weaker focus on school level characteristics on

problem behaviour in comparison to other ecological domains. The effects of

how school level variables can potentially exacerbate risk experienced or

indeed protect against it has often been overlooked within the literature (Reinke

& Herman 2002).

The present study has opted for a particular focus on school level protective

effects. Children spend a substantial amount of time within schools and it is

likely that they have a significant impact upon a child’s behaviour. There is

strong evidence for the position that school level characteristics have an

influence upon an individual’s problem behaviour, in terms as acting as both

risk and promotive factors (Sellstrom & Bremberg 2006, Payne 2008, Bisset et

al. 2007, Evans-Whipp et al. 2004, Aveyard et al. 2004). A criticism of this

research investigating school influences on pupil outcomes is its focus upon

main effects findings, and not deciphering between the different effects school

characteristics have for those children at high compared with low risk. Those

studies that do acknowledge high risk children in investigating schools effects

on pupil outcomes such as behaviour difficulties, often only include this group in

isolation having no comparison with those children at low risk. These studies

have therefore looked for main effects within a group of at risk children rather

than an interaction effect between risk and protective factor.

It could be assumed that if there are school level promotive effects on

behaviour difficulties, then the search for school level protective effects could

also be justified. Whether certain school level influences are more beneficial for

children who are at high compared with those at low risk has been investigated

at least to a degree, with evidence from the ‘short list’ highlighting that effective

schools are salient protective factors. Other school based factors are present in

the shortlist; however, they often include variables such as attending school

128

extracurricular activities (Mahoney 2000) rather than school level variables such

as poverty levels within the school. The present study will therefore attempt to

fill this gap within the literature and investigate how school level variables such

as school level attendance, school level academic achievement, school size

etc. can be particularly important for those children experiencing high degrees

of risk, reducing their chance of displaying behaviour difficulties and therefore

acting as protective factors.

As schools “provide a unique opportunity for the state to deliver psychologically-

informed interventions that can span almost entire generations for the

betterment of individuals, communities and society” (Bulman 2012, page 18) it

is likely they will have salient influences on pupil behaviour. Identifying

protective factors within this domain will be important in reducing the chance of

behaviour difficulties for those most at risk. There is also evidence to suggest

that protective factors may have differing influences within specific contexts

(Fergus & Zimmerman 2005, Vanderbilt-Adriance & Shaw 2008, Sameroff et al.

2003, Masten, Best & Garmezy 1990), within certain populations (Tiet et al.

2001), for specific outcomes (Rutter 2000) as well as across distinct

developmental periods (Vanderbilt-Adriance & Shaw 2008). If this is true, that

protective factors have differing effects depending on context, outcome,

population, and development then a study investigating specifically school level

protective factors with children with SEND looking at the outcome of behaviour

difficulties across different developmental periods would add to the literature

base.

129

4.8 Conclusions

Within the context of this brief review two important issues are noted, firstly a

difference may need to be made with so called general protective factors that

are able to protect against risk from various sources and more specific

protective factors that only operate in the context of a single type of risk

(Erikson et al. 2011). Secondly, some protective factors may operate effectively

in the context of a specific outcome such as behaviour difficulties, but fail to

offer protection when the outcome is distinct such as academic outcomes

(Erikson et al. 2011). There is a clear need for studies to be specific about the

types of factors that provide protection to which outcome and under what

degree and type of risk. This will not only enhance the research within this area

but also allow more effective interventions to develop.

Any research conducted into protective factors and resilience needs to be

conducted within the context of a theoretical framework, (Luthar et al. 2000a).

Without such an approach, and with an over reliance on empirical evidence the

development of these constructs will be somewhat limited (Schoon 2006).

Researchers have often organised protective factors and processes relevant to

the study of resilience into three broad domains, i.e. individual, family and

community (Werner & Smith 1992), whereas others have adopted a more

ecological approach i.e. Bronfenbrenner’s ecological systems theory (Windle

2011) and accounted for the interactions between the different domains and the

developing child. Ultimately the ecological systems model attempts to

understand the individual person in the context where they live and therefore

accounts for the interactions they have with their physical, social and

environmental contexts. This model will therefore be utilised within the context

of the present study and will be further discussed within the methodology

section.

130


• An investigation of protective as well as promotive factors needs to be

undertaken within research, so interactive alongside main effects of

variables should be tested.

• A justification of how risk is conceptualised and what constitutes positive

adaption and whether it equates to a single or multiple outcome should

be acknowledged.

• There is a need for research to acknowledge the specific protective

factors within schools that can overcome risk and reduce the likelihood of


• These effects can be tested using a variable based design, looking for

interaction effects between risk and the protective factor in influencing


• Protective-stabilizing and protective-reactive effects should be

acknowledged within the broader protective model.

• The research questions that have emerged from this chapter (labelled

research questions 3a and 3b) are presented below:

a) Which school level predictor variables have a statistically significant




131

5. Methodology


The intention of this chapter is to provide a detailed description of the

methodology selected for this study, justifying it as the most appropriate to

answer the research questions presented in section 1.5.

The chapter is split into eight broad sections. Section 5.2 begins with describing

and justifying the adoption of Bronfenbrenner’s Ecological Systems Theory, for

the present study; as a means of organizing the predictor variables across

different ecological levels. Section 5.3 frames the present study within the

context of a wider study from which it emerged (the AfA evaluation study,

Humphrey et al. 2011). The aim here is not only to acknowledge where the

present study originated, but to justify it as an independent, original and distinct

piece of research.

The third part of the methodology is section 5.4, the design section. This

highlights the importance of epistemology, and provides justification for a

quantitative methodology, using a survey design. Within this section details of

the variables measured within the present study are also provided. Section 5.5

provides information on the participants of the study, including sampling,

attrition issues and how characteristics of the final sample compare to national

averages. Section 5.6 presents information around the materials used within the

study, including a detailed description of the measurement tool (i.e. the WOST)

and information about its psychometric properties.

The sixth part, section 5.7, includes the procedure of the present study, being

clearly described in order to allow for replication. Within section 5.8 the ethical

considerations for the study are presented, with a discussion of how they have

been met, this certifies the integrity of the study. The penultimate section, 5.9,

132

includes a discussion of multi-level modelling (MLM), which is the major

analytical technique applied within the present study. This section provides a

brief description of this complex analytical technique, offering definitions of the

terminology used, and explanations of how to interpret model outputs. Finally in

section 5.10, summary statements of the methodology are presented.

5.2 Theoretical framework

Adopting a theoretical model for the present piece of research is deemed

appropriate as a means of organising the numerous variables measured within

the study. These variables all have empirical justifications, (as evidenced within

the literature review), for being included as either: a risk factor; a promotive

factor; or a protective factor for behaviour difficulties. However, organising these

variables into a coherent whole within a theoretical framework provides a

stronger justification for including them within the study.

5.2.1 Ecological Systems Theory

The present study will be framed within the context of Bronfenbrenner’s (1979)

Ecological Systems Theory. This is a theoretically compelling approach to

development, and more specifically can be utilised to account for childhood and

adolescent behaviour difficulties. The theory acknowledges how multiple risk,

promotive and protective variables for behaviour difficulties in children with

SEND, can be organised across the different ecological levels within the study.

Ecological Systems Theory (Bronfenbrenner 1979), or Bioecological Theory as

it has become known, (Bronfenbrenner 2005), is a developmental approach

with a specific acknowledgment of the multiple contextual influences that affect

a child’s behaviour. This theory views “the child as developing within a complex

system of relationships affected by multiple levels of the surrounding

environment” (Berk 2009, page 26). The environmental context has been

133

described as similar to “a set of nested structures, each inside the next, like a

set of Russian dolls” (Bronfenbrenner 1979 page 3).

Bronfenbrenner's theory therefore argues that an individual is found nested

within various ecological systems. Some are more proximal, having a direct

influence e.g. the family or school (these have been termed ‘Microsystems’).

Other are more distal having an indirect effect on behaviour, e.g. cultural norms

and values (these have been termed ‘Macrosystems’). The relationships

between these various levels are interdependent, involving reciprocal

interactions rather than unidirectional influences (Bronfenbrenner 1979).

Therefore each individual system does not account for development in isolation,

but the interactions between and within systems have a profound influence

upon the individual and their behaviour and development. The theory not only

recognizes the importance of environmental variables, but also biological

influences within the child that interact with the environmental factors in order to

shape development.

Figure 5.1 shows Bronfenbrenner’s Ecological System Framework, and how the

various environmental contexts i.e. school, family, peer group (as well as

biological influences within the person), and the more distal influences interact

with one another to shape development. A brief description of the various terms

used within the model is also included.

134

Figure 5.1: Bronfenbrenner’s Ecological Systems Framework

The Microsystem comprises the individual child’s immediate social context such

as their family, peer group and school. These relationships have a direct

influence upon their behaviour, and as they are bidirectional, the child’s

behaviour also has an influence upon those within their Microsystem,

(Bronfenbrenner 2005).

The Exosystem refers to the wider setting in which the child is not directly

involved, but nonetheless has an influence upon their development

(Frederickson & Cline 2009). Parent’s social networks could have an indirect

influence on their child’s behaviour difficulties, as friends and relatives who give

advice on bringing up children may influence parenting behaviours which in turn

has an effect on a child’s behaviour (Beck 2009).

Individual

School

Family

Peer group

Macrosystem

Exosystem

Microsystems Mesosystems

135

The Macrosystem is the outermost level on the diagram and is comprised of

wider social systems such as: the cultural context; the political system; or

economic patterns. These systems encompass norms, values, beliefs and goals

that influence and shape development (Boyd & Bee 2009). For example,

cultural or political values on the importance of childcare will have some

influence on how a child is brought up within their immediate setting.

The Mesosystem refers primarily to the relationships among Microsystems

(Berger 2011), such as the relationships between school and family or parents.

If a parent is more involved with their child’s school, this relationship could

promote behavioural competence in the child as both school and parent

become consistent in their approach to managing behaviour. The Mesosystem

also refers to the dynamic relationships between Microsystems and the

Exosystem/Macrosystem and how these shape development.

The Chronosystem (not shown in the diagram) is an additional system within the

theory that broadly refers to the temporal dimension within the model.

Environmental influences are not seen as static but as always changing. New

events such as births within the family or starting school influence pre-existing

relationships between the child and their immediate environment (Berk 2009).

Furthermore, cultural norms and expectations can and will change over time.

5.2.2 Framing the present study within Ecological Systems Theory

The present study will be framed within Ecological Systems Theory. An

advantage of being guided by this theory is its ability to acknowledge multiple

factors involved in influencing behaviour difficulties. It is not one single factor

that is uniquely important but the combination and interaction of multiple factors

across different ecological levels of development that account for these

problems, (Ribeaud & Eisner 2010). The theory is able to recognise variables

beyond those embedded within the individual (and their biological

predispositions) and to include variables occurring in multiple settings such as

136

within the peer group, family, school, and neighbourhood, as well as those

cultural and wider family values affecting the child.

Adopting the Ecological Systems Theory (Bronfenbrenner 1979, 2005) is further

justified by the evidence that it is a prominent model of child development that

has been adopted by numerous researchers investigating different types of risk

factors that have an influence on a child’s behaviour difficulties (e.g. Atzaba-

Poria et al. 2004, Ribeaud & Eisner 2010, Trentacosta et al. 2008, Gerard &

Buelher 2004a). Indeed Atzaba-Poria et al. (2004) point out that if risk factors

are investigated within this framework a more comprehensive inspection of

possible variables can be ensured. Furthermore, if the multiple ecological levels

that potentially influence behaviour difficulties are ignored it will not be possible

to fully account for this behaviour.

The present study however, has opted to investigate specifically variables within

the various Microsystems (i.e. family, peer group and school level), as well as

variables that reside within the individual, such as gender or ethnicity. Variables

within the broader ecological levels (Exosystem and Macrosystem levels) go

beyond the scope of the present study and have not been included here.

The present study’s major focus however is upon the school microsystem level.

This emphasis on schools in preference to other Microsystems has been

justified on the premise that according to Bronfenbrenner (1994), there has

been considerably less research on school influences in comparison to family

influences on child development. Therefore a more school specific study will be

offering a significant contribution to knowledge in this area. With only a few

variables measured at the familial level (i.e. FSM) and peer group level (i.e.

positive relationships), in order to make for a more coherent analysis these are

included within the individual level. Therefore, two ecological levels are evident;

the individual level; and the school level.

Figure 5.2 shows part of Bronfenbrenner’s Ecological System Framework,

specifically focusing on the individual and school levels, and the factors

measured within the present study at these levels.

137

Figure 5.2. Predictor variables within the present study incorporated into Bronfenbrenner’s Ecological System Framework

As the diagram shows, the individual level is nested within the school level. The

nesting of individuals within schools assumes a hierarchical structure to the

data that requires an appropriate analytical strategy to analyse the data. Within

the present study the chosen method was multi-level modelling (see section

5.9). It is also noted here that modelling data within such a framework allows for

both the proximal (i.e. individual level), as well as more distal (i.e. school level)

influences on behaviour difficulties to be taken into account.

School size

School urbanicity

School % SA School % EAL

School % FSM

School level achievement

School absence

School exclusions

School % SAP & ST

SEND Category

FSM

SEND Support

Attendance

Academic Achievement

Positive relationships

Bullying

Bully Role Gender

Year Group

Season of Birth

Ethnicity

School

Individual

138

5.3 Context of the present study

This section has been included in order to provide an overview of the context

for the present study and acknowledge from where it emerged. The present

study utilised data collected from a larger evaluation project that was carried out

by a research team at the University of Manchester, on behalf of the

Department for Education DfE (formerly the Department for Children Schools

and Families, DCSF). The research evaluated Achievement for All (AfA), a

Special Educational Needs and Disabilities (SEND) initiative. As such, it is

deemed necessary to briefly7

describe this project, and how the evaluation was

conducted, before highlighting the differences between the present study and

the evaluation, in order to justify it as an original and independent piece of

research.

5.3.1 What is Achievement for All?

Achievement for All (AfA) was a SEND project, established by the Department

for Children, Schools and Families (DCSF), (now the Department for Education

DfE) with the aim of improving outcomes for these children. It was piloted in ten

local authorities (LAs) across England (sampled to be representative of LAs

nationally), and four hundred and fifty-four schools were selected from these

areas (including primary, secondary, and special schools, and also pupil referral

units). Certain year groups were specifically targeted within these schools,

years 1 and 5 within the primary sector, and years 7 and 10 within the

secondary sector 8

. Pupils, schools and LAs were involved for the two-year

duration of the project, from September 2009 until August 2011. The total

funding budget was £31,000,000.

7 The aim here is to provide a brief overview of the project. For a more detailed explanation of the AfA project and its national evaluation refer to DCSF (2009a) & (2009b) and Humphrey et al. (2011). 8 In year 1 children are between five - six years old, in year 5 children are between nine-ten years old, in year 7 children are between eleven-twelve years old and in year 10 children are between fourteen - fifteen years old.

139

The aim of the project was to support LAs and schools to present the best

inclusive practice, providing opportunities to aid children with SEND to fulfil their

potential. LAs and schools were encouraged to reflect, and build upon, their

pre-existing practices, and the strategies they used, for children and young

people with SEND. It was envisaged that LAs and schools would also be able to

provide the resources to further strengthen provision in areas that had the most

significant impact on the SEND population, and ultimately improve outcomes for

these children. Schools were supported in the implementation of the project by

their LA lead co-coordinator, a team of advisory teachers and by National

Strategies, (DCSF 2009a, 2009b, Humphrey et al. 2011). Specifically the

project included three broad strands (see figure 5.3).

Figure 5.3: Achievement for All diagram found on all guidance for schools and LAs (DCSF, 2009a page 4)

Strand 1: Assessment, tracking and intervention. This strand aimed to improve

the achievement and progress of children with SEND. Within this strand schools

were required to use Assessing Pupil Progress (APP) to track the progress of

their pupils, set targets and implement interventions to support their children.

140

Strand 2: Structured conversations with parents. This strand aimed to improve

the confidence in, and engagement of, parents of children with SEND in the

school. Within this strand teachers were trained in the techniques of holding a

structured conversation with parents, which involved the teacher using active

listening skills, identifying priorities, then planning and agreeing with the parent

on targets for how best to support the child with SEND. The structured

conversations allowed parents to open up and talk about their child so teacher

and parent could plan together the best means of support.

Stand 3: Provision for developing wider outcomes. The aim of this stand was to

improve outcomes in two of a possible five areas9

. Schools had considerable

flexibility and focused on the two areas that they felt would have the biggest

impact upon their pupils, tailoring interventions to meet their school and

individual needs.

5.3.2 Achievement for All evaluation study

The University of Manchester were commissioned to carry out the formal

evaluation of this project. The aim of the evaluation was to firstly, assess the

impact of AfA on outcomes for pupils with SEND, and secondly to identify the

processes and practices at LA, school and classroom levels that are most

effective in meeting and improving these outcomes.

The first research question was analysed quantitatively, assessing the impact of

AfA on outcomes for pupils with SEND, in terms of a) attainment in English and

mathematics, b) in relation to wider outcomes such as behaviour, bullying,

attendance, positive relationships, and wider participation, and c) in relation to

parental engagement. These variables were measured utilising teacher and

parent surveys, as well as collecting data on academic and attendance levels,

and other background and contextual variables at the individual and school

9 These included, attendance, behaviour, bullying, positive relationships and participation in extended services.

141

level over the course of the project. Parent and teacher surveys were completed

at three time points throughout the project, January 2010, January 2011 and

June 2011. Analyses compared the impact of AfA on these outcomes

longitudinally and also compared the impact of AfA on these outcomes with a

group of children in comparison schools10

.

The second research question attempted to identify the main processes and

practices that were particularly effective in improving the outcomes within

research question 1. This section of the research was primarily qualitative and

comprised of various research methods including: interviews with local authority

co-coordinators and National Strategy regional advisors; school case studies

(assessing the implementation, perceived impact and sustainability in specific

schools); and pupil case profiles (showing how AfA was working for specific

individuals). The findings of this study as an interim report (Humphrey et al.

2010) and final report (Humphrey et al. 2011) have been published by the

DfE11

.

5.3.3 Overview of methodology

The present study emerged from the AfA evaluation study, utilising some of the

data collected. It is very much a distinct piece of research however, with the

specific details of the methodology included within the relevant sections below.

Presented here, is a very brief overview of the study in order to aid

understanding, before verifying how it is different from the AfA evaluation study.

The present study aimed to investigate risk, promotive and protective factors for

behaviour difficulties in children with SEND. A longitudinal survey design was

utilised with approximately eighteen months between baseline and follow-up. A

measure of behaviour difficulties were collected via teacher reported surveys

that also measured predictor variables used within the study such as bullying 10 Comparison schools were recruited within the same LAs but not participating within the AfA project. 11 The author of the present study worked as a researcher on this project and was involved in analysing the data and writing the interim and final reports

142

and positive relationships. Background and contextual data were collected from

schools, and were also employed as further predictor variables. The sample

included all children with SEND in years 1, 5, 7 and 10. Analysis was carried out

using multi-level modelling investigating risk, promotive and protective factors

for behaviour difficulties. Behaviour difficulties at follow-up was the dependent

variable, with baseline behaviour difficulties as a control12

. All variables at the

school and individual level were included in the model and significant predictors

were reported.

5.3.4 How is the present study different from the AfA evaluation study?

Although the data generated from the evaluation study of the AfA project was

utilised by the present study, the aims, methods and analyses of these two

studies are separate and distinct, enabling it to be both an original and valid

piece of research. These differences are presented in table 5.1.

Table 5.1: An overview of how the present study differs from the AfA evaluation. THE PRESENT STUDY THE EVALUATION OF AFA Overview The present study is an individual

doctoral piece of work lasting 3 years

The AfA evaluation study was a 2 year pilot study, with a large research team

Aims The principle aim of the present study is to investigate the influence of risk, promotive and protective factors for behaviour difficulties of children with SEND

The principle aim of the AfA evaluation was assessing the project’s effectiveness as an intervention for children with SEND

12 A decision was taken to control statistically for initial levels of behaviour difficulties i.e. the baseline measure, with the outcome measure being follow up behaviour difficulties. This is consistent with Gutman & Midgley (2000) who state in accordance with Kenny (1979) that this approach “offers a stronger basis for inferring possible causal relationships than do simple zero-order longitudinal correlations” (page 231).

143

Aims The present study focuses exclusively upon a single outcome i.e. behaviour difficulties in children with SEND

The AfA evaluation study had a broader focus, looking at various types of outcomes including behaviour difficulties but also bullying, attendance, parental engagement, positive relationships, academic attainment and attendance of children with SEND

Aims The present study comments upon the theoretical as well as the practical issues, being able to contribute to both theory and practice

The AfA evaluation was concerned primarily with the practical implications rather than any theoretical concerns

Methods The present study is exclusively investigating variables that can be generalised across the entire SEND population and found within all schools (i.e. not specific to AfA) e.g. gender, FSM, ethnicity. Furthermore the study computed variables not included within the AfA evaluation such as season of birth and school exclusion rates

The AfA evaluation study includes variables specific to the project e.g. number of structured conversations carried out, and variables regarding target setting within the project

Methods The present study focuses specifically upon teacher-reported behaviour difficulties

The AfA evaluation acknowledged both parent and teacher-reported behaviour difficulties

Methods The present study focuses specifically on mainstream schools (both primary and secondary)

The focus of AfA evaluation was on all schools, including both mainstream and special schools, as well as pupil referral units

Analysis The present study utilises a more in-depth and complex analysis to investigate risk factors, promotive factors, protective factor and the cumulative effect of risk on behaviour difficulties in children and young people with SEND

The AfA evaluation carried out a fairly basic analysis of predictors of behaviour difficulties, which was limited in only investigating risk factors for children and young people with SEND

Analysis The present study analysed data entirely quantitatively

The AfA evaluation involved analysing data quantitatively as well as qualitatively

Target Audience

The principle audience the present study is aimed at is the academic community

The DfE and schools are the intended target audience

144

5.4 Design

Within this section the design is discussed, beginning with epistemology and

how this influences the broad methodology and individual methods applied to

the study. The specific variables measured within the study are also described

in this section.

5.4.1 Epistemology

Epistemology refers to the theory of knowledge. It asks questions about what it

signifies to know something or to research something and whether facts do

really exist (Jones & Forshaw 2012). There is much discussion around

epistemology and its various frameworks and conceptualisations across

numerous disciplines. Within the context of the present study, the definition set

out by Morgan (2007) will be utilised: this states that epistemological stances

are “distinctive belief systems that influence how research questions are asked

and answered and takes a narrower approach by concentrating on one’s

worldviews about issues within the philosophy of knowledge” (Morgan 2007,

page 52). Ascribing to any specific epistemology will therefore have

considerable influence on the soundness of any research carried out, and how

its findings are reported (Crotty 1998). Acknowledging this, the view point being

taken within the present study, is deemed appropriate and will ultimately guide

the methodological decisions and have influences on how results are

interpreted. There are four main epistemological paradigms found within

educational and psychological research. These are: post-positivist;

constructivist; transformative; and pragmatic (Mertens 2005). These are ways

in which the world is viewed and their overarching philosophical views shape

how research is undertaken. Whilst it is beyond the scope of this study to

discuss these paradigms, the author recommends Mertens (2005) for further

reading on the matter.

145

5.4.2 Post-positivist approach

In a great deal of published literature, (at least within psychology) a researcher’s

epistemological position for their study is not explicitly acknowledged. It is

mentioned here nonetheless, to allow the present study to be interpreted more

accurately.

The present study is being framed within a post-positivist epistemological

position. This position argues there is a definitive reality, but this truth can only

be known imperfectly, and inferred on the basis of probabilities (Robson 2011).

Ultimately this approach conforms to a more scientific method, in which: a

theory is presented; precise hypotheses are generated; and data is then

collected and analysed which will support, refute or redefine the original theory

(Robson 2011). Although any one individual study is not able to find the truth in

the situation, the post-positive approach emphasises that the more studies that

exist in support of a certain finding, the more likely it is that the truth has been

discovered. Objectivity is a key issue for this approach, with as many aspects of

the study being strictly controlled and the researcher having as little influence

on the participants as possible, whilst being aware that total objectivity is an

impossibility (Muijs 2011). This approach relies on the assumption of deductive

reasoning: assuming a phenomenon can be reduced to something

measureable. It can therefore be criticised for ignoring the complex detail in

everyday life, preferring a broader approach that can be generalised.

146

5.4.3 Quantitative methodology

In adopting a post-positive view, the present study utilises a quantitative

methodology. The advantages of such an approach will allow a substantial

number of participants to be included within the study, which will ultimately

result in a more reliable data set and improve the study’s generalizability.

Quantitative approaches are also easily replicable and objective conclusions

can be drawn from the research. As the focus of the present study is upon

behaviour difficulties, (i.e. behaviours that are observable and tangible, see

section 1.2.1 for a description), a quantitative methodology would be more

appropriate for this study. In adopting a deductive approach, pre-existing ideas,

predictions and hypotheses grounded within the literature can to be tested

(Robson 2011). Ultimately in adopting this approach the present study is

consistent with previous research within this area, adding to and extending the

literature base.

However, it is acknowledged that there are weaknesses to such approaches, as

they take limited information about the context where the phenomenon is being

investigated (Robson 2011) and are often criticised for producing inadequate

findings due to the closed type of questions asked (Jones & Forshaw 2012).

Reducing participants’ experiences down to their measured outcome on a

specified variable (Howitt & Cramer 2011) is also considered a disadvantage.

Nonetheless, despite these criticisms it is maintained that for the purposes of

the present study a quantitative approach is the most appropriate methodology,

whereas future research could look at this study from a more qualitative angle.

147

5.4.4 Survey design

Within the quantitative framework a survey design was selected as a means of

collecting the data. Survey designs are a particularly useful method within

research, and are perhaps the most prolific quantitative research method, within

social research (Robson 2011). A bespoke survey was designed for the

purposes of the study (see section 5.6) which measured the dependent variable

- behaviour difficulties, as well as other independent variables within the study,

i.e. positive relationships, bullying and role within bullying incidents. The other

variables within this study were collected by contacting schools, LAs, or

National Strategies, or taken from the DfE website, the National Pupil Database

or EduBase13

.

Specifically within the survey, three separate rating scales were used to

measure behaviour difficulties, positive relationships and bullying. These scales

are dimensional systems as they can acknowledge that behaviour lies across a

continuum and varies in degree. These measures describe specific behaviours

to which an informant who knows the child well (e.g. a teacher) rates each

descriptive statement in terms of the extent to which it relates to the child.

Rating scales are the most common way of assessing children with behaviour

difficulties, and this method has dramatically increased in popularity since the

mid-1980s (Merrell 2003).

There are a number of advantages in collecting data through a survey design,

not least because it is an easy, efficient and effective means of collecting large

amounts of information. This method was also selected as it could be flexibly

implemented, and options were provided to complete the surveys on paper or

on-line, whilst maintaining its standardised format. Using a survey approach

allows many different types of variables to be measured at the same time with

relative ease, and findings are therefore more generalizable, as information

comes directly from the ‘real’ world (Muijs 2011). The rating scales within the

13 The National Pupil Database contains details about pupil demographic characteristics and information such as academic progress. EduBase is a school level data base, containing information about all educational establishments in England and Wales.

148

survey are able to take into account a wide range of behaviours that may have

occurred over a long period of time and across various situations,

acknowledging lower frequency behaviours as well as asking about the

behaviour of the child in general. The standardised approach this method

adopts allows for objective and reliable data to be generated and comparisons

to be drawn between studies.

Rating scales however, could be considered subjective, as they comprise of the

respondents’ judgments of behaviour displayed, and measure perceptions

rather than actual behaviour, relying on the memory of the person completing

the survey (Robson 2011). This is not conceived to be a concern within the

present study however, as a reliable and valid measurement tool was created

for the purposes of the study that asked about specific behaviours displayed

(see section 5.6), and used a teacher report who have daily contact with their

children are know them well. A further criticism of this type of design is that

surveys using rating scales can also suffer from low response rates. However,

in the present study despite the high attrition rates predicted the sample sizes

were deemed to be appropriate for the analyses undertaken (see section 6.3).

Using a survey design was therefore deemed appropriate for the purposes of

the study.

149

5.4.5 Teacher, parent or self-report

The decision to utilise teacher report over parent or self-report report was made

on a number of grounds. First, the focus of the present study is upon behaviour

difficulties displayed within schools; teachers are therefore arguably better

placed to comment on behaviour in this context. Secondly, teachers are less

likely to be biased in reporting on negative behaviours than parents - who might

be more inclined to portray their own child in an overly positive light. There may

also be considerable variation between parents in what is deemed acceptable

or unacceptable behaviour, which may reflect ethnic differences (Guttmannova

et al. 2007). Teachers, however, are likely to adopt a more consistent approach

as they witness a wide range of behaviours displayed from numerous children

across the school. Finally, the present study incorporated a wide age range of

children with various disabilities, including children as young as 5 and those

with multiple and profound learning difficulties. It would be inappropriate to ask

these children to accurately comment on their behaviour difficulties displayed.

Therefore, in order to maintain consistency and accuracy across the study, a

decision was taken for teachers to provide the rating of their pupils’ behaviour.

150

5.4.6 Longitudinal study

The majority of studies in developmental psychology have utilised a cross-

sectional approach (Harris & Butterworth 2002). However, this design can be

considered somewhat limited, as it does not provide any information about the

temporal order of events. This is a particularly salient issue in risk research and

Kraemer et al. (2005) have argued that a risk factor cannot be called as such if

it cannot be established it came before the outcome.

Consequently the present study has opted for a longitudinal approach which

can be defined as “a research design in which data are collected on a sample

(of people, documents etc.) on at least two occasions”, (Bryman 2012, page

712). Specifically, the design can be noted as a Cohort Study, as the entire

cohort of SEND children (within the sampled schools) in year groups 1,5,7 & 10

were sampled, with the dependent variable being measured at multiple time

points, in January 2010 (baseline) and again in June 2011 (follow-up).

There are numerous advantages of a longitudinal design, including being able

to acknowledge the chronological order of events, and sometimes establish

cause and effect, (if cause is noted to precede the effect, and the effect is noted

as coming after the cause). These designs generate rich data that are able to

better account for the complexity of human behaviour (Cohen et al. 2011).

Longitudinal designs are considered particularly valuable in explaining

predictors of outcomes and the relationships between variables within studies

(Nazroo 2011). Disadvantages within longitudinal designs could include

significant outlays in terms of cost and time. There are also issues with attrition,

whereby collecting data on multiple occasions will often reduce the sample size

drastically (Cohen et al. 2011). Furthermore, these studies may be impacted by

other unmeasured factors such as any major life events which could have

occurred in between time points which affect the outcome. Nonetheless, despite

these limitations a longitudinal approach was utilised within the present study as

it is deemed a sophisticated design that can “separates real trends from chance

occurrence” (Cohen et al. 2011, page 219).

151

5.4.7 Variables in the study

The dependent variables within the present study are teacher reported

behaviour difficulties at baseline (January 2010) and follow-up (June 2011) from

the WOST (see materials section). Scores ranged from 0-3 with higher values

indicating greater behaviour difficulties. For the purposes of the analysis

behaviour difficulties at follow-up was added as the dependent variable and

behaviour difficulties at baseline was added as a control. Therefore any

influence on behaviour difficulties at follow-up that were above and beyond the

effect of baseline behaviour difficulties may be attributable to the various

predictor variables within the study.

The independent variables in the study comprised of twelve predictor variables

at the individual level, and nine predictor variables at the school level. Table 5.2

shows the individual level variables, a description of each, and how they were

measured. Table 5.3 displays the same information for the school level

variables.

152

Table 5.2: Pupil level predictor variables: descriptions and sources of data collection.

Predictor Variable

Description Source

Year group Year 1 or Year 5 (in primary schools), Year 7 or Year 10 (in secondary schools).

National Pupil Database

Season of birth Pupils’ month of birth was converted to a season; either Autumn (September through November), Winter (December through February), Spring (March through May), or Summer (June through August). 14

National Pupil Database

Gender Male or Female National Pupil

Database Eligible for Free School Meals (FSM)

Yes or No15 National Pupil Database

Ethnicity White British or Other 16 National Pupil Database

Academic achievement (English17

Pupils’ academic achievement in English was recorded as either: predicted GCSE grades; National Curriculum sub levels; or P-Levels. These were transformed to a points score (PS) that ranged from 1 (comparable to P-Level 1) to 65 (comparable to NC Level 10a/GCSE A*) (See Humphrey et al. 2011). Z scores were then created within each year group, and an individual’s relative position within their year was used as the predictor variable, to allow meaningful comparison across year groups.

)

Teachers recorded this information on a termly basis. It was sent by their school to the National Strategies, and forwarded to the researcher, who converted the data to a points score.

14 Season of Birth was chosen instead of month of birth in order to reduce the number of predictor variables within the model, but maintaining the variable as a potentially important predictor of behaviour difficulties.

15 FSM is often taken as a proxy for socio-economic status (Hobbs & Vignoles 2007), with children eligible for and claiming FSM being of lower social-economic status, compared with those not eligible for FSM.

16 This variable was limited to two categories, as breaking it down into further, diverse categories would result in some groups having insufficient numbers to conduct a meaningful analysis.

17 Data were available for English and maths, however, as they were highly correlated and showed evidence of multicollinearity, only the English score was included in the analysis.

153

Attendance Proportion of days in attendance at school displayed as a percentage from 0-100, with values nearer to 100 representing better attendance.

Local Authority

Positive relationships

WOST mean score on positive relationships sub scale, ranging from 0-3 with higher scores indicating better relationships with teachers and pupils.

Teacher survey at baseline, (see materials section)

Bullying WOST mean score on the bullying sub scale ranging from 0-3 with higher scores indicating a greater exposure to bullying.


Bully role Teachers recorded a pupil’s general role in bullying incidents as either Bully, Victim, Bully-Victim, Bystander, or Not Involved.


SEND category of need

The categories of SEND need were transformed into broad areas of need (DfES 2001a). 18Either Cognition & Learning (including SpLD, MLD, SLD, PMLD), Behaviour, emotional and social development (including BESD), Communication & Interaction (including SLCN, ASD), Sensory and/or Physical (including HI, VI, MSI, PD) or Other19

.

Category of need was calculated from the category recorded on the teacher survey at baseline, (see materials section).

SEND support School Action, School Action Plus, or Statement20

Level of support was recorded on the teacher survey at baseline, (see materials section).

18 See section 1.3 for a description of these terms.

19 The category Other was used if teachers felt none of the other categories were applicable although the child was still recorded as having SEND

20 See section 1.3 for a description of these terms

154

Table 5.3: School level predictor variables: descriptions and sources of data collection

Predictor Variable Description Source School urbanicity Whether the school is located in a

rural or urban area. EduBase

School size Number of pupils on roll at the school (this figure was divided by 100 to allow a more meaningful interpretation of the coefficients in the results section).

EduBase

School % eligible for Free School Meals (FSM)

Proportion of pupils receiving FSM, recorded as a percentage from 0-100.

Local Authority

School % English as an Additional Language (EAL)

Proportion of pupils speaking English as an additional language, recorded as a percentage from 0-100.

Local Authority

School % of pupils at School Action (SA)

Proportion of pupils receiving School Action level of support, recorded as a percentage from 0-100.

DfE Performance Tables

School % of pupils at School Action Plus (SAP) and Statements for SEND (ST)

Proportion of pupils receiving School Action Plus or Statement for SEND level of support, recorded as a percentage from 0-100.


School level overall achievement

In primary schools the proportion of pupils attaining at least NC Level 4 in English and maths. In secondary schools the proportion of children achieving at least 5 A*-C GCSE grades including English and maths. Recorded as a percentage from 0-100.


School level overall absence

The average rate of pupil absence from school, recorded as a percentage from 0-100 with higher rates indicating more instances of absence.


School exclusion rates

Pupils with one or more incidents of fixed period exclusions as a percentage of total school size, ranging from 0-100.

School Census

155

5.5 Participants

5.5.1 Local Authorities

Participants for the present study were sampled through the AfA evaluation

project (Humphrey et al. 2011) (see section 5.3 for an overview). At the outset

of the project the Department for Children Schools and Families (DCFS)

selected ten Local Authorities in England to be involved in the project. These

LAs varied in terms of geographical location, population density, and proportion

of ethnic minorities, and are deemed representative of England (DCSF 2009b).

5.5.2 Schools

Each LA sampled approximately forty-five schools to take part in the project.

They were selected to “reflect the range and proportion of schools in the area

and the current range of provision” (DCSF 2009b, page 9). These schools

comprised mainly primary and secondary schools with a handful of special

schools and pupil referral units (PRU). They were deemed representative of the

local authority, including schools in affluent as well as more deprived areas. As

part of the evaluation project it was also necessary to recruit a sample of

comparison schools, and these were recruited from the same local authorities

involved in the AfA project. Data from special schools and PRUs were excluded

at this stage, as the focus of the present study was solely upon mainstream

schools, i.e. primary and secondary schools. The total number of schools

represented within the present study numbered 305 (248 of which were primary

schools, and 57 if which were secondary schools).

5.5.3 Pupils

The participants within this study comprised children already identified by their

school as having a special educational need or disability and were accordingly

placed on the SEND register. Participants included all SEND children from

156

years 1 and 5 in primary schools and 7 and 10 in secondary schools. Teachers

completed a survey for each child within the sample and socio-demographic

data was also collected for each pupil from the school or National Pupil

Database (NPD). The total possible sample was 13223 although the final

sample for the present study only included 4288 of these pupils (2660 in

primary schools, and 1628 in secondary schools), due to the attrition rates.

5.5.4 Attrition

Background socio-demographic information was obtained from 13223 pupils

attending either primary or secondary schools within the AfA project. This

represented the maximum number of possible participants. From this number

4374 did not have a valid survey completed on their behalf at baseline and were

therefore excluded from the study 21. Of the remaining sample 474 did not

complete the behaviour difficulties sub scale in sufficient detail for a mean to be

generated and were also excluded22

. After baseline data were collected the

sample stood at 8375. The attrition rate was therefore 36.7% (i.e. 36.7% of

participants eligible for the study were excluded after wave one was completed).

From this number 3767 did not have a valid survey completed on their behalf at

follow-up and were therefore excluded. Of the remaining sample a further 137

did not have a completed behaviour difficulties sub-section, or it was completed

in insufficient detail to generate a mean score. The sample after follow-up stood

at 4471. The attrition rate was therefore 46.7% (i.e. 46.7% of participants

eligible for the study after baseline were excluded after follow-up was

completed).

Finally, a further 183 were excluded based on the data requirements and

assumptions (see section 6.3) leaving a final sample of 4288 (2660 in primary 21 The behaviour difficulties sub-domain on the survey was required to be competed for each pupil at baseline in order to be a valid case within the study, (as this represents the dependent variable). If pupils had any other missing data they were still included within the study. See section 6.2 for a more in depth analysis of the missing data. 22 If more than 2 items on the behaviour scale were missing a mean score could not be generated

157

and 1628 in secondary). This equates to 32.4% of the total possible sample and

therefore an attrition rate of 67.6%. Table 5.4 summarises the above

information and breaks the numbers down by primary and secondary schools.

Table 5.4: Participant attrition rates by total sample and school type.

As can be seen the attrition rates for secondary schools are considerably higher

than primary schools, with only 23.8% of pupils in secondary schools remaining

in the final sample from the optimal sample, compared with 41.7% of primary

school pupils.

5.5.5 Characteristics of final sample

The characteristics of the final sample are presented below. These descriptive

statistics have been split by data set (primary and secondary models) and

include proportions of the sample within each group for the categorical variables

and means and standard deviations for the continuous variables. Data is also

presented where possible with national averages, and effect sizes (using

Cohen’s d (Cohen 1992)). Effect sizes show the magnitude of the difference

between the present study sample and national averages, allowing meaningful

comparisons to be made between the data sets and giving an idea of the

representativeness of the sample. On the categorical variables the percentage

difference was calculated between the two samples allowing for a comparison

of the data sets. For both the primary and secondary school models a total of 9

predictor variables (8 at individual, 1 at school level) were categorical variables,

Optimal Sample

Valid cases after

baseline

Valid cases after follow-

up

Valid cases after data screening

Percentage remaining

from optimal sample

Total

13223 8375 4471 4288 32.4%

Primary

6377 4609 2788 2660 41.7%

Secondary 6846 3766 1683 1628 23.8%

158

and 12 predictor variables (4 at individual, 8 at school) were continuous

variables.

For the primary model: Table 5.5 shows the proportion within each group for the

categorical variables, compared with the national averages, with the percentage

difference also displayed. Table 5.6 shows the means and standard deviations

for each of the continuous variables, alongside national averages and effect

size differences.

For the secondary model: Table 5.7 shows the proportion within each group for

the categorical variables, with the percentage difference also displayed. Table

5.8 shows the means and standard deviations for each of the continuous

variables, alongside national averages and effect size differences.

The source of the national average data come from DfE (2010a) indicated by ♠;

from DfE (2010b) indicated by ♥; from DfE (2010c) indicated by ♦; from DfE

(2010d) indicated by ♣; from DfE (2011b) indicated by∇; or when this data was

not available indicated by ⊗.

159

Table 5.5: Sample proportion within each group of the categorical predictor variables at the individual and school level (primary schools)

Predictor Variable (n)

Groups of Predictor variable

Percentage of cases

per group

National Average

% difference between the

groups Individual Level Predictors

♠Year group (n = 2660)

Year 1 42.7% 43.1% 0.4 Year 5 57.3% 56.9% 0.4

⊗Season of birth (n = 2660)

Autumn 20.2% - - Winter 23.7% - - Spring 26.0% - - Summer 30.0% - -

♣Gender (n = 2660)

Male 65.6% 64% 1.6 Female 34.4% 36% 1.6

♣Ethnicity (n = 2659)

White British 76.6% 73.8% 2.8 Other 23.3% 26.2% 2.9

♣FSM (n = 2659)

Yes 34.9% 30.2% 4.7 No 65.1% 69.8% 4.7

⊗Bully role (n = 2505)

Victim

7.5%

-

-

Bully 6.1% - - Bully-Victim 11.9% - - Bystander 3.8% - - Not Involved 70.7% - - ♣SEND category23

(n = 2572)

Cognition and learning

58.7% 37% 21.7

Behaviour emotional & social development

15.3% 18.5% 3.2

Communication and interaction

20.0% 33.0% 13.0

Physical and/or sensory needs

2.1% 7.5% 5.4

Other 3.9% 4.0% 0.1 ♣SEND support (n = 2603)

School Action 62.4% 60% 2.4 School Action Plus 33.1% 33% 0.1 Statement 4.6% 7% 2.4

School Level Predictors ⊗Urbanicity (n =2660)

Rural 14.0% - - Urban 86.0% - -

23 Differences here reflect that national averages are computed only for those children at school action plus and statement, whereas in the present study those at school action were also included.

160

Table 5.6: Means and standard deviations of continuous predictor variables, with national averages and effect size comparisons (primary schools)

Predictor variable Mean score

Standard deviation

National Average

Effect Size 24

(Cohen’s d)

Individual Level Predictors ⊗Academic achievement (n = 2514)

025 1.00 - -

♦Attendance (n =2598)

93.35 5.91 94.7 0.22 (small)

⊗Positive relationships (n = 2647)

2.07 0.56 - -

⊗Bullying (n = 2628)

0.54 0.59 - -

School Level Predictors

♠School size (n = 2649)

3.32 2.16 2.33 0.46 (small)

♠School % FSM (n = 2609)

26.04 16.13 18.5 0.47 (small)

♠School % EAL (n = 2609)

21.00 28.15 26.2 0.18 (small)

♠School % SA (n = 2472)

14.55 7.10 19.9426

(All SEND)

-

♠School % SAP and ST (n = 2472)

10.46 5.79 -

♥School level overall achievement (n = 2374)

68.62 15.23 73 0.29 (small)

♦School level overall absence (n = 2466)

6.09 1.38 5.34 0.54 (medium)

∇School exclusion rates (n = 2660)

0.56 1.26 0.91 0.28 (small)

24 According to Cohen (1992) ≥ 0.20 equates to a small effect, ≥0.5 to a medium effect and ≥0.8 to a large effect 25 Due to the way in which this variable was measured (using z scores to allow cross year group comparison) it was not possible to compare the sample mean with national averages (see table 5.2) 26 Figures were only available for total SEND and not split by percentage at school action and percentage at school action plus & statement as in the present study

161

Table 5.7: Sample proportion within each group of the categorical predictor variables at the individual and school level (secondary schools).

Predictor Variable (n)

Groups of Predictor variable

Percentage of cases per

group

National Average

% difference between the

groups Individual Level Predictors

♠Year group (n = 1628)

Year 1 54.9% 51.7% 3.2 Year 5 45.1% 48.3% 3.2

⊗Season of birth (n = 1628)

Autumn 22.8% - - Winter 21.2% - - Spring 27.8% - - Summer 28.3% - -

♣Gender (n = 1628)

Male 57.7% 61% 3.3 Female 42.3% 39% 3.3

♣Ethnicity (n = 1628)

White British 84.4% 77.3% 7.1 Other 15.6% 22.7% 7.1

♣FSM (n = 1628)

Yes 29.5% 25.8% 3.7 No 70.5% 74.2% 3.7

⊗Bully role (n = 1476)

Victim 12.0% - - Bully 9.2% - - Bully-Victim 12.7% - - Bystander 3.6% - - Not Involved 62.5% - -

♣SEND category (n = 1603)

Cognition and Learning

60.1% 41.8% 18.3

Behaviour emotional and social development

23.3% 30.3% 7.0

Communication and interaction

8.8% 14.4% 5.6

Physical and/or sensory needs

4.2% 7.2% 3.0

Other 3.6% 6.2% 2.6 ♣SEND support (n = 1569)

School Action 54.2% 61% 6.8 School Action Plus

34.4% 30% 4.4

Statement 11.4% 9% 2.4 School Level Predictors

⊗Urbanicity (n =1628)

Rural 10.4% - - Urban 89.6% - -

162

Table 5.8: Means and standard deviations of continuous predictor variables, with national averages and effect size comparisons (secondary schools)

Predictor variable (n)

Mean score

Standard deviation

National Average

Effect Size (Cohen’s d)

Individual Level Predictors

⊗Academic achievement (n = 1465)

0 1.00 - -

♦Attendance (n =1617)

92.25 7.88 92.75 0.06 (small)

⊗Positive relationships (n = 1607)

2.08 0.59 - -

⊗Bullying (n = 1542)

0.50 0.66 - -

School Level Predictors

♠School size (n = 1628)

10.67 3.68 9.77 0.24 (small)

♠School % FSM (n = 1628)

20.70 16.13 15.5 0.32 (small)

♠School % EAL (n = 1628)

13.81 28.15 22.7 0.32 (small)

♠School % SA (n = 1609)

16.02 7.10 21.61

(All SEND)

-

♠School % SAP and ST (n = 1609)

11.11 5.79 -

♥School level overall achievement (n = 1609)

46.20 15.23 53.4 0.47 (small)

♦School level overall absence (n = 1609)

7.97 1.38 6.84 0.82 (large)

∇School exclusion rates (n = 1628)

4.27 3.18 4.70 0.14 (small)

163

As can be seen from the above tables, for both primary and secondary schools

the differences between the present study and national averages are broadly

comparable. The majority of effect size differences are considered small (Cohen

1992), with only one medium effect difference (primary school attendance) and

one large effect difference (secondary school attendance). In both cases,

schools in the sample were more likely to have higher overall absence rates

than national averages. The differences here could reflect that schools selected

by their LA to be part of AfA could have been biased in favour of schools with

poorer attendance rates. These schools would benefit to a greater extent from

the AfA initiative which ultimately aimed to improve attendance.

In terms of the categorical variables, the vast majority of variables showed less

than 5% difference between the present study and national average. There

were a couple of notable exceptions, however. The secondary data showed a

7.1% disparity in ethnicity where the present study had a higher number of

white British students compared with the national average. There was also a

6.8% difference in numbers of children at the school action level of SEND

support, where fewer children were recorded at this level in the present study

compared with the national average. The most notable difference across both

primary and secondary data sets was for the SEND category variable. This is

likely to reflect that data for national averages was computed for those at school

action plus and statement, whereas for the present study these groups were

computed together along with school action.

164

5.6 Materials

5.6.1 Overview

The present study utilised a teacher-reported survey to collect a number of the

variables under investigation. The survey included the Wider Outcomes Survey

for Teachers (WOST) (see below) as well as additional information including;

year group; date of birth (in which the variable season of birth was generated);

gender; SEND category, (in which the broad area of SEND was calculated);

SEND level of support and typical role in bully incidents (i.e. Victim, Bully, Bully-

victim, Bystander, or not involved) (See appendix 1).

The remaining factors under investigation included, (at the individual level)

FSM, ethnicity, attendance, academic achievement, and (at the school level)

urbanicity, school size, school % of FSM, school % of EAL, school % of SA,

school % of SAP and ST, school level overall achievement, school level overall

absence and school exclusion rates. These were collected either through the

school directly, the LA, via EduBase, the DCSF statistics Gateway or from the

DfE performance tables on-line (see section 5.4).

5.6.2 Wider Outcomes Survey for Teachers (WOST)

The Wider Outcome Survey for Teachers (WOST) (Humphrey et al. 2011) (see

appendix 1), was originally designed as a bespoke measure27

27 A bespoke measure was called for as no pre-existing measured covered the concepts required within a single format. Utilizing separate surveys for these concepts would be longer and waste teacher time.

for the national

evaluation of AfA. It is a teacher report questionnaire that assesses three broad

areas of pupils’ social outcomes. These include behaviour difficulties, positive

relationships, and bullying (specifically victimisation rather than perpetration).

The behaviour difficulties sub-section comprises 6 items (including: “The pupil

cheats and tells lies” and “The pupil gets into fights with other children”). The

165

positive relationships sub section incorporates 7 items (including: “The pupil is

kind towards others”, and “The pupil makes friends easily”). The bullying sub

scale also comprises 7 items (including: “The pupil is picked on by other

children” and ”The pupil is actively disliked by other children”). Specifically on

this sub domain a definition was given which stated “By bullying we mean

‘behaviour by an individual or group, usually repeated over time, that

intentionally hurts another individual or group either physically or emotionally’

(DCSF, 2008b, page 1)”. This definition was given to avoid confusion with a

potentially ambiguous and contentious topic. Items for all sub domains are rated

on a four point Likert scale using the categories, Never, Rarely, Sometime, and

Often, (for the behaviour difficulties and bullying scales), and Strongly Disagree,

Disagree, Agree, and Strongly Agree, (for the positive relationships scale).

5.6.3 Scoring

A score was composed for each of the three broad sub domains. For the

behaviour difficulties, and bullying sub sections a score was given for each item

as follows: Never = 0; Rarely = 1; Sometimes = 2; and Often = 3. For the

positive relationships scale a score was given for each item as follows: Strongly

Disagree = 0; Disagree = 1; Agree = 2; Strongly Agree = 3. A mean score was

then created for each individual, which ranged from 0-3. A decision was made

in accordance with other similar questionnaires (i.e. SDQ, Goodman 2001), that

if two or less items were missing, a mean score would still be generated but

three or more items missing a mean score would not be generated for that

subscale.

5.6.4 Psychometric properties of the WOST

Information regarding the psychometric properties of the WOST are required in

order to justify it as a reliable and valid tool. Presented below is evidence of the

criteria which the WOST was measured against, drawing upon some of the

criteria set out by Terwee et al. (2007). The information draws heavily upon the

same information, which has been presented in Humphrey et al. (2011) (the

166

evaluation of AfA page 117-122), the reader is referred there for more detailed

information and validation of this measurement tool.

Content Validity Content validity refers to “the extent to which the domain of interest is

comprehensively sampled by the items in the questionnaire” (Terwee et al.

2007, page 39). The WOST is considered to have good content validity and

was designed specifically to measure behaviour difficulties, positive

relationships and bullying. Items for each of the sub domains were generated

using a number of techniques. Firstly, there was discussion within the research

team and ‘brainstorming’ of potential topics within these sub domains (i.e. within

behaviour difficulties, lying, fighting, stealing etc.). Pre-existing measures were

then referred to which measured similar constructs (i.e., for the behaviour

difficulties sub-domain the Conduct problems scale within the Strengths and

Difficulties Questionnaire was referred to). The literature, which related

specifically to these sub domains, was also extensively researched and key

concepts and ideas for items were generated from within it. The team of

researchers then generated approximately ten questions per each sub domain,

which were designed to be comprehensible by the lowest possible reading age.

After an initial pilot of the survey and a basic psychometric analysis, the survey

was deemed valid and fit for purpose (Humphrey et al. 2011). The initial version

of the WOST contained 28 items (9 for behaviour difficulties, 10 for positive

relationships and 9 for bullying). These were later reduced once a confirmatory

factor analysis was carried out after data collection was complete. The final

version contains 20 items (6 behaviour difficulties, 7 positive relationships and 7

bullying) see appendix 1b.

Construct validity Construct validity refers to “the extent to which scores on a particular

questionnaire relate to other measures in a manner that is consistent with

theoretically derived hypotheses concerning the concepts that are being

measured” (Terwee et al. 2007, page 39). Humphrey et al. (2011) tested the

167

pre-specified hypotheses (based on previous literature) that the sub domains

for behaviour difficulties and bullying would be positively correlated and also

both negatively correlated with positive relationships. Testing this hypothesis

can be taken as a broad measure of construct validity. Findings showed that the

correlation between behaviour difficulties and bullying was 0.616, the

correlation between behaviour difficulties and positive relationships was -0.562,

and the correlation between bullying and positive relationships was -0.483. All

of these correlations were significant at the p <.001 level and therefore

supported the hypothesis and provided justification for the WOST’s construct

validity.

Internal consistency Internal consistency refers to “the extent to which items in a (sub) scale are inter

correlated, thus measuring the same construct” (Terwee et al. 2007, page 39).

This phenomenon can be assessed by Cronbach’s Alpha. Values of α > 0.7 are

deemed acceptable, and the three sub domains met this level (behaviour

difficulties 0.903, positive relationships 0.920 and bullying 0.903). A

confirmatory factor analysis (CFA) was also carried out upon the data to assess

whether the data observed fit the structure of the items within each sub domain.

Table 5.9 shows the 7 most commonly used CFA fit indices with their ideal fit

criteria and the measure recorded for the WOST.

Table 5.9: Confirmatory Factor Analysis fit indices for the WOST

CFI TLI GFI AGFI RMR RMSEA χ²/df

Ideal fit >0.9 >0.9 >0.95 >0.9 0 0 <2

WOST 0.922 0.909 0.884 0.850 0.035 0.086 49

The CFI and TLI met the ideal fit requirements, GFI; AGFI; RMR; and RMSEA

can be considered close fits, whereas the χ²/df shows a wide deviation between

actual and ideal fit (although this measure is affected by sample sizes). Debate

exists whether CFA fit indices should produce an exact (i.e. better than ideal fit)

168

or close fit (i.e. approaching the ideal fit figure) when confirming the internal

consistency of a measure (Fife-Schaw 2006). In the case of the present study,

and consistent with Humphrey et al. (2011), adopting the stance that a close fit

is sufficient was taken as this will reduce the likelihood of Type 2 errors.

Inter-rater reliability Inter-rater reliability refers to the degree to which two individuals are consistent

and produce similar results when rating an outcome. This agreement between

raters is considered an aspect of reproducibility (Terwee et al. 2007). Humphrey

et al. (2011) have reported on the inter-rater reliability between teachers and

parents in reporting behaviour difficulties, positive relationships and bullying.

They found significant (p < .001) Pearson Product Moment Correlations of

0.483 for behaviour difficulties, 0.344 for positive relationships and 0.368 for

bullying. The authors suggest that as these values are above 0.27 (the average

correlation between teachers and parents in a meta-analysis assessing similar

sub domains (Achenbach, McConaughty & Howell 1987)) they are deemed

acceptable in meeting the inter-rater reliability criteria.

Floor and ceiling effects Floor and ceiling effects occur when a large percentage of the sample obtain

scores at the highest (ceiling) or lowest (floor) possible score of a scale. It has

been argued that where more than 15% of the sample has the highest possible

or lowest possible score a floor or ceiling effect has occurred (Terwee et al.

2007). These can be problematic in a data set and limits the model’s sensitivity.

Floor and ceiling effects for the WOST were only observed for behaviour

difficulties (35% floor) and bullying (37% floor). It should however, also be

pointed out that the subscales in question measure behaviour difficulties and

bullying where it is expected a floor effect would exist, as most children do not

have any behaviour difficulties or are bullied. Furthermore in another well-

validated behaviour scale, the Conduct Problem scale on the SDQ (measured

by teachers) there is a floor effect of about 64.2% (www.sdqinfo.org). Despite

some observed floor effects the WOST was therefore retained.

http://www.sdqinfo.org/�

169

Interpretability Interpretability refers to “The degree to which one can assign qualitative

meaning to quantitative scores” (Terwee et al. 2007, page 39). The WOST

provides normative information in terms of means and standard deviations for

the various sub groups of primary SEND (see Humphrey et al. 2011). This

information will allow comparisons to be made between a certain score

obtained, and the normative score, or what would be expected. Therefore

allowing interpretability of the data.

170

5.7 Procedure

In October 2009 schools participating in AfA were sent information about the

evaluation study, to be carried out by the University of Manchester (see

appendix 2). In January 2010 participating schools within the 10 local

authorities were sent parental information letters about the study (see appendix

3). Schools were requested to forward this information to parents of children on

the SEND register in years 1 and 5 (within primary schools) and 7 and 10

(within secondary schools). The information sheets provided parents with the

precise details of the study, including how and when data was going to be

collected and what would be done with this data once collected. An opt-out form

was provided at this stage so parents not wishing for their child to be included in

the study could withdraw. The opt-out forms were either returned directly to the

researcher or via the child’s school.

Teachers also received consent forms, and information letters about the study

(see appendix 4). These information letters also included the specific details of

the study, and explained what information the study was asking teachers to

provide. The opt out form allowed teachers not wishing to complete the survey

to inform the researcher directly, Teachers also received information on

precisely how to complete the survey (see appendix 5), and were encouraged

to complete the surveys on-line wherever possible. If this was not going to be

achievable then copies of the survey were either emailed (so they could be

printed) or posted so teachers could complete these on paper and return them.

A final option to aid survey completion was offering teachers the option to

complete the surveys over the telephone. Each school was given a unique

password so teachers wishing to complete the survey on-line could do so. It

was highlighted that a teacher completing the survey on behalf of any child

should be someone who knows the child well and would be able to comment on

their behaviour, positive relationships and bullying. Technical support was made

available if teachers had concerns or difficulties with the survey completion.

Schools were reminded of pupils opting out so surveys for these children were

not completed.

171

The baseline survey was open for 2 months from the middle of January to the

middle of March 2010 (see appendix 1). Teachers provided the child’s UPN,

and the first initial of the child’s first and surname in order for surveys to be

matched to the later time point and with the additional information collected for

the study. The survey included 3 sub-domains, which included items for

behaviour difficulties (the dependent variable) positive relationships and

bullying. There were additional questions on the survey which asked about a

pupil’s typical role in bullying incidents, the child’s level of SEND support and

their category of need. The child’s gender, year, date of birth were also

required.

Academic data for individual pupils involved in the project was collected

alongside the survey data. Schools were asked to provide children’s academic

achievement in English, according to the National Curriculum sub levels. This

data was sent directly to the National Strategies and was subsequently

forwarded to the researcher. These levels were converted into a standardised

point score (see Humphrey et al. 2011). Individual attendance data were

collected from the Local Authority once it had been calculated for the baseline

year of 2009/2010.

In July and August 2010, the additional variables measured within the study

were collected; these included the majority of the school level28 and individual

level29

variables, collected via the National Pupil Database, EduBase or the

School Census. The remaining variables i.e. school level overall achievement,

school level overall absence and school exclusion rates were collected from the

DfE website when this data was made available for the baseline year in

question (2009/2010).

Approximately eighteen months later in June 2011, letters were sent out again

to schools. The information that teachers received asked them to complete the

same survey for an additional time for the same pupils as at the baseline

28 School urbanicity, school size, school % FSM, school % EAL, school % SA, and school % SAP & ST, 29 Year group, season of birth, gender, ethnicity, and FSM,

172

survey. The procedure for this was equivalent to the baseline survey as

presented above.

Once the data were collected a matching procedure took place, where a pupil’s

baseline survey, was matched with their follow-up survey, and with the

additional pupil and school level variables collected. Matching was done based

on UPN provided. Once the final data base was established UPNs were deleted

in order to maintain confidentiality. Data were then analysed within SPSS using

multi-level modelling.

173

5.8 Ethics

The data collected from the present study formed part of a larger study 30

(Humphrey et al. 2011) and ethical approval was given via the University of

Manchester Ethics Committee (Ref: 09226). The ethical considerations relevant

to the present study have however been acknowledged and addressed here

and are discussed below. Recognizing the importance of these ethical

considerations and attending to them is a salient issue that ultimately enhances

the study’s quality, strength, and legitimacy.

Fully informed consent In order to gain fully informed consent from the participants within the study,

information letters were sent out to parents and teachers of the selected pupils.

These letters clearly outlined the purpose of the study, including how and what

data would be collected (see appendix 3). These letters were available in 10 of

the most commonly spoken languages within the 10 Local Authorities sampled

in the study31

. This was implemented to ensure all parents were fully informed

about the nature of the study. Teachers were also sent an information letter

explaining the nature of the study and specifically what was expected of them,

as they would be completing the surveys on behalf of their pupils. Consent was

established through an opt-out form (see appendix 4), and anyone not

completing this form consented to being part of the study.

Right to withdraw An opt-out form (see appendix 3) was presented along with the information

letter to parents available in the same 10 languages. This opt-out consent form

asked parents to sign and return, either to the school directly, or to the

University of Manchester (via telephone, email or post) if they did not wish their

child to be part of the study (and have a teacher complete a survey on behalf of

their child). It was made clear to parents at this stage that they could withdraw

from the study at any specified time in the future if they so desired, and they 30 This study was the AfA evaluation study 31 The ten languages were Arabic, Bengali, Chinese Traditional, Chinese Simplified, English, French, Gujarati, Polish, Somali & Urdu,

174

would not suffer any penalty or have to give a specific reason. A total of 217

parents returned the withdrawal form indicating they did not give consent for

their child to be involved in the study. Teachers were also given the opportunity

to withdraw from the study, (by contacting the researcher directly) and in such

as case another key adult would complete the survey for that specified child.

No teacher did contact the researcher directly to withdraw from the study.

Anonymity The design of this study initially meant it was necessary for teachers and

parents to provide some potentially identifying information about their children,

such as date of birth, unique pupil number and first letter of forename and

surname. However, this information was only necessary to allow for data to be

matched correctly, i.e. across the different survey time points within the study

and with background information collected through the school. Once the data

set was completed, this information was destroyed so pupils and school were

completely anonymous within the study.

Confidentiality Parents and teachers were informed (through the information letters) that their

answers to the questions on the survey would be kept confidential and would

not be seen by anyone outside the University of Manchester. They were also

informed that the data from the questionnaire might be published although it

would not be possible to identify individual responses from this. The raw data

for the study was kept strictly confidential and stored in a password-protected

file.

Incentives for participation No incentive was given to the parents for agreeing to their child being involved

in the study, so opting out did not result in withdrawal of a potential reward. For

teachers however, although not directly rewarded for their participation, the

school in which they worked were receiving significant funding to be part of the

AfA project, which could have affected their decision about withdrawal.

Nonetheless there were a significant number of schools that despite receiving

175

the funding still did not return any of the surveys. This suggests that funding

received was not a significant factor on whether surveys were completed or not.

Protection of Participants The role of teachers within the study could have implications concerning

participant protection. This is because teachers were ‘obliged’ to complete a

survey for the evaluation project as specified by their LA32. The pressure placed

upon teachers could have had a potentially harmful effect. They may have felt

their individual practice was being assessed through their pupils’ performances,

or feel the stress that the effectiveness of the AfA project was dependent upon

their responses. These issues were addressed by the present study as far as

possible and their potential impact reduced. Firstly, teachers were not

pressurised into completing the survey by the researcher; secondly, they were

given the choice to opt out by contacting the University directly33

; thirdly, they

were informed that their individual responses were not being analysed in

isolation; and finally, teachers were made aware that any feedback given to the

school was at a school level and completely anonymised.

Best practice The study was conducted using the guidelines and principles of best practice

laid out in the Code of Ethics and Conduct (British Psychological Society, 2009)

and the Ethical Guidelines for Educational Research (British Educational

Research Association, 2011).

32 Schools received funding for the AfA project had a requirement placed upon them by the LA that they completed surveys for the evaluation. (These surveys were used within the present study). They were however, told of the opt-out option. 33 In such a case a colleague of the teacher in question would be asked to complete the survey in their place, which may have made opting out unlikely to occur.

176

5.9 Analytical strategy

5.9.1 Introduction

The purpose of section 5.9.2 is to justify why the analyses for this study were

carried out on the primary and secondary data sets separately. Section 5.9.3

provides a brief overview of the chosen analytical strategy, which is multi-level

modelling. This section will discuss what multi-level modelling (MLM) is and a

justification for its use within the present study. In a similar vein, section 5.9.4

gives definitions for key terminology within MLM, and in doing so shows how the

models can be interpreted.

5.9.2 Primary and secondary schools

The data for the present study has been split between primary and secondary

schools. The generation of two separate and distinct analyses, which run

throughout the present study is justified by the fact that there are important and

significant differences between these school types.

Compared with secondary schools, primary schools are generally smaller in

size, both in terms of physical space and numbers of pupils on roll (DfE 2010a).

Students in primary schools often have one class teacher for a whole academic

year, whereas in secondary schools students move to different classrooms in

the day and are taught by a number of specialist teachers. Primary school

teachers generally have fewer students in total compared with secondary

teachers. Therefore each pupil they have will received a greater proportion of

their time. As a result primary teachers may know their children better, and be

more accurate in rating their behaviour, positive relationships and bullying

experiences.

In primary schools, pupils are generally together in one class throughout the

day and may know one another better than students in secondary schools. In

secondary schools however, pupils have more chance to interact with a

177

multitude of other students, and there is likely to be a wider range of students

travelling in from further away to attend the school. Finally, and perhaps most

obviously, the majority of primary schools contain pupils from 5-11 years of age

whereas secondary schools usually include pupils 11-16 (DfE 2010a).

These differences are important to note, not least as the present study is an

educational-based piece of research with a salient focus on school effects.

These differences are likely to have distinct influences on the pupils who attend

them, particularly in terms of their behaviour difficulties displayed. Furthermore,

an advantage of splitting data between primary and secondary schools, allows

the effect of different developmental stages that influence behaviour difficulties

in distinct ways to be acknowledged. Lahey & Waldman (2003) have argued

there are multiple developmental pathways for behaviour difficulties, with at

least two groups emerging. These have been termed early onset i.e. within

childhood (accounted for by the primary model) and late onset i.e. within

adolescence (accounted for in the secondary model).The causes or risk factors

associated across these levels maybe significantly different. Splitting the data

between school type is therefore essential in order to maintain a coherent and

accurate analysis throughout the study.

5.9.3 What are multi-level models?

As multi-level modelling (MLM) is a complex statistical method, (Tabachnick &

Fidell (2007), some discussion is warranted. However, due to spatial limitations,

this will be kept to a minimum for further information on the underlying theory

and mathematical explanation the reader is referred to (Rasbash, Steele,

Browne, & Goldstein, 2009).

MLM is seen as an extension of multiple regression: an analytical tool which

aims to assess how much variance in the dependent variable can be accounted

for by the independent variables measured. The difference with MLM is that

predictor variables can be accounted for across a number of levels within the

analysis (Tabachnick & Fidell 2007). For example the influence of individual

178

level variables such as gender, age, ethnicity can be modelled alongside school

level variables such as urbanicity, size, levels of attainment etc. These levels

are organised hierarchically as the individual level is nested within the school

level. See figure 5.4 for an example of a two level hierarchical data structure.

Figure 5.4: An example of a two-level data set

As can be noted from figure 5.4 individual pupils are at level 1 and are found

nested within schools at level 2. Pupils are also clustered, meaning that those

who attend the same school are more similar than those attending different

schools and their scores on a certain variables are more likely to be correlated.

MLM is also able to account for more than two levels of analysis, so in the

example above schools can also be found nested within local authorities, which

in turn could be found nested within certain geographical regions.

The advantage of MLM is that it allows real world data (which is often

hierarchically organised) to be acknowledged and studied accurately (Field

2009). Indeed ignoring the fact that data are organised in this way and treating

them as if they were part of a single level leads to biased estimates (Heck,

Thomas & Tabata 2010), producing dubious results with statistical errors

(Tabachnick & Fidell 2007).

The reason for why this could occur is that traditional statistical analyses such

as multiple regression rely upon the assumption of independence of

observations, i.e. if participants are sampled from distinct contexts there is likely

to be independence in the data, and individual cases will not be related to each

179

other in any way. Data, which are hierarchical in nature however, violate this

assumption of independence and in the case of the present study, two

individuals who attend the same school are more likely to be similar in terms of

socio demographics or attitudes than two children attending different schools

(Field 2009, Heck et al. 2010). MLM is therefore able to account for this inherent

clustering or dependency within the data and factors in the importance of

contextual variables (i.e. school level variables) (Field 2009). MLM is then able

to account for individual scores by adjusting for school differences and also

predict school scores by accounting for the individuals within the school

(Tabachnick & Fidell 2007).

Using MLM allows predictor variables to be included at each level of the

analysis and also breaks down the variance in the dependent variable to the

various levels of interest within the study. Portioning variance in this way to the

different levels within the study is the first step of the analysis, before searching

for the significant predictors (Heck et al. 2010). The statistic that measures this

has been called the intraclass correlation coefficient (ICC) (see below) and can

also be seen as a measure of dependency within the data (Field 2009). It

represents the proportion of variance from total variance explained attributable

to the school and individual level. If this figure is large for the contextual level

(i.e. school level) it would suggest schools do have a large influence on

behaviour difficulties, and children who go to the same school are likely to

behave in a similar manner.

MLM is fast becoming the standard analytical procedure used within

psychological and educational research (Heck et al. 2010). It is only fairly

recently however, that this technique has become so popular, due to the

development of appropriate software programmes, enabling data to be

analysed hierarchically (Twisk 2006). MLM can be applied to numerous types of

research design (Heck et al. 2010) and as such it is deemed the most

appropriate for the present study adopting a two level (school and individual)

longitudinal design.

180

MLM is being used with the present study as it is considered the most

appropriate and indeed the most effective analytical strategy or tool in

‘uncovering’ the salient factors involved (Twisk 2006). It was selected as it fits

the design of the study and enables the research questions to be answered in

the greatest detail. Nonetheless the emphasis of the present study remains

upon the individual children within it and how their behaviour is affected by the

various risk, promotive and protective factors. The focus persists on the

practical application of the results and the variables of interest and how they

can be generalised to other children with SEND, rather than the technique

applied which leads to the uncovering of the results.

5.9.4 How to interpret the outcome: key concepts in multi-level models

The aim of this section is to provide a basic overview of the terminology used

within a multi-level model, in order to make the outputs presented later in the

results chapter readable and interpretable.

Empty and Full models. Multi-level models are usually built up in a number of

stages (Heck et al. 2010). Firstly the empty or null model is computed where no

predictor variables are included but the amount of unexplained variance that is

attributable to each level of the analysis in the study (i.e. individual and school

levels) is recorded. This is conducted to see what proportion of variance in

behaviour difficulties lies within the individual and between schools. This figure

is the ICC (see below) giving a percentage of variance to be explained at each

level of the analysis. Secondly, the full model is generated whereby all predictor

variables are added at their relevant levels i.e. gender at the individual level and

school size at the school level. At this stage the significant predictors are noted.

The empty and full models can be compared using the -2*log likelihood statistic

(see below), with the difference between the coefficients being a measure of

total amount of variance explained.

181

Intraclass correlation coefficient (ICC) refers to the coefficients that are used to

explain the amount of unexplained variance that is ascribed to each level of

analysis. In the context of the present study this equates to the individual or

school levels, before the predictors are added. This can be converted to a

percentage at each level, and therefore equates to the proportion of variance at

each level compared to total variance explained (Heck et al. 2010).

The coefficients β signify the quantity of variance in the dependent variable,

which can be ascribed to each of the predictor variables. The values indicate

the amount of change in the dependent variable as a result of one unit change

in the predictor variable. For example a coefficient value of -0.09 in positive

relationships signifies that a 1 point increase in positive relationships would

result in a 0.09 unit decrease in behaviour difficulties. As multi-level models only

give unstandardized coefficients these need to be interpreted as raw scores

and are not directly comparable across predictors. Consequently an

understanding on the scale to which the variables are measured is essential

this has been provided in table 5.2 and 5.3, (see section 5.4.7). The advantages

of using unstandardized coefficients are that they display actual scores on the

dependent variable. The resulting coefficient is therefore an estimate of the

changes in the dependent variable as a result of one unit change in the

predictor variable (Heck et al. 2010).

The standard error refers to the average amount of variance that coefficients

vary from the mean. If the standard error is small in comparison to the

coefficient then the predictor is likely to be significant within the model.

Specifically if the standard error is multiplied by 1.96 and results in a larger

value than the coefficient, the variable will be a non-significant predictor in the

model.

The p significance statistic assesses the statistical significance of the

coefficients within the model. It can be calculated by first dividing the coefficient

by its standard error to give the t statistic which is then assessed with the

appropriate degrees of freedom to generate the p value. The t statistic and

degrees of freedom are shown in the appendices (see appendix 6 for model

182

outputs). Within psychological research p values are regarded as being

significant when they are less than .05, being reported as p <.05 however, in

the case of the present study, actual values will be acknowledged in order to

highlight any marginal non-significant but potentially important trends i.e. values

that are not less than .05 but are less than .10. There is an argument that

deciding whether a value is significant or not based upon being greater or less

than .05 is ultimately arbitrary, although this is the consistently held view within

published psychological research. P values will be recorded up to 3 decimal

places, i.e. p < .001 or p = .345 consistent with SPSS and MLwiN outputs.

The -2*log likelihood statistic provides a value taken as the overall fit of the

model. It results from the comparison of observed data with the expected

values. If the statistic has a large value this can be interpreted as a poor fitting

model as there will be major deviations from expected values (Field 2009). In

such a case the model is likely to be comprised of considerable amounts of

unexplained variance and non-significant predictors. This statistic is used to

compare across models i.e. from the empty to full model and establish the best

fitting model.

The Chi Square statistic is a measure of whether a significant difference exists

between two models. In the case of the present study, this would be between

the empty and full models by comparing the -2*log likelihood statistic. It can

therefore note whether there has been any significant change when all

predictors are included within the model.

The intercept β0ij is displayed in the heading of each multi-level model and

reflects the overall average score for the average pupil (i) within the average

school (j) before the addition of any predictor variables. A precise score for any

individual could be calculated by using this figure and adding each of the

relevant coefficients noted in the table. Standard deviations are shown in the

brackets.

183

5.10 Summary statements

This section provides a brief overview of the eight main sections included within

the methodology section.

• Theoretical perspective: this section provided a justification for framing

the study with Bronfenbrenner’s Ecological Systems Theory.

• Context: this section provided a context for the present study, giving a

brief description of the AfA evaluation study from which the present study

emerged, but justifying the present study as an independent and original

piece of research.

• Design: this section justified the use of a longitudinal quantitative

methodology using a survey to collect the data. It also provided a detail

description of the variables within the study.

• Participants: this section provided an overview of the participants within

the study, how they were sampled and how they compare with national

averages. Information was also reported on attrition rates.

• Materials: this section provided a description of the WOST and justified

its use by highlighting its psychometric properties.

• Procedure: this section gave a detailed description of the study to allow

future replication.

• Ethical issues: this section showed that the study met the basic ethical

requirements.

• Analytical strategy: this section provided a description of multi-level

modelling including the terminology used in its outputs, and also

provided a justification for its use as an analytical strategy.

184

6. Results


In this chapter the findings of the current study are presented. For purposes of

clarity the chapter has been broken up into a number of sections. Section 6.2 is

an analysis of missing data within the study. Section 6.3 focuses on the

assumptions for multi-level modelling (the analytical strategy selected for this

study). Section 6.4 comprises research question 1 and its sub questions,

investigating risk and promotive factors for behaviour difficulties in children with

SEND. Section 6.5 comprises research question 2 and its sub questions,

investigating cumulative effects of risk on behaviour difficulties for children with

SEND. Finally Section 6.6 comprises research question 3 and its sub questions

investigating school level protective factors for behaviour difficulties in children

with SEND.

6.2 Missing data

Missing data refers to observations of any variable where no value is recorded.

It is a fairly common phenomenon within psychological and educational

research (Peugh & Endes, 2004) and a particularly salient issue within

longitudinal studies, where attrition rates can be high (Twisk & Vente, 2002).

There has been a call for studies to explicitly acknowledge missing data as

failing to account for it could lead to biases in reported results (Wilkinson 1999).

Nonetheless, many studies still do not provide any information on this issue

(Schlomer, Bauman & Card 2010).

185

A detailed missing data analysis was carried out in the present study. Due to

spatial limitations these analyses are not included within the main body of this

report; instead they appear in appendix 8.

In summary, this analysis was conducted in three phases. Phase 1 assessed

whether those participants with a valid survey34 at baseline differed from those

who did not have a valid survey at this time point. Results revealed that the

difference between the optimum sample and the sample with a baseline survey

completed can be considered comparable. Phase 2 assessed whether those

participants with a valid survey at baseline and follow-up differed from those

who had a valid survey at baseline but not at follow-up. The results revealed

that the difference between the sample with solely a valid survey at baseline

and those with a valid survey at baseline and follow-up were comparable, with

only one notable exception; that pupils attending larger schools were less likely

to have a survey completed at follow-up. Phase 3 assessed whether any

meaningful patterns in missing data existed between the final sample35

across

all the predictor variables. Results revealed that data were not missing

completely at random (MCAR), but this was likely to reflect a whole school not

completing and returning the school level data rather than being related to a

specific pupil. In conclusion, missing data is unlikely to have had a detrimental

effect upon the results found within the present study.

34 Valid surveys refer to any survey that was submitted which generated sufficient information for a behaviour difficulties mean score to be generated i.e. not more than 2 items missing. 35 Due to the longitudinal nature of this study, the final sample consisted of only those individuals with a valid survey at baseline and follow-up (with a behaviour difficulties mean at both these time points).

186

6.3 Data assumptions for Multi-Level Models

Introduction There are a number of data requirements and assumptions that are integral to

analyses of data using multi-level modelling, the majority of which are

synonymous with those of multiple regression analyses (Tabachnick & Fidell

2007). It is important to check these assumptions and interpret results in the

light of these findings, noting the possible implications they have upon the

study. Violations of these assumptions however, are fairly widespread in

psychological research, but failing to meet them weakens rather than

invalidates results generated from the data (Field 2009).

The requirements for the present study are presented below with a brief

description of the concept, followed by whether they were met in the primary

and secondary data set.

Multicollinearity Multicollinearity occurs when predictor variables are highly correlated with one

another. The acceptable limit for bivariate correlations is considered to be .7

(Tabachnick & Fidell 2007). Anything below this level can be included within a

model, but with correlations above this level one of the predictors should be

removed. Evidence of multicollinearity can also be examined more precisely

using the Variance Inflation Factor (VIF). Values greater than 10 suggest there

is multicollinearity between variables (Pallant 2010).

Primary and Secondary Models: No variable for either the primary or secondary

model exceeded the correlation of .7. The VIF statistic also showed the values

were within an acceptable range (1.020-3.511, for primary and 1.032-9.292 for

secondary) and did not exceed 10, which would display evidence of

multicollinearity. Therefore this assumption was met.

187

Normality Normality is evident in the data when the difference between the ‘observed’ and

‘expected’ values (i.e. the residuals) is normally distributed with a mean of 0.

Looking at a histogram (displaying whether the standardized residuals are

normally distributed), and the Normal P-P plot of Regression Standardised

Residuals (showing whether the values fit along a straight line) can assess the

assumption of normality.

Primary and Secondary Models: See appendix 7. Although there is some

deviation from normality and a slight positive skew for both data sets, the

histograms show evidence that residuals are approximately normally

distributed. The overall trend in the graphs shows the points fit along a straight

line, with marginal deviation from normality. Furthermore, the standardised

residual mean is 0.02 (standard deviation of 0.98) for the primary model and

0.02 (standard deviation of 0.96) for the secondary model. Therefore it can be

concluded this assumption was met.

Linearity The relationship between variables being assessed should be linear in nature,

as this is the principle underlying regression analyses. This assumption

therefore assesses whether the residuals conform to a linear relationship

(Pallant 2010). Observing the scatterplot measuring the standardised residual

versus the standardised predicted values can assess this assumption. If the

overall shape of the scatterplot is rectangular rather than curved in any degree

this displays evidence of linearity, (Tabachnick & Fiddell 2007). When linearity

cannot be inferred the overall model is weakened but not invalidated (ibid).

Primary and Secondary Models: See appendix 7. The overall pattern of the

scatterplots are rectangular in shape although a very slight deviation from this

pattern in the bottom left corner of both plots, suggesting a slight deviation from

a linear pattern.

188

Homoscedasticity Homoscedasticity assesses whether residuals have approximately equal

variances across all predicted dependent variable scores (Tabachnick & Fiddell

2007). This assumption is assessed by observing whether the points within the

scatterplot (measuring the standardised residual versus the standardised

predicted value), are approximately equal in width at all values of the expected

dependent variable scores.

Primary and Secondary Models: See appendix 7. The overall pattern of the

scatterplot shows that points on the graph are approximately equal across all

expected dependent variable scores for both models. There are slight

deviations in the bottom left corners of the graphs, although this is not

considered serious because the thinnest point on the graph is not three times

smaller than the widest point, which would signify a more serious violation of

this assumption (Tabachnick & Fidell 2007).

Independence of residuals This assumption is tested using the Durbin-Watson statistic and refers to

whether the relationship between the residuals of any two observations are

independent from one another and do not correlate (Field 2009). The Durbin-

Watson statistic varies between 0 and 4, with a value between 1 and 3 being

deemed adequate and displaying evidence of independence of residuals (ibid).

Primary and Secondary Models: The Durbin-Watson statistic was 1.856 for the

primary model and 2.028 for the secondary model therefore the data for both

models can be assumed to have independence of residuals.

Multivariate outliers An outlier is so called when the case differs considerably from the pattern or

trend within the data. Outliers can be identified in the data set by noting the

cases that the overall model predicts poorly. As 99.9% of cases should fall

between 3.29+/- (on a measure of standardized residuals) cases outside of this

range are considered outliers, (Field 2009). The Mahalanobis distance also

189

identifies multivariate outliers and specifically measures the distance between

scores and mean values of predictor variables. Cases that score above the

critical value (χ² (df 40) = 73.402, p<.001) (which is based on the number of

predictors in the model) are removed (Tabachnick & Fiddell 2007: Table C4).

Primary and Secondary Models: 12 cases in the primary and 3 cases in the

secondary model had standardised residual scores above the critical level and

were therefore removed. 114 cases in primary and 52 cases in secondary had

a score above the critical value of the Mahalanobis distance and were also

removed from the model.

Influential cases Influential cases refer to those that have an excessive influence on a model’s

predictions. Cook’s distance measures the influence of each case on the model;

values above 1 are cause for concern (Field 2009).

Primary and Secondary Models: Cook’s distance showed all cases within both

models had scores less than 1, suggesting no cases had an excessive

influence on the model’s predictions.

Random intercepts This assumption is additional to that of multiple regression which states that the

random intercepts must be normally distributed around the overall model (Field

2009). The random intercepts are referring to the schools in the study.

Primary and Secondary Model: The overall pattern of the histogram showed the

intercepts are broadly normally distributed for both the primary and secondary

models, therefore meeting this requirement.

Sample size The final sample size contained 4288 participants: 2660 participants nested in

248 primary schools, and 1628 participants nested in 57 secondary schools. An

analysis was undertaken to assess whether the number of participants and

190

schools were sufficient to detect a small effect at an Alpha level of 0.05. Sample

sizes within multi-level model studies are conducted by calculating a standard

sample size and then adding a correction criteria (Twisk 2006).

The standard sample size was calculated according to the figures reported in

Cohen (1992). An alpha level of 0.05 was selected for a small effect size f2 =

0.02. With a total of 40 predictors a sample of 1404 participants would be

required.

The correction equation was calculated according to Twisk (2006).

𝑀 =𝑁

1 + (𝑛 − 1) (1 − 𝑝)

M = Number of schools required

N = Number of participants (from standard sample size)

p= Intraclass correlation coefficient36

n = Average number of participants per school

Primary model

𝑀 =𝑁

1 + (10.7 − 1) (1 − 0.109)

14049.643

= 145.598

36 The ICC was calculated on the data set after baseline survey completion, but before the follow-up data collection. Behaviour difficulties mean score at baseline was taken as the dependent variable for these analyses, which resulted in an ICC in primary of 0.109, and in secondary 0.111. These figures reflect others found in the literature, i.e. Sellstrom & Bremberg (2006) investigating school effects found the ICC for problem behaviour was around 0.08.

191

At least 1404 participants and 146 schools are needed within the primary model

to obtain a small effect size at .05 level. The final primary model contained 2660

participants and 248 schools and therefore the sample size was sufficient for

the analyses

Secondary model

𝑀 =𝑁

1 + (28.6 − 1) (1 − 0.111)

1404

25.536= 54.981

At least 1404 participants and 55 schools are needed within the secondary

model to obtain a small effect size at .05 level. The secondary model contained

1628 participants and 57 schools and therefore the sample size was sufficient

for the analyses.

Floor and ceiling effects Floor and ceiling effects occur when a large percentage of the sample have

scores at the highest (ceiling) or lowest (floor) possible score of a scale. It has

been argued that where more than 15% of the sample has the highest possible

or lowest possible score, a floor or ceiling effect has occurred (Terwee et al.

2007). These can be problematic in a data set because they limit the model’s

sensitivity.

Primary and Secondary Models: In the primary model 36% of the sample

achieved a score of 0, whereas in the secondary model this was 39%. Although

these are clear floor effects within the data set, it should be pointed out that the

scale in question measures behaviour difficulties where it is expected a floor

effect should exist, as most children do not suffer with these problems.

Furthermore, floor effects can be justified as in another well validated behaviour

scale; the conduct problem scale on the Strengths and Difficulties

Questionnaire (SDQ), a floor effect of 64.2% exists on the teacher reported

measure of these problems (www.sdqinfo.org). One potential means of

http://www.sdqinfo.org/�

192

overcoming floor effects is to transform the scores. However, it has been

argued transforming scores can lead to limitations such as reduced

interpretability (Field 2009, Tabachnick & Fidell 2007). It was therefore decided

to keep the sample size to a maximum, including all scores, thereby retaining

adequate power levels.

Summary of assumptions and requirements As can be seen from table 6.1 the majority of the assumptions and

requirements for multi-level modelling were met. Two notable exceptions were

the lack of homoscedasticity and presence of a floor effect.

Table 6.1. Summary of data assumptions and requirements

Data Assumption Criteria Met?

Primary Model Secondary Model

No Multicollinearity Normality Linearity Homoscedasticity Х Х

Independence of residuals No Multivariate outliers No Influential cases Random Intercepts Appropriate Sample size No Floor and ceiling effects Х Х

193

6.4 Risk and promotive factors

6.4.1 Introduction to the section

The aim of this section is to answer the 3 sub-research questions within

research questions 1. These are:







The section begins with a presentation of the relevant descriptive statistics,

before moving on to the inferential analyses to answer each of these questions.

6.4.2 Descriptive Statistics

Table 6.2: Mean and standard deviations for the behaviour difficulties score at baseline and follow-up within the primary and secondary data sets.

Primary school model N =2660

Secondary school model N =1628

Mean score Standard deviation

Mean score Standard deviation

Follow-up score

0.57 0.67 0.62 0.75

Baseline score

0.67 0.71 0.56 0.76

Table 6.2 shows that the primary model is considerably larger than the

secondary model, comprising 2660 compared with 1628 participants. In the

primary model the behaviour difficulties mean score dropped from 0.67 at

baseline to 0.57 at follow-up, whereas in the secondary model behaviour

194

difficulties mean score increased from 0.56 at baseline to 0.62 at follow-up. The

large standard deviations reflect a fairly large spread in the data. The minimum

possible score for each model, and at both time points was 0, with the

maximum possible score 3.

Table 6.3. Bivariate correlations between the behaviour difficulties mean score (at baseline and follow-up) and all predictor variables within the primary and secondary school models 37

.

Primary school model

Secondary school model

Fo

llow

-up

Bas

elin

e

Follo

w-u

p

Bas

elin

e

Urbanicity (Urban vs. Rural) .01 .00 .10** .02

School size .04* .02 .04 -.05

School % FSM .15** .16** .03 .06*

School % EAL .06** .11** -.04 .00

School % SAP and ST .07** .03 .05* .03

School % SA .01 .01 -.03 -.03

School level achievement -.20** -.16** -.12** -.13**

School level absence .06** .06** .17** .11**

School exclusion rates .09** .06** .01 .06*

Year group (Year 5 vs. Year 1) .05** -.01 .00 .08**

Season of birth: Autumn .05* .05* -.01 -.02

Winter .01 -.01 .00 -.02

Spring .01 .00 -.01 .01

Summer -.06** -.04 .02 .02

Gender (Female vs. Male) -.14** -.14** -.20** -.17**

37 Correlations were computed using SPSS, Pearson’s r was used for the continuous predictor variables, for categorical variables the correlations reflect a point biserial correlation coefficient.

195

Ethnicity (White British vs.

Other) .01 .04* -.05* -.07**

Free School Meals (Yes vs.

No) .16** .16** .11** .10**

Attendance -.01 .01 -.22** -.18**

Academic achievement -.08** -.06** -.18** -.16**

Positive relationships -.40** -.56** -.40** -.61**

Bullying .41** .64** .32** .61**

Bully role: Victim

-.05**

-.04

-.07**

-.03

Bully .29** .39** .31** .45**

Bully-Victim .28** .42** .27** .48**

Bystander .04* .03 .08** .06*

Not involved -.34** -.50** -.31** -.54**

SEND category: Cognition & Learning

-.17**

-.21**

-.18**

-.24**

Behaviour, emotional and

social development

.38** .42** .25** .34**

Communication & Interaction -.07** -.07** -.03 -.05

Sensory and/or Physical -.02 -.04* -.07** -.07*

Other -.07** -.06** -.03 -.03

SEND support: School Action

-.13**

-.12**

-.10**

-.14**

School Action Plus .12** .11** .04 .05*

Statement .03 .04 .06* .08*

* Correlation significant at p <.05 (one tailed)

** Correlation significant at p <.01 (one tailed)

196

Table 6.3 has been included here to justify the inclusion of these predictor

variables into the present study. The table shows that all the predictor variables

(or at least one group within the categorical variables) displayed a significant

relationship with the behaviour difficulties score at follow up within either the

primary or secondary data sets. The only exception was for school % SA38

. As

all but one variable displayed a least some relationship with the outcome

variable they were included within the study.

There are some key differences between the primary and secondary models.

Significant correlations between behaviour difficulties at follow up and season of

birth, school size and school % EAL were noted within the primary but not the

secondary model. Conversely, significant correlations between behaviour

difficulties at follow up and attendance and school urbanicity were noted within

the secondary but not the primary model.

Looking at the strength of these correlations it can be seen that the variables

bullying and positive relationships had the highest correlation coefficients.

Positive relationships at follow-up in the primary and secondary models were

both -.40. The bullying mean had a correlation coefficient at follow-up in the

primary model of .40 and in the secondary model of .32. Other variables with

high correlation coefficients within the data set were bully role and the BESD

group on SEND category variable, as well as gender, FSM eligibility and school

level achievement.

It should be noted that although the majority of the correlations were significant

at the p < .01 level, they were mostly weak in nature. Few reached beyond +/-

0.30 which would indicate a medium effect, (Cohen 1992). The reason for the

weak, but nonetheless statistically significant correlations may in fact reflect

issues with the sample size. In studies with larger sample sizes (as in the case

of the present study) there is a heightened sensitivity to detect effects (Cohen

1992).

38 This variable was still included within the final analyses as a school’s percentage of children at school action remains a pertinent variable when explicitly investigating a special educational needs population.

197

6.4.3 Research question 1a.

What proportions of variance in behaviour difficulties are attributable to the school and individual levels?

This question was analysed by running separate multi-level models on the

primary and secondary school data sets (see section 5.9.2 for a justification why

the data was split by school type). Typically when analysing data with multi-level

models, empty (or null) models are calculated in the first instance. From such

models the total amount of unexplained variance within each of the levels of the

model can be accounted for (Heck et al., 2010). This statistic is known as the

ICC (see section 5.9.4), and within the context of the present study shows the

percentage of variance in behaviour difficulties that is attributable at the

individual and school levels respectively.

Empty multi-level models for the primary and secondary schools are presented

below, see tables 6.4 and 6.5. These models show how much variance in

follow-up behaviour difficulties (after controlling for baseline behaviour

difficulties) is attributed to the individual and school levels, before adding in any

of the predictor variables.

Table 6.4: Empty multi-level model for primary school data

Empty Model: Primary Schools (β0ij = 0.197 (0.019) -2*log likelihood = 3973.122

Co-efficient Standard error

p value ICC

SCHOOL LEVEL 0.043 0.006 <.001 15.4%

INDIVIDUAL LEVEL 0.237 0.007 <.001 84.6%

Behaviour difficulties mean score: baseline

0.587 0.014 <.001 --

198

As can be seen from table 6.4 a multi-level analysis showed that for the primary

model there was a significant school (µ0j = 0.043, p <.001) and individual effect

(µ0ij = 0.237, p <.001) on behaviour difficulties. The ICC showed that the overall

school variance accounted for 15.4% of total variance in follow-up behaviour

problems (after controlling for baseline behaviour), whereas 84.6% of the

variance lay at the individual level.

Table 6.5. Empty multi-level model for secondary school data

As can be seen from table 6.5 a multi-level analysis showed that for the

secondary model there was a significant school (µ0j = 0.050, p <.001) and

individual effect (µ0ij = 0.336, p <.001) on behaviour difficulties. The ICC

showed that the overall school variance accounts for 13.0% of total variance in

follow-up behaviour difficulties (after controlling for baseline behaviour

difficulties), whereas 87.0% lay at the individual level.

Empty Model: Secondary Schools (β0ij = 0.327 (0.038) -2*log likelihood = 2926.001


p value ICC

SCHOOL LEVEL 0.050 0.014 <.001 13.0%

INDIVIDUAL LEVEL 0.336 0.012 <.001 87.0%

Behaviour Difficulties Mean Baseline

0.531 0.020 <.001 --

199

6.4.4 Research question 1b.

Which school and individual level predictors explain a statistically significant proportion of variance in behaviour difficulties?

The second step with multi-level models is to run a full model with all predictors

at the various levels included. From such models the significant predictors can

be acknowledged at both the individual and school levels. As before, the

behaviour difficulties mean score at follow up was the outcome, controlling for

baseline behaviour difficulties. Variables at the school level and individual level

were then added to the model. Continuous and binary variables are easily

added whereas for categorical variables with more than two groups a reference

category is required. In each case the group with the largest sample size was

selected (Field 2009). This was the ‘summer’ group, for the season of birth

variable; the ‘cognition and learning’ group for SEND category variable; ‘school

action’ for the SEND support variable; and ‘not involved’ for the bullying role

variable.

The full multi-level models indicate whether the independent variables are

significant predictors of behaviour difficulties. Variables that are significantly

related to behaviour difficulties in a positive direction, i.e. predicting increases in

these problems have been termed risk factors. The factors that are related to

behaviour difficulties in a negative direction, i.e. predicting decreases in

behaviour difficulties have been termed promotive factors.

This conceptualisation between risk and promotive factors is straightforward for

continuous variables (i.e. high attendance versus low attendance) and binary

variables (i.e. male versus female) as risk is labelled at one end and promotive

at the opposite end. The issue becomes more complex with categorical

variables with multiple groups, as these categories do not have a clearly

defined opposite group. Categorical variables are however, added to MLM by

using a reference group. Therefore, the groups within the categorical variables

that remain significantly significant and related to increases in behaviour

difficulties are noted as risk factors, (and the reference group becomes a

200

promotive factor). Alternatively the groups of the categorical variables that are

statistically significant but related to decreases in behaviour difficulties are

noted as promotive factors (and the reference group therefore becomes a risk

factor).

Two full multi-level models have been produced for the primary school data,

and two for the secondary school data. The different models have been

presented to show the effect when the SEND category variable is included and

when it is excluded from the analysis. This decision was made in order to better

understand whether BESD (one of the groups of the SEND category variable) is

a synonymous term with behaviour difficulties as measured in the present

study, or whether it is conceptually distinct. See section 2.2.9 for a further

justification for including both models.

The primary school models are display in table 6.6a and 6.6b, and the

secondary school models is table 6.7a and 6.7b. Results are compared before

a decision is taken as which full multi-level model will be selected for

interpretation for the primary data and which model will be selected for

interpretation for the secondary data.

201

Table 6.6a. Full multi-level model for primary school data (including SEND

category variable)

Full model: Primary Schools (β0ij = 1.152 (0.282) -2*log likelihood = 2588.105


p value

SCHOOL LEVEL

(ICC = 13.7%)

0.032

0.006

<.001

Urbanicity (if Urban) 0.001 0.052 .983

School size 0.001 0.017 .639

School % FSM 0.001 0.002 .692

School % EAL -0.000 0.001 .743

School % SA -0.004 0.003 .128

School % SAP and ST -0.001 0.003 .830

School level achievement -0.006 0.002 <.001


-0.035 0.018 .056

School exclusion rates 0.002 0.020 .902

INDIVIDUAL LEVEL

(ICC = 86.3%)

0.202

0.007

<.001

Behaviour mean baseline 0.419 0.026 <.001

Year group (if Year 5) 0.097 0.025 <.001

Season of birth:

0.071

0.031

.020 Autumn

Winter 0.026 0.028 .354

Spring 0.028 0.029 .333

202

39*’s = Reference categories

Summer *39 * *

Gender (if Male) 0.081 0.023 <.001

Ethnicity (if White British) -0.015 0.033 .658

Free school meals (if Yes) 0.070 0.025 .004

SEND category:

*

*

* Cognition & Learning

Behaviour, emotional and social development

0.269 0.036 <.001

Communication & Interaction -0.014 0.033 .677

Sensory and/or Physical -0.009 0.100 .927

Other 0.095 0.064 .138

SEND support:

*

*

* School Action

School Action Plus 0.041 0.027 .123

Statement 0.048 0.061 .430

Attendance -0.001 0.002 .540

Academic achievement -0.028 0.012 .024

Positive relationships -0.096 0.025 <.001

Bullying 0.032 0.030 .295

Bully role:

-0.076

0.048

.113 Victim

Bully 0.221 0.053 <.001

Bully-Victim 0.035 0.045 .429

Bystander 0.046 0.057 .409

Not involved * * *

203

Table 6.6b Full multi-level model for primary school data (excluding SEND

category variable)

40*’s = Reference categories

Full model: Primary Schools (SEND Category removed) (β0ij = 1.137 (0.281) -2*log likelihood = 2730.444


p value

SCHOOL LEVEL

(ICC = 13.3%)

0.032

0.006

<.001

Urbanicity (if Urban) -0.004 0.052 .936

School size 0.007 0.017 .684

School % FSM 0.001 0.002 .678

School % EAL -0.000 0.001 .704

School % SA -0.004 0.003 .170

School % SAP and ST 0.007 0.003 .841

School level achievement -0.006 0.001 <.001


-0.030 0.018 .103


INDIVIDUAL LEVEL

(ICC = 86.7%)

0.208

0.007

<.001



Season of birth:

0.077

0.031

.012 Autumn

Winter 0.030 0.028 .282

Spring 0.042 0.029 .147

Summer *40 * *

204

Gender (if Male) 0.092 0.023 <.001

Ethnicity (if White British) -0.025 0.033 .455


SEND support:

*

*

* School Action

School Action Plus 0.047 0.025 .060

Statement 0.029 0.056 .605

Attendance -0.001 0.002 .676



Bullying 0.017 0.030 .580

Bully role:

-0.069

0.048

.147 Victim

Bully 0.242 0.053 <.001

Bully-Victim 0.049 0.044 .273

Bystander 0.037 0.056 .512

Not involved * * *

205

Model comparison: Primary schools Comparing between the models (when SEND category variable is included and

excluded), the results are fairly similar; all but one predictor variable remained

significant in both models. This variable academic achievement was significant

when the SEND category variable was included (β0ij = -0.028, p < 0.024)

although failed to reach significance when it was excluded (β0ij = -0.008, p <

0.462). This suggests there is a confounding relationship between SEND

category and academic achievement, with academic achievement only being a

risk factor when SEND category is acknowledged. For all other variables there

was minimal change in the strength of the coefficients between these two

models. None of these coefficients either became significant or ceased being

significant when the SEND category variable was removed.

Table 6.6a shows that when the SEND category variable was included in the

analyses it became a significant predictor of behaviour difficulties; specifically

being in the BESD group was related to significantly higher behaviour difficulties

(β0ij = 0.269, p < .001). This is not a surprisingly finding and provides evidence

that the dependent variable in the present study i.e. behaviour difficulties, is

strongly related to a SEND measure of behaviour difficulties i.e. being a

member of the BESD group. As this was a significant statistical relationship,

alongside a strong theoretical reason that these concepts are broadly

measuring the same thing, (despite the BESD category potentially incorporating

children displaying internalising as well as externalising type behaviour

problems) using the SEND category variable cannot reliably be conceived as a

risk factor for behaviour difficulties. It could more appropriately be

conceptualised as an alternative measure of these problems. Indeed a child in

the BESD group is likely to display temper tantrums, be aggressive, disruptive

and defiant (DfES 2003) and these are key behaviours measured within the

present study. As SENS category is a potential confound in these models it was

removed from any further analyses, and significant results are reported from the

model with it excluded i.e. table 6.6b.

206

Final predictions (SEND category removed): Primary schools As can be seen from the above model in table 6.6b, there is one significant

predictor at the school level, school level achievement (β0j = -0.006, p < .001).

The negative coefficient shows that as school level achievement increases by

1% (i.e. improves) there is a resulting 0.006 decrease in the behaviour

difficulties mean score. This indicates that higher school achievement is a

promotive factor and lower school achievement is a risk factor for behaviour

difficulties.

There are a total of six significant predictors at the individual level. One of these

is a continuous predictor variable: positive relationships (β0ij = -0.115, p <

.001). The coefficient for this variable is negative showing that a one point

increase in this variable results in a 0.115 decrease in behaviour difficulties.

Higher levels of positive relationships are therefore a promotive factor, with

lower levels are a risk factor, for behaviour difficulties.

There were three significant binary predictor variables of behaviour difficulties.

Firstly, gender, (β0ij = 0.092 p < .001). As the coefficient was positive it shows

being male is a risk factor and being female promotive. Specifically being a

member of the risk group results in a 0.092 increase in the behaviour difficulties

mean score. Secondly, year group (β0ij = 0.091, p < .001). As the coefficient

was positive it shows being in year 5 is a risk factor and being in year 1 is

promotive. Specifically, being a member of the risk group results in a 0.091

increase in the behaviour difficulties mean score. Thirdly, FSM eligibility (β0ij =

0.084, p = .001). The coefficient was positive again, showing eligibility for FSM

is a risk factor and non-eligibility is promotive. Being a member of the risk group

therefore results in a 0.084 increase in the behaviour difficulties mean score.

Finally, two categorical variables emerged as significant predictors of behaviour

difficulties. Firstly, season of birth (β0ij = 0.077, p = .012). This variable showed

that being born in the autumn was a risk factor and the reference category

being born in the summer is promotive. Being a member of the risk group

therefore results in a 0.077 increase in the behaviour difficulties mean score.

Secondly, bullying role was noted as a significant predictor, with being classified

207

as a bully a risk factor (β0ij = 0.242, p < .001). The behaviour difficulties mean

score increasing by 0.242 for those in this group. The reference category was

not involved in bullying which is therefore a promotive factor.

Finally, the coefficient for baseline behaviour difficulties was significant (β0ij =

0.465, p < .001), showing that behaviour difficulties at baseline are a powerful

predictor of later behaviour difficulties at follow-up. As this score was used as a

control in the model, variables that explain follow-up behaviour beyond

controlling for baseline are particularly salient.

The coefficients presented above for the different predictor variables can be

treated as effect sizes. It should be pointed out that the majority are fairly small

in magnitude. This means that large changes on the predictor variables may

only bring about relatively small changes in behaviour difficulties displayed.

Each coefficient however, needs to be interpreted independently on the scale

on which it was measured (see table 5.2), in order to assess its importance in

accounting for behaviour difficulties. Indeed it should be recalled here that the

behaviour difficulties mean score was measured on a scale from 0-3. The risk

and promotive factors are summarised in figures 6.1 and 6.2.

208

Table 6.7a: Full multi-level model for secondary school data (including SEND category variable)

41 *’s = Reference categories

Full model: Secondary Schools (β0ij = 0.299 (0.539) -2*log likelihood = 2002.602


p value

SCHOOL LEVEL

(ICC = 9.3%)

0.030

0.011

.008


School size 0.027 0.013 .036

School % FSM 0.009 0.008 .248

School % EAL -0.005 0.003 .178

School % SA -0.002 0.006 .799

School % SAP and ST -0.007 0.009 .439

School level achievement 0.008 0.004 .065


0.062 0.039 .123

School exclusion rates -0.003 0.011 .758

INDIVIDUAL LEVEL

(ICC = 90.7%)

0.291

0.012

<.001


Year group (if Year 10) -0.087 0.034 .011

Season of birth:

0.008

0.045

.859 Autumn

Winter -0.020 0.045 .650

Spring -0.003 0.048 .955

Summer *41 * *

209

Gender (if Male) 0.091 0.036 .011

Ethnicity (if White British) 0.032 0.053 .539


SEND category:

*

*

* Cognition & Learning Behaviour, emotional and

social development 0.075 0.043 .080

Communication & Interaction

-0.005 0.060 .938

Sensory and/or Physical

-0.049 0.093 .601

Other -0.053 0.101 .601

SEND support:

*

*

* School Action

School Action Plus -0.016 0.039 .688

Statement -0.058 0.061 .344

Attendance -0.010 0.002 <.001


Positive relationships -0.067 0.038 .077

Bullying -0.065 0.041 .116

Bully role:

-0.052

0.062

.406 Victim

Bully 0.199 0.073 .006

Bully-Victim 0.128 0.074 .085

Bystander 0.195 0.086 .023

Not involved * * *

210

Table 6.7b: Full multi-level model for secondary school data (excluding SEND category variable)

42 *’s = Reference categories

Full model: Secondary Schools (SEND Category removed) (β0ij = 0.262 (0.553) -2*log likelihood = 2047.946


p value

SCHOOL LEVEL

(ICC = 10.3%)

0.034

0.013

.007


School size 0.031 0.013 .018

School % FSM 0.014 0.008 .092

School % EAL -0.006 0.004 .072

School % SA -0.003 0.006 .579

School % SAP and ST -0.009 0.009 .338



0.056 0.041 .174


INDIVIDUAL LEVEL

(ICC = 89.7%)

0.297

0.012

<.001


Year group (if Year 10) -0.092 0.034 .007

Season of birth:

0.001

0.045

.980 Autumn

Winter -0.027 0.045 .543

Spring -0.011 0.048 .814

Summer *42 * *

211

Gender (if Male) 0.092 0.036 .010

Ethnicity (if White British) 0.046 0.053 .384


SEND support:

*

*

* School Action

School Action Plus -0.009 0.038 .816

Statement -0.065 0.059 .266

Attendance -0.010 0.002 <.001


Positive relationships -0.072 0.038 .059

Bullying -0.049 0.041 .232

Bully role:

-0.068

0.062

.271 Victim

Bully 0.189 0.072 .009

Bully-Victim 0.111 0.073 .130

Bystander 0.209 0.085 .014

Not involved * * *

212

Model comparison: Secondary schools Comparing between the models (when SEND category variable is included and

excluded), the results are fairly similar. There was no change in whether any of

the predictor variables became or ceased becoming significant as a result of

exclusion of the SEND category variable.

Table 6.7a shows a surprising finding that when the SEND category variable

was included within the analyses, being in the BESD category was not

established as a risk factor. Being in this group did approach statistical

significance, although remained a marginal non-significant trend (p = .080). One

explanation why this factor failed to emerge as significant is that the BESD

group does not only include those children who display externalising problems,

such as the behaviour difficulties measured within the present study, but also

internalising behaviours such as anxiety and depression (DfES 2003), and

these behaviours were not accounted for within the present study. Furthermore,

internalising behaviours such as anxiety and depression are more abundant in

secondary compared with primary students (Green et al. 2005), which could

account for the null finding in the secondary compared with the primary model.

If the BESD group of the SEND category variable (which is designed to include

those with behaviour difficulties), fails to become a significant predictor of these

behaviours something is not right. There is perhaps a requirement for an

overhaul of SEND categorisation, particularly around the BESD group, which is

an issue that has been recently acknowledged by the Department for

Education’s green paper entitled ‘support and aspiration: A new approach to

special educational needs and disability’ (DfE 2011c). Conceptualising the

BESD group as a predictor for behaviour difficulties is therefore not the most

useful finding. It is confounded as it does include those children displaying

behaviour difficulties as measured in the present study although alongside

internalising problems such as anxiety and depression. As such the SEND

category variable is seen more as an alternative measure of behaviour

problems rather than a risk factor. Therefore, as in the primary model this

variable was excluded from any further analyses, and significant results are

reported from the model with it excluded i.e. table 6.7b.

213

Final predictions (SEND category removed): Secondary schools As can be seen from the above model in table 6.7b, there is one significant

predictor at the school level, school size (β0j = 0.031 p = .018). The positive

coefficient shows that as school size increases by 100 pupils on roll there is a

resulting 0.031 increase in the behaviour difficulties mean score. This indicates

that larger schools are a risk factor and smaller schools a promotive factor for


There are six significant predictors at the individual level. Two of these are

continuous predictor variables: academic achievement (β0ij = -0.044, p = .017)

and attendance (β0ij = -0.010, p < .001). The negative coefficients show that a

one point increase in academic achievement results in a 0.044 decrease in the

behaviour difficulties mean score, and a one percent increase in attendance

results in a 0.010 decrease in behaviour difficulties mean score. Higher

academic achievement and higher attendance are promotive factors and lower

academic achievement and attendance are risk factors for behaviour difficulties.

There were three significant binary predictor variables of behaviour difficulties.

Firstly, gender, (β0ij = 0.092 p = .010). As the coefficient was positive it showed

that being male is a risk factor and being female promotive. Specifically, being

male is a risk which results in a 0.092 increase in the behaviour difficulties

mean score. Secondly, year group (β0ij = -0.092, p = .007). As the coefficient

was negative it shows being in year 7 is a risk factor and being in year 10

promotive. Specifically, being a member of the risk group results in a 0.092

increase in the behaviour difficulties mean score. Thirdly, FSM eligibility (β0ij =

0.082, p = .029). The coefficient was positive again showing eligibility for FSM is

a risk factor and non-eligibility is promotive. Specifically, being eligible for FSM

results in a 0.082 increase in the behaviour difficulties mean score.

One categorical variable emerged as a significant predictor of behaviour

difficulties although two groups within it acted as risk factors. Bullying role was

the significant predictor and specially being a bully (β0ij = 0.189, p = 0.009) or a

bystander to bullying (β0ij = 0.209, p = .014) were risk factors. In both cases not

214

involved in bullying was the reference category and therefore acts as a

promotive factor.

Finally, the coefficient for baseline behaviour difficulties was significant (β0ij =

.430, p < 0.001), showing that behaviour difficulties at baseline is a powerful

predictor of later behaviour difficulties at follow-up. As this score was used as a

control in the model, variables that explain follow-up behaviour difficulties

beyond controlling for the baseline measure are especially important.

As mentioned above, the coefficients represent effect sizes; with the figures

showing each predictor variable’s relative impact on the behaviour difficulties

mean score. The coefficients are not directly comparable with one another as

account needs to be taken of the scale on which they were measured. As in the

primary model the coefficients are relatively small in comparison to the

behaviour difficulties mean score, which can range from 0-3. Therefore, fairly

large changes in these variables are needed to have an observable impact on

behaviour difficulties. Nonetheless, these variables remain significant risk and

promotive factors for behaviour difficulties. The risk and promotive factors are

summarised in figures 6.1 and 6.2.

215

The Venn diagrams display the significant risk and promotive factors for

behaviour difficulties. These diagrams allow an easy comparison between the

primary and secondary models to be made as included are those factors unique

to the primary and secondary schools respectively and also the factors that are

salient for both school types.

Figure 6.1: Risk factors for primary and secondary schools

Figure 6.2: Promotive factors for primary and secondary schools

216

6.4.5 Research question1c.

Of the variance initially attributed to the individual and school levels, how much of this is explained by the predictors used in the study?

In order to answer this question, a comparison was taken between the empty

and full models (both displayed above) for primary and secondary schools

models separately. These comparisons assessed the amount of variance to be

explained within the empty model that can be explained by the full model.

Subtracting the variance accounted for in the full model (with SEND category

removed) from the total variance to be explained in the empty model, allowed

for a percentage of total variance explained to be calculated which can be used

as an overall model fit estimate.

A Chi Square analysis assessed the differences between the -2*log likelihood in

the empty and full models in order to assess whether the full model, including all

the predictors, was significantly better than the empty model in accounting for

behaviour difficulties. The results of these analyses are presented in tables 6.8

and 6.9.

Table 6.8: Comparison between the empty and full models for primary schools

Primary school model

Empty model Full model % Variance explained

School level variance

0.043 0.032 25.6%

Individual Level variance

0.237 0.208 12.2%

Total Variance 0.280 0.240 14.3%

-2*log likelihood 3986.083 2730.444

Chi Square χ² (26, n = 2660) = 1255.639, p <.001

217

The decrease in -2*log likelihood from empty to full models indicates that adding

in the predictors to the model resulted in a better model fit. This is confirmed by

the Chi Square test showing this decrease to be significant. Subtracting the

total variance accounted for in the full model from the variance to be explained

from the empty model allows a percentage of total variance that the full model

can explain to be calculated. This figure was 25.6% at the school level and

12.2% at the individual level. This equates to a total model fit of 14.3%. That is

when all the predictors were added they could explain 14.3% of the variance in

behaviour difficulties across both school and individual levels.

Table 6.9: Comparison between the empty and full models for secondary

schools

The decrease in -2*log likelihood from empty to full models indicates that adding

in the predictors to the model resulted in a better model fit. This is confirmed by

the Chi Square test showing this decrease to be significant. Subtracting the

total variance accounted for in the full model from the variance to be explained

from the empty model allows a percentage of total variance that the full model

can explain to be calculated. This figure was 32.0% at the school level and

11.6% at the individual level. This equates to a total model fit of 14.2%. That is,

when all the predictors were added they could explain 14.2% of the variance in

behaviour difficulties across both school and individual levels.

Secondary school model

Empty model Full model % Variance explained

School level variance

0.050 0.034 32.0%

Individual Level variance

0.336 0.297 11.6%

Total Variance 0.386 0.331 14.2%

-2*log likelihood 2926.001 2047.946

Chi Square χ² (26, n = 1628) = 878.055, p <.001

218


• Within the primary and secondary models both school and individual

levels significantly contributed to explaining variance in behaviour

difficulties at follow-up.

• Within both primary and secondary school models, the individual level

explains a considerably larger proportion of this variance compared with

the school level (primary = 15.4% at school level versus 84.6% at

individual level, secondary = 13.0% at school level versus 87% at

individual level).

• In the primary school model, one significant risk factor was found at the

school level: low school level achievement. Six significant risk factors

were found at the individual level. These were: being male, being born in

the autumn, being in year 5, eligible for FSM, being a bully, and having

poor positive relationships.

• In the secondary school model, large school size was the only significant

risk factor at the school level. Seven significant risk factors were found

within the individual level. These were, being male, being in year 7,

eligible for FSM, being a bully or bystander, having low academic

achievement, and poor attendance.

• The size of the coefficients are all relatively small, suggesting that large

changes in the predictor variables are needed for relatively small

changes in the behaviour difficulties mean score.

• The full primary model can account for 14.3% of total variance in

behaviour difficulties at follow-up after controlling for baseline behaviour,

whereas the full secondary model can account for 14.2%.

219

6.5 Cumulative effects of risk factors:

6.5.1 Introduction to section

The aim of this section is to answer the 3 sub-questions within research

question 2, these are:








difficulties?

The section begins with a description of how a cumulative risk score is

generated before presenting the descriptive statistics followed by each sub

question and the analysis undertaken to answer it. The section concludes with a

summary of the findings.

6.5.2 Cumulative risk score

A cumulative risk score was generated in accordance with previous studies laid

out in the literature review (see section 3.4). The first step involved selecting the

significant risk factors for behaviour difficulties highlighted in research question

1. There were a total of seven risk factors in the primary school model (6 at the

individual level and 1 at the school level) and eight risk factors within the

secondary school model (7 at the individual level and 1 at the school level).

In keeping with previous literature (Ribeaud & Eisner 2010, Deater-Deckard et

al. 1998, Atzaba-Poria et al. 2004) cumulative risk scores were generated within

a specified ecological domain, i.e. within the individual level or within the school

220

level, but not across levels. With only one significant predictor at the school

level for both primary and secondary schools it was not possible to generate a

cumulative score for at this level. Although these factors are important in

explaining variance in individual behaviour difficulties, the pertinent issue within

this section remains the cumulative effects of predictors and how they work

together within a single ecological level. Secondly only contextual risk factors

are used within cumulative risk scores (Flouri & Kallis 2007). Therefore, three

risk variables from the primary school model (i.e. gender, year group and

season of birth) and two from the secondary model (i.e. gender and year group)

were not added to the cumulative risk score but added to these models as

covariates.

The contextual risk factors in the primary model that comprise the cumulative

risk score were low positive relationships, being eligible for FSM and being a

bully. In the secondary model the contextual risk factors composed of low

academic achievement, low attendance, eligibility for FSM and being a bully or

bystander. These risk factors were therefore calculated into a cumulative risk

score. For binary and categorical variables, the group which denoted risk was

coded as 1 and all other categories as 0. For continuous variables the top (or

bottom) 25% of cases that were related to increased behaviour difficulties were

coded as 1 and other scores as 0. Risks were added together to generate a

cumulative score for each participant. Higher numbers indicated the individual

was at an increased risk.

221

6.5.3 Descriptive statistics

Table 6.10: Total number and percentage of participants per each risk group for the primary and secondary school models

Table 6.10 shows that within the primary and secondary data sets, the vast

majority of participants had between 0 and 1 risk. In both models as the number

of risks increases the numbers of participants in the risk group falls. In the

secondary model less than 1% of participants were found at the extreme end of

the risk scale (i.e. with high levels of risk). As this could potentially skew the

findings, these were recoded. Those with 4 risks were recoded as 3 and the

category becoming 3+ risks. This procedure is consistent with Gerard & Buehler

(2004a), Appleyard et al. (2005), and Raviv et al. (2010).

Table 6.11. Mean cumulative risk scores for the primary and secondary models

Mean Standard deviation

Maximum Minimum

Primary 0.60 0.72 3 0

Secondary 0.92 0.88 3 0

As can be seen from table 6.11 the mean cumulative risk score in the primary

model was 0.60 whereas for the secondary model this was higher at 0.92.

Primary schools Secondary schools

Number of risks

Number of sample with

each risk

Percentage of risks

Number of sample with

each risk

Percentage of risks

0 1390 52.3% 621 38.1%

1 965 36.3% 606 37.2%

2 271 10.2% 310 19.0%

3 33 1.2% 82 5.0%

4 - - 9 0.6%

222

6.5.4 Research question 2a

Is there a cumulative effect of contextual risk factors on behaviour difficulties, where higher numbers of risk factors present are associated with increased levels of behaviour difficulties?

Multi-level models were then utilised to analyse the data in order to establish

whether a cumulative effect existed within the data. Behaviour difficulties at

follow up was the outcome variable (controlling for baseline behaviour), the

predictor variables included the cumulative risk score, and the non-contextual

risk factors were added as covariates. Tables 6.13 and 6.14 show these results

for the primary and secondary models.

Table 6.12: Cumulative risk multi-level model for the primary school data

Primary cumulative risk model β0ij = 0.224 (0.039) -2*log likelihood = 3903.131

Coefficient Standard error p value

SCHOOL LEVEL (ICC = 15.4%)

0.042 0.006 <.001

INDIVIDUAL LEVEL (ICC = 84.6%)

0.231 0.007 <.001

Behaviour baseline 0.533 0.016 <.001

Gender (if male) 0.091 0.021 <.001

Year group (if year 5) 0.090 0.021 <.001

Season of birth:

Autumn

0.049

0.028

.077

Winter 0.077 0.031 .029

Spring 0.030 0.028 .229

Summer - - -

CUMULATIVE RISK SCORE

0.089 .016 <.001

223

Table 6.13: Cumulative risk multi-level model for the secondary school data

The above two models show that the cumulative risk score is a strongly

significant predictor of behaviour difficulties for both the primary and secondary

models. Specifically a positive relationship emerges for both the primary (β0ij =

0.089, p = <.001) and secondary models. (β0ij = 0.124, p = <.001). This

evidence shows that each additional risk increases the behaviour difficulties

score at follow-up by 0.089 in primary schools and 0.124 in secondary schools.

Results therefore support the cumulative risk hypothesis and that number of

risks within an individual’s background is predictive of increased behaviour

difficulties.

Secondary cumulative risk model β0ij = 0.418 (0.061) -2*log likelihood = 2864.192

Coefficient Standard error p value


0.042 0.012 <.001


0.325 0.012 <.001


Gender (if male) 0.099 0.032 .002

Year group (if year 10) 0.066 0.030 .027

CUMULATIVE RISK SCORE

0.124 0.018 <.001

224

6.5.5 Research question 2b

What is the nature of the relationship between exposure of cumulative risk and behaviour difficulties?

In order to answer this question and assess whether increasing cumulative risk

has a proportional increase on behaviour difficulties or whether there is any

evidence of an accelerative effect on behaviour difficulties, two further models

were computed. The cumulative model (assessing a linear relationship) was

assessed against the quadratic model (assessing a non-linear accelerative

relationship) to establish the best fitting model.

Table 6.14: The mean (and standard deviations) of behaviour difficulties at baseline and follow up for each risk group

Primary schools Secondary schools

Number of risks

Mean baseline

behaviour

Mean follow up

behaviour

Mean baseline

behaviour

Mean follow up

behaviour

0 0.38 (0.50) 0.39 (0.54) 0.29 (0.54) 0.39 (0.58)

1 0.71 (0.69) 0.63 (0.66) 0.59 (0.76) 0.61 (0.71)

2 1.46 (0.76) 1.15 (0.79) 0.87 (0.84) 0.89 (0.85)

3 2.05 (0.62) 1.66 (0.72) 1.18 (0.91) 1.30 (0.90)

One way to test statistically for a non-linear relationship over a linear

relationship would be to add a squared term of the cumulative risk score into the

model (Aiken & West 1991). If the squared cumulative risk score (i.e. a

quadratic term) accounts for additional variance beyond the cumulative risk

score (i.e. a linear term) and results in a better overall model fit, it could be

concluded that an disproportional relationship between risk and behaviour

difficulties is present within the data. Due to the issue of multicollinearity

however, it was essential to mean centre the cumulative risk score before

squaring it to create the additional variable.

225

This analysis was conducted in two stages, firstly the squared term of the

cumulative risk score was generated. Secondly, this term was added to the

model after accounting for the cumulative risk score, the resulting model has

been termed a quadratic model (Gerard & Buelher 2004a). If the squared term

accounts for variance beyond the cumulative risk score it can be concluded it is

a better fit of the data, and a non-linear i.e. quadratic relationship exists.

Results of these analyses are presented below in tables 6.15 and 6.16. These

tables show the empty model, the cumulative model (which comprises the

cumulative risk score) and the quadratic model, (which adds in the cumulative

risk score squared as an additional predictor in the model). For each model, the

behaviour mean score at baseline is included as a control, along with the

significant non-contextual risk factors from research question 1 added as

covariates.

Within the primary school data, and as can be seen from table 6.15 the best

model fit occurred for the quadratic model, as there was a significant reduction

in the log likelihood value from the cumulative to the quadratic model (χ²(1, n

=2660) = 4.731 p < .05). The squared cumulative risk term within the quadratic

model was significant (β0ij = 0.035, p = .029), even after accounting for the

cumulative risk score. This suggests that a quadratic relationship is a better fit

of the primary school model and accounts for variance beyond the linear term.

Within the secondary school data, and as can be seen from table 6.16, the best

model fit occurred for the quadratic model, as there was a significant reduction

in the log likelihood value from the cumulative to the quadratic model (χ²(1, n

=1628) = 3.842 p < .05). The squared cumulative risk term within the quadratic

model was significant (β0ij = 0.034, p = .049), even after accounting for the

cumulative risk score. This suggests that a quadratic relationship is a better fit

of the secondary school model and accounts for variance beyond the linear

term.

226

Table 6.15: Empty, Cumulative and Quadratic multi-level model for primary school data

Empty model

(β0ij = 0.197 (0.019)

Cumulative model

β0ij = 0.224 (0.039)

Quadratic model

β0ij = 0.263 (0.040) Coefficient Standard

error p

value Coefficient Standard

Error p


error p

value School Level (ICC = 15.4%)

0.043 0.006 <.001 School Level (ICC = 15.4%)

0.042 0.006 <.001 School Level (ICC = 15.3%)

0.041 0.006 <.001

Individual level (ICC = 84.6%)

0.237 0.007 <.001 Individual level (ICC = 84.6%)


0.227 0.006 <.001

Behaviour baseline

0.587 0.014 <.001 Behaviour baseline


0.528 0.016 <.001

Gender 0.091 0.021 <.001 Gender 0.089 0.021 <.001

Year group 0.090 0.021 <.001 Year group 0.088 0.021 <.001

Season Season

Autumn 0.049 0.028 .077 Autumn 0.049 0.028 .076

Winter 0.077 0.031 .029 Winter 0.029 0.026 .259

Spring 0.030 0.028 .229 Spring 0.031 0.026 .239

Cumulative risk score

0.089 .016 <.001 Cumulative risk score

0.067 0.019 <.001

Cumulative risk score squared

0.035 0.016 .029

-2*log likelihood = 3973.122 -2*log likelihood = 3903.131 -2*log likelihood = 3898.400 χ²(6, n =2660) = 69.991, p<.001 χ²(1, n =2660) = 4.731 p < .05

227

Table 6.16: Empty, Cumulative and Quadratic multi-level model for secondary school data

Empty model

(β0ij = 0.197 (0.019)

Cumulative model

β0ij = 0.224 (0.039)

Quadratic model

β0ij = 0.263 (0.040) Coefficient Standard

error p


Error p


error p

value School Level (ICC = 15.4%)

0.531 0.020 <.001 School Level (ICC = 11.4%)

0.042 0.012 <.001 School Level (ICC = 15.3%)

0.042 0.012 <.001

Individual level (ICC = 84.6%)



0.324 0.012 <.001

Behaviour baseline



0.476 0.021 <.001

Gender 0.325 0.012 <.001 Gender 0.099 0.032 .002

Year group 0.476 0.021 <.001 Year group 0.063 0.030 .035

Cumulative risk score

0.099 0.032 .002 Cumulative risk score

0.105 0.020 <.001

Cumulative risk score squared

0.034 0.017 .049

-2*log likelihood = 2926.001 -2*log likelihood = 2864.192 -2*log likelihood = 2860.320 χ²(3, n 1628) = 61.809, p<.001 χ²(1, n =1628) = 3.842 p < .05

228

6.5.6 Research question 2c. Is the number of risks present within an individual’s background more important than the specific types of risks in accounting for behaviour difficulties?

In order to answer this question a comparison was made between the

cumulative risk model assessing the importance of number of risk factors within

an individual’s background versus an independent additive model that accounts

for the specific types of risks. Both models acknowledge the same risk factors,

however, in the cumulative model (presented in section 6.5.4) contextual risks

are dichotomised and entered as a single predictor, and non-contextual risks

are added as covariates. In the independent additive risk model, risks are

added in their original form as separate predictors.

229

Table 6.17: The independent additive models for the primary school data

A comparison of the independent additive model and cumulative model is

conducted by observing the differences in Schwarz’s Bayesian criterion. This is

an adjusted version of the -2*log likelihood values and accounts for the fact that

these two models are not nested (Flouri 2008). Schwarz’s Bayesian criterion for

cumulative model (3981.988) – independent additive model (3679.777) =

608.107, (χ² (5, n =2660) = 302.211, p < .001). There was a significant

reduction in this value from the cumulative to the independent additive model.

This indicated that the independent additive model is a significantly better fitting

model.

43 Summer is the reference category, other groups of this variable were non-significant (i.e. winter and spring) and were added to the model although are not reported here. 44 Not involved is the reference category, other categories of this variable were non-significant (i.e. victim, bully-victim, bystander) and were added to the model although are not reported here.

Primary independent additive risk model (β0ij = 0.494 (0.064) -2*log likelihood = 3562.452 Schwarz’s Bayesian Criterion = 3679.777

Coefficient Standard error

p value


0.043 0.006 <.001


0.221 0.007 <.001



Season of birth (Autumn vs. Summer43

0.059

) 0.029 .042

Gender (if Male) 0.089 0.021 <.001

Free school meals (if Yes) 0.089 0.022 <.001


Bully role: (Bully vs. not involved)44 0.199 0.049 <.001

230

Table 6.18: The independent additive models for the secondary school data

A comparison of the independent additive model and cumulative model is

conducted by observing the differences in Schwarz’s Bayesian criterion. This is

an adjusted version of the -2*log likelihood values and accounts for the fact that

these two models are not nested (Flouri 2008). Schwarz’s Bayesian criterion for

cumulative model (2864.192) – independent additive model (2325.537) =

608.107, (χ² (6, n =1628) = 538.655, p < .001). There was a significant

reduction in this value from the cumulative to the independent additive model.

This indicated that the independent additive model is a significantly better fitting

model.

45 Not involved is the reference category, other categories of this variable were non-significant and were added to the model although are not reported here.

Secondary independent additive risk model (β0ij = 1.523 (0.224) -2*log likelihood = 2232.186 Schwarz’s Bayesian Criterion = 2325.537

Coefficient Standard error

p value


0.051 0.016 .001


0.302 0.012 <.001


Year group (if Year 10) -.0.096 0.033 .004

Gender (if Male) 0.100 0.035 .004


Attendance -0.011 0.002 <.001


Bully role: (Bully vs. not involved)45 0.215 0.071 .002

Bully role: (Bystander vs. not involved)

0.211 0.085 .013

231


• Increasing levels of cumulative risk result in more severe behaviour

difficulties displayed within both primary and secondary schools.

• This relationship represents a non-linear quadratic pattern as with

increasing risk, there is a disproportional increase in behaviour difficulties

displayed. This suggests that as risk increases there is a point of mass

accumulation where the cumulative risk scores had a greater effect than

the sum of its individual parts.

• The independent additive model was a statistically significant better

fitting model than the cumulative model in both primary and secondary

schools. This suggests that the specific type of risk is more important

than number of risk factors present, in accounting for behaviour

difficulties displayed.

232

6.6 Protective factors

6.6.1. Introduction to section

The aim of this section is to answer the 2 sub-questions within research

question 3, these are: a) Which school level predictor variables have a statistically significant




The purpose of this section is to establish if any of the school level variables

within the study act as protective factors (moderators), that reduce the influence

risk has on behaviour difficulties. Moderators will therefore be particularly

beneficial for those experiencing high degrees of risk.

The following analyses looked for interaction effects between the school level

predictor variables and the variables which made up the contextual cumulative

risk score. The decision to look at these risk variables in isolation rather than as

a cumulative risk score was taken on the basis that the independent additive

model was a better fitting model than the cumulative model in accounting for

behaviour difficulties displayed (see section 6.5.6). Therefore, in the primary

school model, risk is associated with being eligible for FSM, being a bully and

having low positive relationships. In the secondary school model risk is

associated with being eligible for FSM, being a bully or a bystander, low

academic achievement and low attendance.

The focus within the present study remains on the school level variables

investigating whether factors at this ecological level can moderate the risk

experienced at the individual level. This focus has been justified within the

literature review (see section 2.1 and section 4.7), as little research has

explicitly acknowledged the importance of school influences in reducing

behaviour difficulties, particularly for those at high degrees of risk.

233

6.6.2 Research question 3a Which school level predictor variables have a statistically significant interaction effect with contextual risk?

The variables within these analyses included, firstly measures of risk which

were noted as the contextual risk factors in research question 2 and comprised

the cumulative risk score46. Secondly, the significant risk factors which were

non-contextual and did not make up the cumulative score and were added as

covariates47. Thirdly, all the school level variables48

. Fourthly, the interaction

terms between the contextual risk factors and the school level variables, (all

centred before being added to the model). Finally, the other variables measured

within this study and included in the full model in research questions 1 (see

section 6.4.4) were also added as covariates.

Two multi-level models (one for primary and one for secondary schools) were

then computed to find any statistically significant interaction effects between the

school level variables and the contextual risk factors. Significant interaction

effects can be taken as evidence that the variables are acting in a protective

manner, moderating the influence risk has on outcome. Results are presented

in table 6.19 and 6.20.

46 These variables included in primary schools, FSM eligibility, being a bully, low positive relationships, and in secondary schools FSM eligibility, being a bully or bystander, low attendance and low academic achievement. 47 These variables were gender, year group and season of birth for primary schools, and gender and year group for secondary schools 48 These included school urbanicity, school size, school % FSM, school % EAL, school % SA, school % SAP & ST, school level overall achievement, school level overall absence and school exclusion rates.

234

Table 6.19: Protective factor model for primary school data

Protective factor model: Primary Schools (β0ij = 0.286 (0.189) -2*log likelihood = 2676.267


p value

SCHOOL LEVEL (ICC = 13.7%) 0.032 0.007 <.001

Urbanicity (Urban vs. Rural) -0.038 0.057 .505

School size -0.006 0.019 ..746

School % FSM 0.002 0.002 .427

School % EAL -0.002 0.001 .823

School % SA -0.005 0.003 .127

School % SAP and ST -0.002 0.004 .994

School level achievement -0.006 0.001 .529

School level overall absence -0.038 0.020 ..067


FSM Risk * Urbanicity 0.081 0.077 .295

FSM Risk * School size 0.043 0.024 .072

FSM Risk * School % FSM -0.002 0.002 .493

FSM Risk * School % EAL -0.001 0.001 .280

FSM Risk * School % SA 0.001 0.003 .792

FSM Risk * School % SAP and ST 0.007 0.005 .119

FSM Risk * School level achievement -0.001 0.002 .791

FSM Risk * School level overall absence 0.015 0.023 .526

FSM Risk * School exclusion rates -0.024 0.025 .350

Positive Relationships Risk * Urbanicity 0.060 0.063 .334

Positive Relationships Risk * School size

-0.026 0.020 .206

Positive Relationships Risk * School % FSM

-0.001 0.002 .632

Positive Relationships Risk * School % EAL

-0.001 0.001 .187

Positive Relationships Risk * School % SA

0.009 0.003 .004

Positive Relationships Risk * School % SAP and ST

-0.002 0.004 .580

Positive Relationships Risk * School level achievement

0.004 0.002 .012

Positive Relationships Risk * School level overall absence

0.030 0.020 .144

Positive Relationships Risk * School exclusion rates

-0.041 0.022 .065

Bully Risk * Urbanicity 0.236 0.147 .109

235

Bully Risk * School size -0.002 0.051 .974

Bully Risk * School % FSM 0.001 0.004 .902

Bully Risk * School % EAL 0.001 0.002 .522

Bully Risk * School % SA 0.003 0.007 .633

Bully Risk * School % SAP and ST

0.000 0.009 .970

Bully Risk * School level achievement -0.001 0.004 .875

Bully Risk * School level overall absence

0.033 0.047 .403

Bully Risk * School exclusion rates -0.085 0.046 .064

INDIVIDUDAL LEVEL (ICC = 86.3%) 0.202 0.007 <.001


FSM 0.009 0.092 .924

Positive Relationships -0.195 0.076 .011

Bully (vs. Not involved) -0.038 0.183 .836

Primary protective model Comparing between the full model (see section 6.4.4) and protective model

(table 6.19), there was a significant reduction in the -2*log likelihood value. Full

model = 2730.444 – Protective model (2676.267) = 54.177, (χ² (27, n =2660) =

54.177, p < .01). This reduction in value from the full model to the protective

model indicates the protective model is a significantly better fitting model. Table

6.19 shows that the vast majority of the interactive terms were non-significant

predictors in the model.

Only two significant interaction effects were found; these were positive

relationships risk * school % of SA (β0ij = 0.009, p =.004); and positive

relationships risk * school level achievement (β0ij = 0.004, p =.012). These

results suggest that each of these predictor variables moderate the effects of

risk on behaviour difficulties.

236

Table 6.20: Protective factor model for secondary school data

Protective factor model: Secondary schools (β0ij = 0.717 (0.119) -2*log likelihood = 2004.807


p value

SCHOOL LEVEL (ICC = 8.4%) 0.026 0.011 .019

Urbanicity (Urban vs. Rural) 0.123 0.130 .349

School size 0.020 0.013 .136

School % FSM 0.010 0.009 .287

School % EAL -0.004 0.004 .367

School % SA -0.004 0.007 .563

School % SAP and ST -0.009 0.009 .346


School level overall absence 0.046 0.042 .283

School exclusion rates -0.000 0.011 .967

FSM Risk * Urbanicity -0.006 0.134 .500

FSM Risk * School size -0.091 0.018 .736

FSM Risk * School % FSM -0.001 0.011 .963

FSM Risk * School % EAL -0.002 0.005 .662

FSM Risk * School % SA -0.009 0.008 .261

FSM Risk * School % SAP and ST 0.001 0.011 .893

FSM Risk * School level achievement 0.000 0.005 .943

FSM Risk * School level overall absence 0.062 0.048 .197

FSM Risk * School exclusion rates -0.000 0.014 .978

Attendance Risk * Urbanicity 0.004 0.008 .626

Attendance Risk * School size 0.001 0.001 .396

Attendance Risk * School % FSM 0.000 0.001 .674

Attendance Risk * School % EAL -0.000 0.000 .619

Attendance Risk * School % SA -0.000 0.001 .424

Attendance Risk * School % SAP and ST -0.000 0.001 .988

Attendance Risk * School level achievement

0.000 0.000 .597

Attendance Risk * School level overall absence

0.003 0.003 .408

Attendance Risk * School exclusion rates

-0.000 0.001 .928

Academic Risk * Urbanicity -0.138 0.067 .038

Academic Risk * School size -0.002 0.009 .774

237

Academic Risk * School % FSM 0.002 0.006 .710

Academic Risk * School % EAL -0.001 0.002 .758

Academic Risk * School % SA -0.006 0.004 .119

Academic Risk * School % SAP and ST -0.002 0.005 .672

Academic Risk * School level achievement

-0.001 0.003 .771

Academic Risk * School level overall absence

-0.017 0.022 .423

Academic Risk * School exclusion rates 0.006 0.008 .456

Bully Risk * Urbanicity -0.076 0.189 .689

Bully Risk * School size 0.012 0.025 .638

Bully Risk * School % FSM 0.015 0.015 .345

Bully Risk * School % EAL -0.003 0.007 .631

Bully Risk * School % SA 0.006 0.012 .617

Bully Risk * School % SAP and ST

-0.118 0.016 .449

Bully Risk * School level achievement 0.006 0.008 .457

Bully Risk * School level overall absence

-0.020 0.070 .775

Bully Risk * School exclusion rates 0.003 0.026 .916

Bystander Risk * Urbanicity 0.093 0.287 .746

Bystander Risk * School size 0.050 0.030 .097

Bystander Risk * School % FSM -0.012 0.029 .711

Bystander Risk * School % EAL 0.011 0.014 .454

Bystander Risk * School % SA 0.021 0.020 .307

Bystander Risk * School % SAP and ST

0.052 0.029 .074

Bystander Risk * School level achievement

0.006 0.012 .625

Bystander Risk * School level overall absence

0.081 0.115 .483

Bystander Risk * School exclusion rates 0.047 0.027 .089

INDIVIDUAL LEVEL (ICC = 91.6%) 0.287 0.012 <.001


FSM 0.084 0.043 .051

Attendance -0.001 0.003 <.001


Bully (vs. Not involved) 0.213 0.078 .006

Bystander (vs. Not involved) 0.236 0.099 .017

238

Secondary protective model Comparing between the full model (see section 6.4.4) and protective model

(table 6.20), there was a non- significant reduction in the -2*log likelihood value.

Full model = 2047.946– Protective model (2004.807) = 43.139, (χ² (45, n

=1628) = 43.139, p > .05). As the reduction in value from the full model to the

protective model was non-significant, this suggests the protective model should

not be favoured over the full model, as it is a poorer fitting model. It is perhaps

unsurprising that the protective model is no better than the full model in terms of

model fit, as of the additional 45 predictors comprising the interaction terms only

one reached statistical significance. As this model therefore contained many

non-significant terms it was expected to be a poorer fitting model (Field 2009).

Nonetheless, as the 2*log likelihood value did decrease albeit not significantly,

the significant interaction effects were observed. Only one significant interaction

effect was noted (see table 6.20) which shows school urbanicity * academic risk

reached statistical significance, (β0ij = -0.138, p =.038). The result suggests that

this predictor variable moderates the effects of risk on behaviour difficulties.

This result should be interpreted however, with considerable caution. As the

alpha level was set at .05, and with the addition of 45 extra predictors it would

be expected at least one result would emerge as a significant finding. On the

other hand as this protective model only had sufficient power to detect medium

effects (not small effect like in the other models) a number of type 2 errors could

be present, where a finding has been reported as non-significant when in fact

this effect does exist. In conclusion this result needs to be interpreted in the light

of these issues.

239


Do the significant interactions display evidence of protective-stabilising or protective-reactive effects?

Interaction graphs were computed in order to answer this question. These

graphs allow for an easy visualisation of the interaction effects, and show how

behaviour difficulties are affected by different levels of the moderator variable

for those at high and low risk. These graphs also display the direction of the

effect and whether high or low levels of the moderator variable confer the extra

advantage to those at high risk.

The graphs for significant moderating variables for both the primary and

secondary model are presented in figure 6.3-6.5. They show the protective

effects for school related variables on behaviour difficulties at different degrees

of risk. The ‘Y’ axis represents the dependent variable mean score (i.e.

behaviour difficulties). The ‘X’ axis represents the independent variable (i.e. the

risk variable), which was mean centred as is the procedure when looking for

interaction effects (Aiken & West 1991, Tabachnick & Fidell 2007). Two points

were created which represent high and low levels of the risk score that are +/-1

standard deviations above/below the mean for continuous risk variables, or

represented risk being present (high risk) or risk being absent (low risk) for the

categorical variables. The two lines on the graph represent the moderator

variable ‘Z’. This variable was also mean centred with the two lines representing

high levels of the variable, i.e. +1 standard deviation above the mean, and low

levels of the variable i.e. -1 standard deviation below the mean. For binary

moderators the two lines represent the two groups of the moderating variable

i.e. urban and rural for the variable school urbanicity.

Interaction graphs are only interpretable if the interaction term is statistically

significant; therefore graphs are only presented for the significant findings.

These graphs include 2 within the primary school model and 1 within the

secondary school model. They are also able to assess whether a ‘protective-

stabilising’ or a ‘protective-reactive’ effect has occurred on any of these

240

variables (Luthar et al. 2000a). See section 4.5.1 for more details of these

different effects.

a) Primary model: School level achievement

Figure 6.3: Protective factor interaction graphs for primary schools: School level achievement

Figure 6.3 displays an interaction graph between the independent risk variable

positive relationships and the moderator variable school level achievement. The

low risk end of the positive relationships variable indicates good levels of

positive relationships, whereas the high risk end reflects poor positive

relationships.

For school level achievement a protective-stabilising model best describes the

data. Individuals who experience low degrees of risk (i.e. have good positive

relationships) display fewer behaviour difficulties when attending schools with

higher levels of achievement, than those attending schools with lower levels of

achievement. Individuals who experience high degrees of risk (i.e. have poor

positive relationships), also display better behaviour when attending schools

with higher levels of achievement than schools with low levels of achievement,

however, the difference between the schools types is considerably greater in

the high risk context.

0

0.2

0.4

0.6

0.8

1

1.2

Low Risk High Risk

Beh

avio

ur D

iffic

ultie

s

Positive Relationships

Moderator: School level achievement

Low school level achievement

High school level achievement

241

Higher school level achievement is therefore acting as a protective factor,

reducing behaviour difficulties particularly for those experiencing high levels of

risk. A protective-stabilising effect has occurred here, as behaviour difficulties

remain more or less stable when the protective factor is present (i.e. high levels

of school level achievement), however, when it is absent (i.e. low levels of

school level achievement), behaviour difficulties increase more dramatically.

A final analysis was undertaken here to test the significance of the simple

slopes, to acknowledge whether each of lines on the interaction graph differ

significantly from 0 i.e. from the horizontal. These tests revealed that for both

the lines representing high school level achievement (b = 0.190, p = .013) and

low levels of achievement (b = 0.199, p = .009) were significantly different from

0. In order to show a true protective stabilising effect the moderator variable (in

this case high school level achievement) should remain stable despite

increasing risk and therefore not differ significantly from 0. However, as this

slope was less steep than low level achievement, it is offering some protective

function.

b) Primary model: School percentage of SA

Figure 6.4: Protective factor interaction graphs for primary schools: School percentage of children with SA

0

0.2

0.4

0.6

0.8

1

1.2

Low Risk High Risk

Beh

avio

ur D

iffic

ultie

s


Moderator: School % SA

Low school % SA High school % SA

242

Figure 6.4 displays an interaction graph between the independent risk variable

positive relationships and the moderator variable school % SA. The low risk end

of the positive relationships variable indicates good levels of positive

relationships, whereas the high risk end reflects poor positive relationships.

For school % SA a protective-stabilising model best describes the data. There is

little difference between individuals who experience low degrees of risk (i.e.

have good positive relationships) in attending either a school with high or low

levels of children on the SEN register at SA. However, individuals who

experience high degrees of risk (i.e. have poor positive relationships), display

better behaviour when attending schools with higher levels of children on the

SEN register at SA.

Higher school % SA is therefore acting as a protective factor, reducing

behaviour difficulties particularly for those experiencing high levels of risk. A

protective-stabilising effect has occurred here, as behaviour difficulties remain

more or less stable when the protective factor is present (i.e. high levels of

school % SA), however, when it is absent (i.e. low levels of school % SA),

behaviour difficulties increase more dramatically.



significantly from 0 i.e. from the horizontal. These tests revealed that for both

high school percentage of SA (b = 0.186, p = .015) and low school percentage

of SA (b = 0.203, p = .008) the lines differed significantly from 0. In order to

show a true protective stabilising effect the moderator variable (in this case high

school percentage of SA) should remain stable despite increasing risk and

therefore not differ significantly from 0. However, as this slope was less steep

than low school percentage of SA, it is offering some protective function,

although not a full protective stabilising effect

243

c) Secondary model: School urbanicity

Figure 6.5: Protective factor interaction graph for secondary schools: School urbanicity

Figure 6.5 displays an interaction graph between the risk variable academic

achievement and the moderator variable school urbanicity. The low risk end of

the academic achievement variable indicates good levels of academic

achievement, whereas the high risk end reflects poor academic achievement.

For school urbanicity a protective-stabilising model best describes the data.

There is little difference between individuals who experience low degrees of risk

(i.e. have good academic achievement) in attending either a school in a rural or

urban area. However, in individuals who experience high degrees of risk (i.e.

have poor academic achievement); they display better behaviour when

attending schools in urban areas.

Urban schools therefore act as a protective factor, reducing behaviour

difficulties particularly for those experiencing high levels of risk (in terms of poor

academic achievement). A protective-stabilising effect has occurred here, as

behaviour difficulties remain stable when the protective factor is present (i.e.

urban schools), despite increasing risk. However, when it is absent (i.e. rural

0

0.2

0.4

0.6

0.8

1

1.2

Low Risk High Risk

Beh

avio

ur D

iffic

ultie

s

Academic Achievement

Moderator: School Urbanicity

Urban Rural

244

schools), behaviour difficulties increase more dramatically when moving from

low to high risk.



significantly from 0 i.e. from the horizontal. These tests revealed that for rural

schools (b = 0.165, p = .007) the line significantly differed from 0. This effect

was not observed for urban schools (b = -0.027, p = .208), where the line did

not differ significantly from 0. This provides further evidence that in urban

schools, despite increasing risk the level of behaviour difficulties remains stable

and does not increase showing evidence of a full protective stabilising effect.

245


• Significant interaction effects for the primary school model were found

between positive relationship risk and school level achievement; and

positive relationship risk and school level percentage of SA.

o Higher school level achievement reduces behaviour difficulties for both

individuals experiencing high and low levels of risk. The effect of this

variable in reducing behaviour difficulties has an even more profound

and positive effect for those at high compared with low risk.

o This is evidence of a protective-stabilising effect that higher levels of

school level achievement moderates the influence poor positive

relationships has upon behaviour difficulties.

o Higher levels of school percentage of SA reduces behaviour difficulties

for those individuals experiencing high degrees of risk (in terms of poor

positive relationships). However, percentage of SA makes no difference

to an individual’s behaviour difficulties when experiencing low degrees

of risk (i.e. good positive relationships).

o This is evidence of a protective-stabilising effect, that higher levels of

school percentage of SA moderate the influence poor positive

relationships has upon behaviour difficulties.

• Significant interaction effects for the secondary school model where found

between academic risk and school urbanicity.

o Attending urban schools reduces behaviour difficulties for those

individuals experiencing high degrees of risk (in terms of poor academic

achievement), but type of school attended (i.e. urban or rural) makes no

difference to an individual’s behaviour difficulties when experiencing low

degrees of risk (i.e. good academic achievement).

o This is evidence of a protective-stabilising effect, that urban schools

moderate the influence poor academic achievement has upon


246

7. Discussion

7.1 Introduction to the Chapter

This chapter is subdivided into five main sections. The first section, 7.2 provides

a brief summary of the results, broken down by the three main research

questions. The second section, 7.3 then discusses the findings of the three

broad research questions in relation to the existing literature (previously

addressed in chapters 1 to 4), noting whether the results converge or diverge

from these studies. The third section, 7.4 acknowledges the limitations of the

study, from both a methodological and conceptual standpoint. The final section

7.5 highlights possibilities and recommendations for future research. Within the

fifth section, 7.6 summary statements of the chapter are provided.

7.2 Summary of results

7.2.1 Research question 1

The primary model revealed there was a statistically significant school and

individual effect with 15.4% of the variance in behaviour difficulties attributable

to the school level, and 84.6% to the individual level. The secondary model also

revealed a significant school and individual effect with 13.0% of the variance in

behaviour difficulties attributable to the school level and 87.0% to the individual

level.

Within primary schools the significant risk factors at the individual level were:

being in year 5, being born in the autumn, being male, being eligible for FSM,

poor positive relationships and being a bully. The only significant risk factor at

the school level was attending a school with low levels of academic

247

achievement. The significant promotive factors at the individual level were

therefore: being in year 1, being born in the summer months, being female, not

being eligible for FSM, good positive relationships and not involved in bullying.

At a school level, attending a school with a high level achievement was a

promotive factor.

Within secondary schools the significant risk factors at the individual level

included: being in year 7, being male, being eligible for FSM, low levels of

academic achievement, low levels of attendance and being a bully or a

bystander to bullying. The only significant risk factor at the school level was

attending a large school (in terms of pupils on role). The significant promotive

factors at the individual level were therefore: being in year 10, being female, not

being eligible for FSM, high levels of academic achievement, high levels of

attendance, and not involved in bullying incidences. At a school level, attending

a smaller school was a promotive factor.

The percentage of variance in behaviour difficulties that could be explained

when all predictors were added was 14.3% in primary schools (25.6% at the

school level and 12.2% at the individual level). In the secondary school data set,

the percentage of variance which could be explained when all predictors were

added was 14.2% (32% at the school level and 11.6% at the individual level).


The results revealed for both primary and secondary models that the cumulative

risk score was a significant predictor of behaviour difficulties. Specifically when

higher numbers of risk factors are present within an individual’s background,

regardless of their exact nature, this resulted in more severe behaviour

difficulties displayed.

The functional form of the relationship between cumulative risk and behaviour

difficulties was tested by adding in a quadratic term to see if this or a linear term

best described the data. The quadratic term reached statistical significance for

248

both the primary and secondary models. This suggests that with each increment

of risk there is an acceleration of behaviour difficulties displayed. Risks

therefore potentiate one another, exerting a greater influence on behaviour

difficulties collectively than the sum of their individual parts.

Finally, results revealed that type of risk was more important than number of

risk in explaining behaviour difficulties. For both the primary and secondary

school models the independent additive model (when risks were put

independently in to the model) was a significantly better fit of the data, and

accounted for more variance in explaining behaviour difficulties, than the

cumulative model (where risks were either present or absent and then added

together to form a single risk score).


Within the primary school model there were two significant protective variables.

Firstly, the variable school level achievement showed that attending schools

with higher academic achievement was beneficial to all children, although

especially for those considered at high risk. Higher school level achievement is

a key protective factor for behaviour difficulties in children with SEND who are

at risk in terms of having poor positive relationships. This factor showed

evidence of a protective-stabilising effect, that attending schools with higher

levels of academic achievement moderates the relationship between risk and


Secondly, the variable school percentage of SA showed that attending schools

with higher numbers of children at school action on the SEND register was a

key protective factor. High levels of this variable moderate the influence that risk

(in terms of poor positive relationships) has upon behaviour difficulties. In low

risk contexts (i.e. having good positive relationships) attending schools with

either high or low percentages of children at SA makes no difference to

behaviour difficulties displayed. This variable showed evidence of a protective-

249

stabilising effect, that attending schools with higher percentage of children with

SA moderates the relationship between risk and behaviour difficulties.

Within the secondary school model one significant protective factor emerged.

The variable school urbanicity showed that attending schools in urban areas

acted as a protective factor for those children considered at high risk (in terms

of poor academic achievement), but made no difference for those considered at

low risk. This factor showed evidence of a full protective-stabilising effect, with

attending urban schools being particularly beneficial in reducing behaviour

difficulties for those at high levels of risk.

7.3 Discussion of results in relation to previous literature.

7.3.1 Research question 1a:

What proportions of variance in behaviour difficulties are attributable to the school and individual levels?

Empty multi-level models with behaviour difficulties at follow-up as the outcome

(controlling for previous behaviour difficulties at baseline), were computed for

each of the primary and secondary data sets. These findings showed that for

the primary and secondary schools, both the individual level and the school

level significantly contributed to variance in behaviour difficulties. Nonetheless,

the individual level exerted a far greater influence than the school level on


This is not a surprising finding, and is consistent with the vast majority of studies

within this area that suggest the individual level accounts for more variance in a

number of childhood outcomes compared with the school level (Reis, et al.

2007, Maes & Lievens 2003, Gottfredson & DiPietro 2011, Aveyard, et al.

2004). It has been suggested that although schools play some influence in

accounting for problem behaviour, ecological levels that are more proximal to

250

the individual have more profound and direct effects on pupils. Contextual

influences such as the school environment often exert their influence indirectly

being mediated through another variable therefore; their influence is less

pronounced (Mooij et al. 1998).

The ICC at the school level in primary schools was 15.6% and in secondary

schools 13.0%, this statistic shows the amount of variance in behaviour

difficulties attributable to the school compared with individual level. These

figures compare with others found previously, i.e. Gottfredson (2001)

acknowledged that the variance attributable to the school level for individual

behaviour is around 8-15%, and Khoury-Kassabri et al. (2004) who found

schools account for about 9-15% of variance when the outcome was student

victimization.

The figures found within the present study are however, slightly higher than

some previous studies e.g. 5-10% suggested by Felson et al. (1994) in terms of

any childhood outcome, 3% in Kohen et al. (2009) investigating physical

aggression in young children, <8% in Sellstrom & Bremberg’s (2006) review of

multi-level studies, investigating different behaviour outcomes, 2%

acknowledged by Reis et al. (2007) when the outcome is aggressive behaviour,

6.3% by Payne (2008) when using a measure of delinquency, and finally 2-3%

by Anderson et al. (2010) when the outcome was mental health difficulties.

Nonetheless, another study included within the review by Sellstrom & Bremberg

(2006) suggested that a measure of problem behaviour (specifically weapon

carrying) had an ICC at the school level as high as 25%.

Within the previous studies acknowledged above, large differences exist in the

ICC reported at the school level when the outcome concerns behavioural issues

in children and adolescence. This is perhaps unsurprising as often different

studies have used distinct outcomes, chosen with specific populations, across

different types of schools with various countries around the world. Furthermore,

there has been differences in how the outcomes were measured i.e. teacher

verses self-report, and whether previous outcomes were controlled for. All these

issues will affect the actual size of the ICC reported. The slightly higher ICC

251

found within the present study may reflect the population under investigation. As

children with SEND obtain additional support from within their school, they could

build stronger relationships with staff, and therefore become more prone to the

effects of school level influences on their behaviour, than children not receiving

the extra support. However, the ICC within the present study is only a little

higher than the majority of the previous literature, and in fact still within a similar

range. What can be concluded here is that within the present study and the

previous literature the school ecological level remains an important influence on

problem behaviour, although to a lesser degree than at the individual level.

7.3.2 Research question 1b:

Which school and individual level predictors explain a statistically significant proportion of variance in behaviour difficulties?

Full multi-level models were computed by imputing all predictor variables in to

the analyses. Variables that showed a statistically significant positive

relationship with increased levels of behaviour difficulties were termed risk

factors and when that relationship was negative, resulting in decreases in

behaviour difficulties, the term promotive factors was used (see section 1.4 for a

definition of these terms).

a) Consistent findings across primary and secondary models

Gender A salient finding to emerge from the present study was that participants’ gender

was a significant predictor of behaviour difficulties, and remained as such even

after controlling for previous behaviour difficulties displayed. Specifically it was

being male which led to increases in these problems, with reductions related to

being female. Being male is therefore a risk factor for displaying behaviour

difficulties (and being female a promotive factor). These findings are consistent

with other studies that have found such differences (e.g. Brown & Schoon 2008,

Green et al. 2005, & Loeber et al. 2000). It is therefore not only within the

252

general population, but also within the SEND population where being male is a

risk factor for behaviour difficulties.

As gender was a significant predictor across both primary and secondary school

models, this implicates it as a particularly important finding –influencing

behaviour difficulties across different developmental stages. This evidence, is

not that surprising, as Baillargeon, et al. (2007) found such gender differences

in the display of aggressive behaviours to be evident at around 1 and a half

years of age - suggesting they are likely to be even more evident by the time a

child reaches school age. The reasons why being male is a risk factor for

behaviour difficulties lies beyond the scope of the present study, but previous

researchers have suggested that males may be exposed to, or vulnerable to

additional risk factors, which are mediated through their gender (Storvoll &

Wichman 2002). These could include hormones (Book et al. 2001), parenting

practices (Crick & Zahn-Waxler 2003), or the fact that girls often display more

prosocial behaviour which could protect against the display of behaviour

difficulties (Messer et al. 2006). Nonetheless, the present study has shown that

for a group of children with SEND behaviour difficulties will be predicted by

gender, with being male a significant risk factor.

Free School Meals eligibility FSM eligibility was established within the present study as a significant predictor

of behaviour difficulties across both primary and secondary school models.

Those children eligible and claiming FSM are at an increased risk of displaying

more severe behaviour difficulties. FSM is often taken as a proxy for social

economic status (Hobbs & Vignoles 2007), with children eligible for and

claiming FSM being of lower social-economic status, compared with those not

eligible. The findings within the present study therefore provide further support

for the consistently held view that children of lower SES are at an increased risk

of displaying problem behaviours, (e.g. Brooks-Gunn & Duncan 1997, Klaff et

al. 2001, and Huaqing-Qi & Kaiser 2003). More recent research, including the

large cohort studies by Brown & Schoon (2008) and Green et al. (2005) can

also be supported, as these studies found that children living in poverty or

253

residing with parents on lower incomes were at increased risks of displaying

problem behaviours compared to those children from more affluent

backgrounds. Finally, support is offered for a non-correlational study using a

quasi-experimental design, which showed a positive relationship between lower

SES and behaviour difficulties (D’Onofrio, et al. 2009). The consistency

displayed between the results of the present study and those from previous

literature, goes to show that children of lower socio-economic status are at an

increased risk of displaying behaviour difficulties and this is true across a wide

range of populations including children with SEND.

The present study was not able to provide evidence why this relationship

occurred; nonetheless, it has been argued that family income can have

substantial effects on children particularly during the school years (Brooks-Gunn

& Duncan 1997). Those children from poorer backgrounds are more likely to

experience more severe familial stress, live within chaotic and unstable

households, experience poorer parenting behaviours, and have a lack of

cognitive stimulation. These factors will accumulate and their multiple influences

may exacerbate coping resources and result in the display of behaviour

difficulties (Evans 2004).

Bullying Within both the primary and secondary models, being a bully, (as rated by

teachers) was a significant risk factor for behaviour difficulties. Children who

were rated as such were considered the instigator or perpetrator of bullying

incidents and were noted as the children with the most severe behaviour

difficulties. This evidence is consistent with Wolke et al. (2000), and Hampel et

al. (2009) who have also showed a direct relationship between being a bully

and the display of more severe behaviour difficulties. A study by Kim et al.

(2006) (who had a similar methodology with the present study), found children

rated as being a bully at baseline were significantly more likely to display more

aggressive behaviour ten months later. Being a bully is therefore a precursor to

anti-social behaviour. The consistency in evidence displayed suggests this is a

particularly pertinent finding, and extends across distinct populations, with

254

children with SEND no more or any less affected by this variable than typically

developing children. Indeed the evidence that being a bully is a predictor of

more severe behaviour difficulties is perhaps not a surprising one; bullying

behaviour is often comprised of aggressive acts, which at least to some degree

are also salient features of behaviour difficulties measurements.

Within the secondary model, although not the primary model, being rated by

teachers as a bystander to bullying was also a significant risk factor. Bystanders

of bullying are conceptualised as being present in bullying incidents although

not partaking in them directly. It is unclear why being a bystander to bullying is a

risk factor for behaviour difficulties, and with a lack of research specifically

within this area, the issue remains particularly cloudy. Nonetheless, it could be

surmised here that bystanders are friends of bullies in which case they are

present at bullying incidents although not directly involved themselves. They

may be influenced by the negative behaviour they witness and display similar

negative behaviour within other contexts. Bystanders could therefore be rated

as having more severe behaviour difficulties. However, if this was the case why

is being a bystander only a significant risk factor in secondary schools and not

in primary schools? One argument could be that primary school teachers are

more reliable raters of a child’s role in bullying incidents, as they teach the same

pupils throughout the day. They may therefore know these pupils better and can

more accurately rate their child’s role in bullying incidents.

In contrast to a number of previous studies being a victim of bullying did not

emerge as a significant risk factor. This evidence disputes the majority of

previous research in suggesting that those individuals rated as being victims of

bullying have more significant behaviour difficulties, (Gini 2008, Hampel et al.

2009, and Areseneault et al. 2010). The reasons for these differences are not

abundantly clear, although they may well be surrounded within the

measurement issues of bullying roles. Within the present study teachers were

the raters of a pupils role in bullying incidents, as bullying may be a subjective

experience it is difficult to know when someone has been bullied or whether

they were doing the bullying or indeed whether they were a bully-victim or

bystander. As this variable only involved teachers ticking one box for a pupil’s

255

typical role in bullying incidents care should be taken in the interpretation of this

finding. Finally and perhaps unsurprisingly being rated as ‘not involved’ in

bullying incidents emerged as a promotive factor for children in both primary

and secondary schools, and these children experienced fewer behaviour

difficulties

Year Group Within both the primary and secondary models the variable year group (taken

as a measure of age) was a significant predictor of behaviour difficulties.

Previous studies have also showed age to be related to behaviour difficulties

(Green et al. 2005, Maughan et al. 2004). The present study adds to this

literature in suggesting year group or age is an important predictor of behaviour

difficulties in children with SEND. Specifically, within the primary school model

those in year 5 (compared with year 1) were at risk and in the secondary model

those in year 7 (compared with year 10) were at risk. This is an interesting

finding as within primary schools the older children were worse off, whereas in

secondary schools it is the younger children that are more at risk of displaying


The finding in relation to the secondary school model is consistent with

Bongers, et al. (2004) who suggest that there are different age effects on

behaviour difficulties depending on the exact nature of the difficulties displayed.

As children become older they may be less likely to engage in physically

aggressive acts but more likely to display non-aggressive acts such as truancy

and drug abuse. In terms of the behaviour difficulties measured within the

present study from the WOST, 4 of the 6 items were related to aggressive

behaviours and 2 were related to non-aggressive acts. This bias in favour of

more aggressive behaviour difficulties could be used to explain why in the

secondary school model at least, the younger children were considered more at

risk as the majority of items measured behaviours which they were likely to

score worse on compared with the older children. Nonetheless, this would not

explain the finding within primary schools where the older children were most at

risk. Indeed the results of the present study are inconsistent with

256

Papatheodorou (2005) who states younger children may display more

behaviour difficulties, as they will not yet have learnt socially acceptable

behaviours, although this is more relevant to preschool aged children.

In the context of the present study the difference between year 1 and year 5

scores could reflect not the disparity between aggressive and non-aggressive

acts within the measurement tool, but how well the teachers know the children

within their class. There could be an argument that teachers are more familiar

with their year 5 classes having got to know them throughout their school career

compared with teachers of year 1 children who are teaching children fairly new

to the school. Children who have been in the school longer may become

‘known’ with teachers rating their behaviour upon negative reputations and

stereotypes rather than actual behaviours.

A further alternative view is that the decrease of aggressive acts and increase

of non-aggressive acts with increasing age may only come about at a certain

threshold. For example both these types of behaviours may augment with age

until a threshold time (i.e. at the transition time to secondary schools), where

aggressive acts then begin to decrease and non-aggressive acts increase

further. Indeed Lahey et al. (2000) in a study of 9-17 year olds suggested that it

is the middle years of this age range where the problems are most pronounced.

This is consistent with the present study that found those children in years 5

and 7 had more problems with their behaviour than children in years 1 and 10.

In drawing this section to a conclusion it is essential that age effects on

behaviour difficulties are considered from a developmental perspective,

acknowledging so called early onset (within childhood) and late onset (within

adolescence), behaviour difficulties and taking into account that the causes of

both maybe different (Lahey & Waldman 2003).

257

b) Unique predictors within primary schools

Season of birth Season of birth (broken down into the 4 distinct seasons see table 5.2) was an

established risk factor for behaviour difficulties only within the primary school

model. Those born in the autumn months (September through November) had

significantly worse behaviour difficulties than those born within the summer

months (June through August). The academic year within the UK begins in

September; therefore those born in the autumn months are the oldest in their

year, with summer born children the youngest. It is the oldest children within the

school year in primary schools who are at risk and display the most severe


This evidence is in fact contrary to a number of previous studies who have

showed the youngest children within any school year display the most severe

behaviour difficulties (Goodman, et al. 2003, Menet, et al. 2000, Polizzi, and

Martin & Dombrowski 2007). These studies however, did not explicitly use a

SEND population which could account for the difference in findings. The

disparity in findings could reflect issues with maturity levels and that younger

children within the school year are less socially and physically skilled and may

suffer social disadvantages in terms of peer relationships (Martin et al. 2004).

This is not necessarily an explanation as to why younger children display

behaviour difficulties, (as they have not yet learned appropriate behaviours) but

that the younger children’s lack of maturity may cause them to become victims

of aggressive acts as they are seen as the most physically weak children. The

autumn born children may be the perpetrators of such incidents being physically

stronger and often nearly a whole year older than summer born children.

This may be a credible explanation within primary schools where relative age in

a school year has a more pronounced effect, particularly within year 1 where

nearly a year’s difference in age is a substantial proportion of a child’s life.

These differences however, become less pronounced as children get older

(Menet et al. 2000), and differences in maturity levels become less extreme

within a single school year. These null findings within the secondary model

258

support other researchers who have failed to find a relative age within the year

effect with secondary age children (Lien et al. 2005).

Positive relationships Positive relationships became a significant predictor of behaviour difficulties

within the primary school model. Those with lower levels of positive

relationships were significantly more likely to display more severe behaviour

difficulties than those with higher levels. Within the secondary model the

relationship between poorer relationships and behaviour difficulties approached

significance, although this value was not p <.05 therefore cannot be

conceptualised as a risk factor. The present study therefore provides further

support for the consistently held view that children with more positive peer and

teacher relationships will experience fewer behaviour difficulties (Ladd &

Burgess 1999, Hamre & Pianta 2001, Criss, et al. 2002, Dodge, et al. 2003,

Buyse, et al. 2008, Baker, et al. 2008, Silver, et al. 2010). The focus within the

present study on children with SEND shows that for this group of children, like

all children, positive relationships are important in reducing behaviour

difficulties. Indeed Wiener (2004) has reported that for children with learning

difficulties, risks for behaviour difficulties can be reduced with positive social

relationships.

A number of explanations have been offered as to why more positive

relationships with peers and teachers have an effect on behaviour difficulties.

One idea is that the influence of positive relationships on behaviour difficulties is

mediated through attachment relationships. Those children with a more secure

attachment with their primary caregiver will be able to form other secure

attachments with teachers and peers which lead to the increasing likelihood of

displaying positive behaviours, and reducing the likelihood of negative

behaviours (Hamre & Pianata 2001). Nonetheless, whether lower levels of

positive relationships result in increases in behaviour difficulties or whether

behaviour difficulties lead to a decrease in positive relationships is a contentious

issue. Doumen et al. (2009) found that aggressive behaviour at the beginning of

the school year resulted in an increase in child-teacher conflict that then lead to

259

further increases in aggressive behaviour at the end of the school year. This

confusion in the literature would benefit from further longitudinal evidence to

suggest the time order in effects. Possibly there exists a bidirectional

relationship with both concepts (i.e. positive relationships and behaviour

difficulties) impinging on one another.

Finally, it is interesting to note that the influence of positive relationships on

behaviour difficulties was more important within primary than in secondary

schools (indeed failing to be a significant predictor in the secondary school

model). There is a lack of evidence in the previous literature to account for why

differences across age groups could affect how relationships influence

behaviour difficulties. One study has shown that younger children are more

affected by negative peer influences than older children, which may then be

manifested in the display of behaviour difficulties (Laird, et al. (2001).

School level achievement A particularly strong and significant predictor of behaviour difficulties within the

primary model is school level achievement. Lower levels of school achievement

are a risk factor for increases in behaviour difficulties within primary schools.

Although some studies have failed to find that school level achievement is

related to behaviour difficulties (Wilson 2004), the present study is consistent

with a number of previous studies that have showed higher school level

achievement to be related to fewer problem behaviours within schools (Rutter et

al. 1979, Mooij, 1998, Barnes et al. 2006, Bisset et al. 2007). Within primary

schools, pupils (in the same year) are usually taught all together in the same

class and not split into sets of ability. Being in a mixed ability class where the

overall standard is relatively high could result in lower achieving pupils (such as

children with SEND) gaining by having peers of higher ability to help with their

academic difficulties. This could lead away from frustration in their work and

ultimately the display of problem behaviours, when teachers are unavailable to

support them.

260

A concerning finding, however, was that no such significant relationship within

the secondary model was established. In fact although the coefficient

approached significance, this effect was in the opposite direction to primary

schools, suggesting that higher levels of school level academic achievement

would be associated with more severe behaviour difficulties. Felson et al.

(1994) provide a hypothesis for this effect arguing that attending a school with

high achievement amongst its pupils is not always an advantage and could

have some detrimental effects upon student behaviour. This study showed that

when the school culture promoted high academic achievement a negative

influence upon student behaviour was observed. The authors suggested that

those students who are not achieving at a high level and are surrounded by

others who are, may suffer in comparison, seeing themselves as failing. The

sense of frustration they feel is then manifested in the displayed of problem

behaviours. This may explain the situation within the secondary model; as

secondary school pupils, will be aware of their performance in comparison to

others at the school. If overall academic achievement is low, pupils with SEND

(who are often behind their peers in academic achievements) will not consider

themselves as behind when they are in higher achieving schools. Ultimately this

may reduce frustration and the display of problem behaviours.

c) Unique predictors within secondary schools

School size Within the secondary model, although not in the primary model, school size was

a significant predictor of behaviour difficulties. The direction of effect was that

larger secondary schools are considered a risk factor for behaviour difficulties

with those attending smaller schools better off. The previous literature

surrounding the effect of this variable on behaviour difficulties is considerably

mixed with some evidence arguing, as in the present study, that larger schools

are a risk factor for behaviour difficulties and violence (Stewart 2003, George &

Thomas 2000), whereas others maintain no effect of school size on student

victimisation, aggression, or smoking and drinking behaviours (Khoury-Kassabri

et al. 2004, Wilson 2004, Maes & Lievens 2003). As these studies all focused

261

on secondary age school students, their findings only relate to this school type.

Further research is needed specifically on the effect of school size on primary

age students.

There are a number of reasons which can be proposed as to why in the present

study large school size was related to the display of behaviour difficulties. This

could have emerged as a significant finding as being within a larger school may

facilitate a degree of anonymity, where students feel less valuable and involved

within their school. Children with SEND may be particularly vulnerable to these

influences, and these feelings could manifest themselves in the display of

behaviour difficulties, with students rivalling for teacher attention. Within large

schools there is also the opportunity to meet and interact with a wider range of

people from different backgrounds. This can lead to a higher likelihood of

meeting and being influenced by negative peers which could have an influence

on individual problem behaviour.

Despite these ideas, the issue remains why the effects were not observed

within primary schools. It could be argued the variability in size could be an

important explanation. The difference in size between the largest and smallest

secondary schools is considerably greater than the difference between largest

and smallest primary schools. The lack of variability in size within primary

schools could be a reason why no effect was noted. Furthermore, within smaller

schools there maybe the opportunity for students to develop better relationships

with their peers and their teachers and these more positive relationships could

counteract some of the potential risks associated with larger schools

(Gottfredson & DiPietro 2011).

Attendance Within the secondary model, although not the primary model, attendance was a

significant predictor of behaviour difficulties. Those pupils who had poorer

attendance levels in secondary schools were at risk of displaying more severe

behaviour difficulties. This evidence is consistent with others who have found

similar negative effects on behaviour from greater amounts of school

262

unauthorised absences (Malcolm, et al. 2003, Miller & Plant 1999). Numerous

studies have presented evidence that those who have poorer attendance are

also more likely to engage in other problem behaviours such as drug abuse,

smoking and underage drinking (Chou, et al. 2006, Henry 2007, Henry &

Huizinga 2007). Although these behaviours are not synonymous with behaviour

difficulties in schools they are nonetheless problematic behaviours. The

considerable consistency in evidence present here, suggests that poor

attendance is a risk factor that extends beyond children with SEND, and is a

salient issue for all children.

Despite the evidence appearing in support of this relationship it could be

surmised that poor attendance is a consequence of behaviour difficulties and

not a cause. Those children, who have behaviour difficulties in schools to begin

with, attract negative attention from their teachers, leading to an overall

negative experience of school. As they begin to feel less connected they may

be more inclined to not attend school. Although a feasible possibility, the

present study was longitudinal in design and measured attendance levels

before the final outcome of behaviour difficulties was assessed, whilst

controlling for previous behaviour difficulties displayed. It is therefore suggested

here that poorer attendance is a risk factor for the display of behaviour

difficulties.

In secondary schools when children choose not to attend school, they are likely

to be unsupervised by adults and therefore have more opportunity to engage in

negative behaviours (McAra 2004). In such situations they are likely to have a

greater exposure to negative peer experiences and suffer a diminished positive

effect of school influences (Henry & Thronberry 2010). An issue that however

remains is why the relationship between poor attendance and behaviour

difficulties was not evident within the primary model. It could be that primary

school children are supervised to a greater extent than secondary children both

by their parents who often bring them and collect them directly from school, and

also by their teachers within school who see them in the same class throughout

the day. This may give primary school students considerably less opportunity to

abscond from school. Indeed primary school children with SEND may receive

263

additional supervision as they are often perceived as being more vulnerable

than their non-SEND peers.

Academic achievement Lower levels of academic achievement were established as a salient risk factor

for behaviour difficulties within the secondary but not the primary school

models. This finding is consistent with a number of other researchers within the

literature who have all shown that those individuals with lower levels of

academic achievement are also more likely to display behaviour difficulties

(Hinshaw 1992, Reid, et al. 2004, & McIntosh, et al. 2008). In relation to the

present study academic achievement was taken as a measure of an individual’s

point score in English, therefore showing that lower English abilities are a risk

factor for behaviour difficulties. This is consistent with Morgan et al. (2008) who

found those with reading problems (a sub domain of academic achievement in

English) at the beginning of school were more likely to display behaviour

difficulties 2 years later than those without these problems at the beginning of

school.

The major debate within this area concerns the direction of effect and whether

academic achievement predicts behaviour difficulties or if behaviour difficulties

predicts academic achievement. Breslau, et al. (2010) and McLoed & Kaiser

(2004) who both conducted longitudinal studies suggest it is behaviour

difficulties which predict academic achievement. This stance is nonetheless

disputed within the present study, which also implemented a longitudinal design

measuring the academic achievement as a predictor before the outcome

behaviour difficulties (and controlling for behaviour difficulties displayed at

baseline). As the predictor was measured before the outcome, lower levels of

academic achievement can be established as a risk factor rather than a

correlate of behaviour difficulties. The differences in these results could reflect

the population under investigation. The present study in utilising a SEND

population provides support for the argument presented by Morgan et al. (2008)

that the cause of behaviour difficulties is poor academic achievement, and when

children become frustrated with increased difficulty of a task they start to ‘act

264

out’ displaying behaviour difficulties. This argument particularly resonates

concerning children with SEND, as they are more likely to struggle academically

that their non-SEND peers. An alternative argument is that behaviour difficulties

displayed in children and adolescents have negative effects on their teachers

and this ultimately impinges on their learning (Reid et al. 2004, Breslau, et al.

2010).

A surprising finding however is why in the primary school model this effect failed

to reach statistical significance. There could be an argument here when children

become aware they are academically behind their peers this leads to a sense of

frustration which manifests itself in behaviour difficulties. Secondary school

pupils will have better awareness of their relative performance in comparison to

their peers, which could explain why this variable was related to behaviour

difficulties in the secondary schools. In primary schools however, pupils who are

further behind academically may not fully comprehend the situation and

therefore this influence has a lesser effect upon their behaviour. This is a fairly

speculative argument and further research is warranted in this area.

d) Non-significant predictors

There are a number of variables within the study that despite having both

empirical and theoretical justifications for their inclusion, failed to emerged as

significant predictors of behaviour difficulties for either of the primary or

secondary models. Nonetheless, they are discussed here and reasons are

offered as to why they may have remained non-significant predictors.

Ethnicity Ethnicity, failed to become a significant predictor of behaviour difficulties across

either the primary or secondary model. This evidence is contrary to other

researchers in the field, who argue that such differences do exist (e.g. Zwirs, et

al. 2011). Within the UK context Brown and Schoon’s (2008) noted key ethnic

differences in the prevalence of behaviour difficulties. Specifically, Black African

children had the least behaviour problems, followed by White children, with

265

Pakistani, Bangladeshi and Black Caribbean children having considerably

higher levels. The present study is unable to support these findings as found no

such ethnic differences. Nonetheless, it was not possible to break the ethnicities

down to the degree shown by Brown and Schoon, and just two labels White

British and Other were used within the present study. The advantages that

Black African children had within the Brown and Schoon study could have been

obscured in the present study as they were included within the other ethnic

groups that are more prone to such problems. Therefore, how ethnicity was

coded and categorised could be the influencing factor.

A further reason for the null finding within the present study has been argued by

Greenberger et al. (2001), who suggested that ethnic differences in terms of

problem behaviours may not emerge until well into adolescence. This may

explain why no significant effect was found within primary schools, although

fails to account for the non-significant finding in secondary schools, where

pupils are considered adolescents. Another explanation is that within a number

of the previous studies a parental report rather than teacher report has been

used to assess behaviour. Guttmannova et al. (2007) has suggested that

definitions and measurements of behaviour difficulties can vary across ethnic

groups, particularly in terms of what is deemed acceptable and developmentally

appropriate behaviour, with parents of different ethnicities rating children in

different ways. Within the present study teachers did the rating of a child’s

behaviour difficulties, as they are arguably in a better position to accurately

evaluate an individual’s behaviour, and assessing behaviour in relation to all

children in the class. The differences in findings reported across studies may

therefore be due to who was rating the behaviour of the child i.e. parent or

teacher. Finally, as no previous study has explicitly investigated risk factors for

a SEND population it could be reliably conceived that within this group of

children, ethnic differences as a risk factor for behaviour difficulties are non-

existent, and this is consistent with the stance of Dekovic, et al. (2004).

266

Level of SEND support Within both primary and secondary models the SEND level of support variable

(including school action, school action plus, and statement) failed to reach

significance as a predictor of behaviour difficulties. This is perhaps not a

surprising finding; although it is often assumed that children with higher levels of

support are more severely affected by their SEND than those at low levels, this

is not necessarily true. The level of support a child receives reflects the schools

ability to provide for the individual child’s needs, comparing across schools with

children at the same level of support could therefore be almost meaningless; as

considerable variation exists in a how schools define SEND (Ofsted 2005).

Nonetheless, in general, pupils move through the levels of support as their need

becomes greater (Frederickson & Cline 2009, DfES 2001a). It could be reliably

assumed that those with the most severe behaviour difficulties will be at the

statement level of support and those with the least problems at the school

action stage. Although despite this suggestion that children with a statement for

SEND are most likely to display behaviour difficulties, they may however, be

protected from displaying these problems due to their significant support

package in place.

Urbanicity The role of school location and whether it was situated with a rural or urban

setting had no influence on SEND pupil’s behaviour difficulties, either within the

primary or secondary models. The present study provides support therefore for

Wilcox-Rountree and Clayton (1999) and Swahn et al. (2010). These studies

also found no evidence of a school location effect, and that rural students are

not any better or worse off than their urban counterparts in terms of alcohol use

or aggression. Nevertheless, some previous literature has suggested there is a

school urbanicity effect on behaviour, with those in urban schools displaying

more severe behaviour difficulties (Stewart 2003, Larsson & Frisk, 1999, Hope

& Bierman 1998). These effects may occur as urban schools are more likely to

reflect the significantly higher rates of violence displayed within their wider

community (Warner, et al. 1999). Furthermore, urban school children may have

a greater exposure of anti-social and aggressive peers (Hope & Bierman 1998).

Other researchers however, have argued that children attending schools within

267

urban settings are actually at an advantage. Within these schools there is a

more of an abundance of resources available which leads to a more positive

effect on pupils (Sellstom & Bremberg 2006). There is considerable

inconsistency in this debate, but within the present study at least school location

has no effect on SEND children’s behaviour difficulties.

Percentage of FSM students within the school Another non-significant predictor of behaviour difficulties across both the

primary and secondary models was that the percentage of students within the

school eligible and claiming free school meals. It has been suggested that FSM

can be taken as a proxy for socio-economic status (Hobbs & Vignoles 2007)

with those eligible noted as being of lower SES. The majority of previous

evidence surrounding this variable argues that schools with higher average SES

influenced student behaviour in positive ways (Sellstom & Bremberg 2006) with

lower levels of SES within the school considered to be a risk factor for

behaviour difficulties (Felson, et al. 1994, George & Thomas 2000, and Wilcox

& Clayton 2001). Why this is the case however, is not abundantly clear, perhaps

the collective effect of many students who are of lower SES results in them

being at risk individually, resulting in a more aggressive culture within the

school, influencing even those of higher SES.

The present study is not able to offer support for these findings and instead has

returned a null result; suggesting at least for a group of children with SEND the

SES level of the school plays no influence on their behaviour displayed. It is

interesting why this disparity between the present study and previous literature

exists. The difference could reflect the measurement tool used, as FSM

eligibility may not be the most accurate measure of SES. It accounts for not only

those individuals who are eligible, but also those who are actively receiving it.

Concluding this issue the present study offers support for the view posed by

Stewart (2003) who reported a number of school level variables could account

for school problem behaviour although SES was not one of them.

268

Percentage of EAL students within the school Within both the primary and secondary models the percentage of pupils within

the school who have English as an Additional Language (EAL) was a non-

significant predictor of behaviour difficulties. These findings are consistent with

the previous work of Barnes et al. (2006) and Stewart (2003) who failed to find

any support for the idea that behaviour difficulties were related to the number of

children within the school with EAL or from different ethnicities. Nevertheless,

the debate within this area is particularly contentious as other researchers have

found that schools with more than 50% of students from ethnic minority

backgrounds (and therefore presumably higher numbers of EAL students) had

significantly worse behaviour problems than those with fewer students from

ethnic minority backgrounds (Ma et al. 2007).

It is not clear why differences within this variable would predict increases or

decreases in behaviour difficulties. Perhaps those who have EAL initially find it

difficult to understand appropriate behaviour within a classroom and culture they

are just beginning to learn. A greater number of children within the school with

EAL could have an additive effect on behaviour difficulties displayed, and

indeed this could have a negative influence on other children without EAL. This

explanation would not however, account for why older children who are more

likely to be well equipped with the English language and culture and do

understand behavioural expectations within the school. Contrary to the view that

more EAL students in a school reflects greater behaviour difficulties is the

evidence from Kohen et al. (2009). They have argued that actually schools with

a higher percentage of EAL students had significantly lower levels of aggressive

behaviour. There is considerable confusion with this variable, and its effects on

behaviour difficulties are unclear, further research is needed to elucidate this

debate. Nonetheless, the present study maintains that for children with SEND

the percentage of students with EAL makes little difference to their behaviour

displayed.

Percentage of SEND students within the school Within the present study two variables measured the number of students within

the school as having SEND. These were the percentage of students within the

269

school at the school action level of SEND and the percentage of students in the

school at the school action plus and statement level of SEND. Both these

variables were non-significant predictors of behaviour difficulties for both the

primary and the secondary models. This evidence is consistent with others in

the field who have suggested that schools with more SEND pupils within the

classroom have a limited effect on behaviour of other students (Sharpe et al.

1994, Tapasak & Walther-Thomas 1999).

However, previous research has found that children with SEND are more likely

to display behaviour difficulties than those without (Green et al. 2005).

Therefore a school with more SEND pupils is likely to have more significant

behaviour difficulties on the basis of a simple aggregation effect. As the number

of these pupils increase this could influence other students in a negative

manner. Indeed one study found that having more children on the SEND

register was related to school problem behaviour (Barnes et al. 2006). A

number of inclusion based studies have also shown that there are significant

negative effects on pupils when higher proportions of SEND pupils are included

within their classrooms (Daniel & King 1997, Brown 1982). The reason for such

negative influences is, that having more SEND pupils within a school can create

additional work for teachers, leaving considerably less time for other pupils, who

may then become frustrated or bored and act out problem behaviours in

response (Daniel & King 1997). The sample for the present study however, was

children with SEND, not children without SEND as in these studies described

above. Perhaps children with SEND do benefit to a greater extent when more

children also have SEND, as additional resources and interventions will be put

in place which they could benefit from either directly or indirectly. Nonetheless,

despite these suggestions the present study found a null result suggesting no

impact of school level SEND on pupil behaviour.

270

School level overall absence Within the present study differences between schools in their levels of absence

was a non-significant predictor of behaviour difficulties across both primary and

secondary schools. Therefore, whether the school level of overall absence was

high or low makes no difference to SEND pupils behaviour difficulties. This

evidence is contrary to the majority of the previous research which suggests

that children who attend schools with higher rates of absence are at a

heightened risk of displaying behaviour difficulties, than a child attending a

school with lower rates of absence (Bisset, et al. 2007, Maes & Lievens 2003).

Presumably higher rates of truancy or unauthorised absence could affect other

pupils in negative ways as they become annoyed with the disruption to the class

when the truanting pupils return and need additional help (Wilson et al. 2008).

No such effect was witnessed in the present study and the behaviour difficulties

displayed by children with SEND are not affected by the overall absence rate

within their school. The reasons why no such effect was observed within the

present study are hard to attribute. One possible explanation is that there could

be a lack of variability in overall absence scores across the schools in the

sample to detect a significant effect. This may be evidenced by the difference

between the present sample mean and figures found nationally. The difference

equates to a large effect size within the secondary model and a medium effect

size within the primary model. Mean scores within the present study were

considerably higher than national averages, suggesting that schools in the

present study have poorer attendance.

School exclusion rates The final non-significant predictor across both school models is school

exclusion rates. Within the present study this variable was taken as proxy for a

schools level of behaviour difficulties. Both within the primary school and

secondary school models non-significant findings emerged. This suggests that

exclusion rates have no influence on the behaviour difficulties displayed of

children with SEND. The present study is therefore unable to support the

majority of the previous literature which suggests that higher overall levels of

behaviour difficulties within a school will have a more negative influence on the

individual children within them (Mooij 1998). A considerable amount of this

271

research however, is classroom rather than school based suggesting that being

a member of a class with high levels of aggressive classroom for a longer

period of time has a strong influence on individual level aggression scores,

(Thomas & Bierman 2006, Mercer et al. 2009). It may be that peers within these

contexts become effective role models for other students, and reinforce the

display of behaviour difficulties (Barth, et al. 2004). Finally, the most directly

comparable piece of research to the present study comes from Theriot, et al.

(2010) who argued that school level exclusion is a significant predictor to

explain individual level exclusion; however, this can also not be supported.

Although the evidence from previous studies is overwhelmingly in support of the

idea that schools with higher levels of behaviour difficulties have negative

effects on individual behaviour, the present study failed to find such an effect.

This could be for a number of reasons, but perhaps the most likely is the lack of

variability within this predictor. Most schools did not report a single exclusion

over the school year which resulted in this variable having a fairly large floor

effect. Furthermore, the reporting of exclusions is surrounded in controversy

with many inaccuracies recorded, and schools failing to record them

appropriately for fear of negative consequences (Harris, Eden & Blair 2000). In

sum perhaps this variable is not the most appropriate to use when attempting to

gauge a measure of school level behaviour problems, and how it could

potentially influence the behaviour difficulties displayed of children with SEND.

7.3.3 Research question 1c:

Of the variance initially attributed to the individual and school levels, how much of this is explained by the predictors used in the study? An analysis was undertaken to calculate the amount of variance to be explained

(as evidenced by the empty model) that was explained when all the predictors

were added (i.e. the full model). The primary model including both school and

individual predictors can explain 14.3% of variance in behaviour difficulties

whereas the secondary model can account for 14.2%.

272

This is a fairly disappointing finding; a substantial proportion of variance

remains unexplained within the present study. However, as discussed in

chapter 2 there are numerous risk factors for behaviour difficulties that were not

measured within the present study; although still remain important in accounting

for behaviour difficulties. These may include hormonal influences (Book et al.

2001), certain personality traits (Viding, Frick, & Plomin, 2007), low self-esteem,

(Schonberg & Shaw 2007) association with delinquent peers (Laird et al. 2001),

and poor attachment relationships (Moss, et al. 2006). There are also further

school related variables which could be included e.g. school climate (Sellstom

and Bremberg 2006), school connectedness (Bond et al. 2007), and presence

of effective school policies (Evans-Whipp et al. (2004). If all these variables

were added the amount of variance that was explained within the present study

is likely to substantially increase.

Comparing between the primary and secondary models shows that total

variance explained was roughly comparable 14.3% within primary schools and

14.2% within secondary schools. Greater differences emerge when the

explained variance is broken down across the individual and school levels. The

secondary school model was able to account for 32.0% at the school level

compared with the primary school which could account for 25.6% at this level.

The difference between these models at the individual level is more comparable

with the primary model being able to explain 12.2% and secondary model

11.6%. At least two interesting findings are noteworthy here. Firstly, despite the

same variables being used within both primary and secondary models, more

variance could be explained at the school level for secondary compared with

primary schools. This suggest the school level variables selected are perhaps

more pertinent in accounting for behaviour difficulties in secondary school

students compared with primary school students. Secondly, despite more

variance to be explained at the individual level compared with school level (see

research question 1a) proportionally more of this variance was explained at the

school level compared with the individual level, for both the primary and

secondary models. This suggests that the school level predictors selected for

the study are more comprehensive of this level, than the individual predictors

273

are within the individual level in accounting for the variance in behaviour

difficulties. This is not surprising as the individual level is considerably more

complex incorporating many more potential risk factors for behaviour difficulties

than within a contextual level such as the school level.

Sellstom & Bremberg (2006) propose that multi-level studies which investigate

individual and school effects on pupil outcomes should report how much

variance was partitioned between these levels in the null and full models in

order to account for the amount of variance each level can explain. This type of

analysis has not been forthcoming in a number of investigations using a multi-

level approach which makes it difficult to compare the results within the present

study to those found previously in the literature. Nevertheless, although only a

relatively small proportion of variance was explained this is a salient finding, and

these factors do account for some of the variation in behaviour difficulties.

Clearly further study should extend this work adding in additional variables that

may be potentially important.

7.3.4 Research question 2a:

Is there a cumulative effect of contextual risk factors on behaviour difficulties, where higher numbers of risk factors present are associated with increased levels of behaviour difficulties? The findings in relation to the above question showed that for both primary and

secondary models, higher numbers of risk factors present within an individual’s

background, regardless of the specific make up of the risk, resulted in more

severe behaviour difficulties displayed. This is clear evidence of a cumulative

effect of risk factors on behaviour difficulties, and provides support for the

cumulative risk hypothesis, and previous literature which has found this same

effect (Trentacosta, et al. 2008, Raviv et al. 2010, Lima et al. 2010, McCrae &

Barth 2008, Appleyard et al. 2005, Atzaba-Poria et al. 2004, Gerard & Buelher

2004b, Kerr & Black 2000, Forehand et al. 1994). The present study is the first

piece of research to utilise a cumulative risk model to assess behaviour

274

difficulties in a population of children with SEND. In establishing such an effect,

it adds to the literature base, highlighting the salience of cumulative risk models

in accounting for behaviour difficulties across distinct populations.

Previous research has suggested that multiple risk factor models are better than

single risk factor models in accounting for the display of behaviour difficulties

(Dodge & Pettit 2003, Wachs 2000, Ackerman et al. 1999, Whipple, et al. 2010,

Evans, Kim, et al. 2007, Evans 2003, Forehand et al. 1998). This is because

problem behaviours result from numerous negative pathways. It is not any

specific risk factor that leads to these problems but the sheer number of risks

that build up within an individual’s background that overwhelm their coping

mechanisms and lead to problems in their behaviour (Flouri & Kallis 2007).

Within the present study the cumulative risk score was comprised of various

types of risks, with each individual having their unique set. The key issue was

not that any specific kind of risk lead to the problem behaviour but increasing

the number of risks in an individual’s background resulted in more significant


The present study argues that research should be conducted by investigating

risk factors in combination rather than in isolation. This is because risks are not

independent from one another, and cluster together within individuals (Flouri &

Kallis 2007). Risk factors need therefore to be acknowledged within the context

of all other possible risk (Kraemer et al. 2001, MacKenzie et al. 2011, Gutman

et al. 2003), and this will enable a better understanding of how these factors

work in combination to influence behaviour difficulties.

275


What is the nature of the relationship between exposure of cumulative risk and behaviour difficulties?

The functional form of cumulative risk on behaviour difficulties and whether this

relationship is linear or nonlinear remains a particularly disputed aspect with the

literature (Gerard & Buelher 2004a, Appleyard et al. 2005). Within the present

study research question 2b revealed that a quadratic pattern best described the

data. Therefore, as risks accumulate within individuals their combined effect

becomes increasingly detrimental.

This evidence is contrary to the linear pattern observed in a number of previous

studies (e.g. Raviv et al. 2010, Flouri & Kallis 2007, Appleyard et al. 2005,

Gerard & Buehler 1999, 2004a, 2004b, Dekovic 1999) where increments in risk

had a similar additive effect on problem behaviour observed. The present study

also found no evidence of a saturation model signifying a levelling off or plateau

effect, after the threshold is reached (Evans 2003, Morales & Guerra 2006).

Within such models after the threshold further risk causes no increased

negative effect on behaviour, but this was not witnessed within the present

study, where adding in each additional risk increases in behaviour difficulties

were reported.

The present study therefore provides support for the alternative view that a

nonlinear relationship exists between cumulative risk and behaviour difficulties.

Specifically, the quadratic term was significant suggesting that there is an

acceleration of behaviour difficulties as number of risks increase linearly. Risks

therefore potentiate one another with their combined effect being more severe

than the sum of their individual effects.

This means that as risk increases beyond a certain level behaviour difficulties

displayed become increasingly more severe (Raviv et al. 2010). Evidence of a

mass accumulation effect (Gerard & Buelher 2004a) was therefore observed

where the total effect of cumulative risk exerted an influence on behaviour

276

difficulties greater than the sum of its individual parts (Lima et al. 2010, Flouri &

Kallis 2007). The present study therefore provides support for a number of

others within the field who have found the same nonlinear relationship with

cumulative risk and various types of problem behaviours (Rutter 1979,

Bierderman et al. 1995, Forehand et al. 1998, Greenberg et al. 2001, Jones et

al. 2002).

What is particularly remarkable about a number of previous studies is their

apparent consistency in where the threshold occurred; despite being comprised

of various different types of risks, and outcomes. Rutter (1979) showed a

threshold after 4 risks are reached, Bierderman, et al. (1995) suggested it was

around 3, Forehand et al. (1998) noted that as the number of risks increased

from 3 to 4 there was a dramatic increase in the amount of problem behaviour,

Greenberg et al. (2001) showed that when 3 or more risk factors were present

there was a dramatic increase in problem behaviour, and Jones et al. (2002)

proposed that after 4 risks were encountered there would be significant

increases in problem behaviour.

The present study is in agreement that there is an acceleration of problem

behaviour at a critical level of risk although is not able to comment on exactly

where this effect occurred as the cumulative risk score incorporated fewer risks

than the studies mentioned above. Nonetheless, a similar nonlinear effect was

noted that is consistent within these previous studies, despite the present study

using a distinct population of children with SEND and different types of risks. It

appears that regardless of the types of risks included within these models a

point is reached where risks potentiated one another creating a more severe

effect on behaviour difficulties. It could be the coping resources children and

young people have displayed to counter lower levels of risk may fail once the

threshold is reached, and they are unable to overcome the negative influences

in their lives.

277

7.3.6 Research question 2c

Is the number of risks present within an individual’s background more important than the specific types of risks in accounting for behaviour difficulties? Research question 2c found that within both the primary and secondary school

models the independent additive model was a significantly better fitting model in

accounting for behaviour difficulties than the cumulative risk model. These

results suggest that putting multiple risks in to a statistical model independently

can account for more variance in behaviour difficulties than when a single

cumulative risk score (comprised of multiple risks) is used. Therefore, number

of risks (as assessed through the cumulative model) is not more important than

type of risk (as assessed through independent additive model) in accounting for

behaviour difficulties. This evidence that type of risk remains an important issue,

having different effects on outcomes is consistent with a number of other

researchers in this field (Hooper et al. 1998, Ackerman et al. 1999, Hall et al.

2010, Rousse & Fantuzzo 2009).

The evidence presented here suggests that type of risk is more important than

number of risk when attempting to account for variance in behaviour difficulties.

Although this is in direct contrast with the cumulative risk model, (where number

is seen as more important than type of risk), (Morales & Guerra 2006,

Appleyard, et al. 2005), the present study suggest that both types of models are

important. Using a cumulative model acknowledges the number of risks that

are present in an individual’s background and how these interact together to

predict behaviour difficulties. In such a case risks are seen as interchangeable,

and an effective picture of an individual’s behaviour difficulties could be

predicted if the number of risks in their background was acknowledged.

However, the independent additive model argues that if the specific types of

risks are known behaviour difficulties can be better accounted for in such

models.

278

It is perhaps pertinent to mention here that criticisms have been levelled at both

these types of models. Despite the independent additive model being able to

account for the relative importance of each risk variable independently (Gutman

et al. 2003), it is criticised for assuming risks are independent from one another,

whereas in reality they are interdependent (Flouri & Kallis 2007). The

cumulative model has also been criticised in the way in which cumulative scores

are generated which can sacrifice critical information when continuous scores

are dichotomized (Raviv et al. 2010). In fact the differences in measurement

between the two models may make it particularly difficult to reliably compare

between them. Perhaps the best approach to adopt is to use these two types of

risk models alongside one another, that way the intensity or specific type of risk

factors can be acknowledged along with the number of factors combined that

best predict the outcome (Schoon 2006). Indeed these two models are not

mutually exclusive and serve two different functions. The cumulative model is

interested in numbers of risks factors that accumulate within individuals to

predict outcomes. Whereas the independent additive model is useful in

assessing the intensity of specific types of risk factors that have the most

powerful influence.

279

7.3.7 Research questions 3a and 3b

Which school level predictor variables have a statistically significant interaction effect with contextual risk? Do the significant interactions display evidence of protective-stabilising or protective-reactive effects?

Within the primary model two school level protective factors were noted. There

were significant interactions between risk and school level academic

achievement, and risk and the percentage of children within the school at the

school action level of support. Within the secondary model only one significant

interaction term was noted. There was a significant interaction between risk and

the schools’ location (whether urban or rural).

The significant interaction effects show evidence that different levels of these

variables interact with the risk variable and influence behaviour difficulties in

distinct ways. Interaction graphs were computed to assess whether behaviour

difficulties are affected by different levels of the moderator variable for those at

high and low risk. The graphs displayed the direction of the effect and whether

high or low levels of the moderator variable confer an extra advantage to those

at high risk. Within primary schools, higher school level academic achievement

and higher percentage of children at SA within the school both moderate the

effect risk (in terms of poor positive relationships) has upon behaviour

difficulties. Both these interactions conform to a protective stabilising effect. In

secondary schools attending urban schools moderates the relationship between

risk (in terms of poor academic achievement) and behaviour difficulties and also

conforms to a protective stabilising effect.

Primary school level academic achievement The first significant interactive term to emerge was that risk significantly

interacted with school level achievement. This interaction effect indicated that

all students, regardless of their risk status benefit more from attending a school

with high academic achievement. In low risk contexts (i.e. having good positive

relationships) there is only a small difference between behaviour difficulties

280

displayed in those attending high and those attending low achieving schools.

However, in high risk contexts (i.e. having poor positive relationships) the

difference in behaviour difficulties displayed in those attending high and low

achieving schools is considerably greater. Those children at high risk attending

low achieving schools are significantly worse off compared with those attending

high achieving schools. In this context a higher degree of school level academic

achievement acts as a protective factor, and reduces the display of behaviour

difficulties.

This type of interaction, where different levels of the protective factor have a

relatively similar effect in low risk situations, although large differences in high

risk situations, has been termed a protective-stabilising effect (Luthar et al.

2000). For low risk children whether the protective factor was present i.e. higher

levels of school academic achievement did make some improvements to their

behaviour, but less so compared with when it was present amongst the high risk

children. In order for a fully protective-stabilising effect to occur the protective

factor i.e. high levels of school level overall achievement should remain stable

across high and low risk contexts, this was not the case in the present study.

There was a slight increase in behaviour difficulties displayed from children with

low to high risk statuses, when this variable was present. However, in its

absence (i.e. low levels of school level achievement); there was a more

dramatic increase in behaviour difficulties children from low to high risk

statuses.

Evidence has already been suggested for the promotive effects (main effects) of

school level achievement on behaviour difficulties in primary schools (see

section 6.4.4). Indeed a particularly strong effect was noted that higher levels of

achievement within schools are associated with fewer behaviour difficulties for

all children, which supports the majority of previous literature (e.g. Bisset et al.

2007). What needs to be justified here however, is why this variable was a

significant protective factor being particularly beneficial for those with poor

levels of positive relationships. Why does attending a school with high levels of

achievement have an even greater influence on those at high risk compared

with those at low risk? High risk children, who may struggle socially, may benefit

281

from a school that has a focus upon academic outcomes as this in some way

could distract them from their weakness in the social domain and perhaps focus

upon their strengths in a more academic domain. They may be less inclined to

display behaviour difficulties as the focus is upon academic skills. High

achieving schools are also likely to be the ‘better’ schools not solely in terms of

academic outcomes but in terms of how they support their pupils with social and

behavioural difficulties, and in such schools more effective interventions may be

provided which will reduce the display of behaviour difficulties. This is a

speculative argument and further research is needed however, to support to

refute this claim.

Primary school percentage of pupils as school action The second significant interactive term to emerge was risk with school % of SA.

This interaction effect indicated that for those children experiencing low levels of

risk (i.e. having good positive relationships), attending schools within either a

high or low percentage of children at SA has little impact upon their behaviour.

However, for children with high levels of risk (i.e. poor positive relationships)

behaviour difficulties are significantly worse for those attending schools with low

percentages of SA. Schools with a higher percentage of students at SA are

therefore acting as a protective factor. High risk children will have better

behaviour when attending such schools.

It is therefore advantageous for a child experiencing high levels of risk to attend

a school with more children at SA. The type of interaction, where the protective

factors confers advantages but considerably more so in high risk compared with

low risk situations has been termed a protective-stabilising effect (Luthar et al.

2000, Fergus & Zimmerman 2005, Windle 2011). The present study

acknowledges that the protective factor, ‘higher percentage of SA students’

conforms partially to this type of effect, offering more protection for those at high

compared with low risk. This evidence is reflected in the simple slope analysis

that showed that the line representing the moderator higher percentage of SA

was less steep than the line representing lower percentage of SA, and therefore

is offering some protection (see section 6.6.3b).

282

Some justification for why different levels of school % SA have distinct effects

on those children experiencing high and low risk is warranted. As noted above

in section 6.4.4 there was no independent risk or promotive effect of school %

SA on behaviour difficulties, (i.e. no main effect was established). The

significant interaction effect of this variable with risk, suggests that something

distinct is occurring, which influences those at high and those at low risk in

different ways.

Little research has been conducted surrounding the proportion of SEND

children in a school as a protective factor. As there is no explicit research to

compare with, provided below are some speculative reasons for this effect

which could be assessed within a future study. In low risk situations school

percentage of SA makes little difference to behaviour displayed; however, for

high risk children attending schools with a larger percentage of children at SA

has positive impacts on behaviour difficulties. In this case schools with higher

proportions of SEND children may have more resources available within them;

such resources, although of benefit for all children, are particularly beneficial for

high risk children in reducing behaviour difficulties displayed. In these schools it

is likely a more inclusive ethos is adopted that encourages support for peers

and participation in activities, and may help to replenish coping mechanisms

which may have become depleted in high risk children. Indeed this finding hints

towards an inclusivity effect whereby schools with higher degrees of diversity (in

this case, a larger percentage of SEND children) confer advantages to pupils

particularly at high risk. Indeed a similar finding has been presented in terms of

academic outcomes (Humphrey, Wiglesworth, Barlow & Squires 2012).

Secondary school urbanicity

One significant secondary school level protective factor was noted; an

interaction between risk and the schools location, whether urban or rural. This

interaction effect indicated that for those children experiencing low levels of risk

(i.e. having high academic achievement); attending schools within either urban

or rural locations has little impact upon their behaviour. However, for those

283

experiencing high degrees of risk (i.e. poor academic achievement) behaviour

difficulties are significantly worse for those attending rural schools. Urban

schools are therefore acting as a protective factor showing that high risk

children will have improved behaviour when attending such schools.

The interaction graph displayed shows evidence of a protective-stabilising effect

(Luthar et al. 2000), that despite increasing risk, the protective factor when

present (i.e. urban schools) maintained the degree of behaviour difficulties at a

stable level. The simple slope analysis showed that for urban schools,

behaviour difficulties remain stable in low and high risk, and therefore did not

deviate from the horizontal. When the protective factor was absent (i.e. rural

schools) there was a large increase in behaviour difficulties from low to high

risk, and the simple slope analysis showed that this did differ from the

horizontal. In conclusion at low risk a child’s behaviour will be unaffected by

school urbanicity, whereas at high risk a child would have better behaviour in an

urban school.

A discussion needs to take place as to why these effects have occurred and

particularly why urban schools offer some advantage over rural schools to those

at risk in terms of having poor academic achievement. The main effects of

school urbanicity have been discussed previously (see section 7.3.2), and

within the present study at least, this variable failed to emerge as a significant

finding. What is pertinent in this section therefore, is why attending a school in

an urban area would be particularly beneficial for those at high risk? Sellstom &

Bremberg (2006) provided evidence that urban schools have positive influences

on pupil outcomes, and within such contexts more resources are made

available to support the most at risk students. In rural schools, however, the

most at risk students may suffer in comparison, as often within these schools

fewer resources are available and they may have less access to intervention

programmes to support behaviour. Rural school pupils at high risk (particularly

in terms of having poor academic achievement) may not have certain coping

resources replenished (which combat the display of behaviour difficulties) as in

the case of urban schools. Consequently, high risk children are worse off and

display greater behaviour difficulties in rural compared with urban schools.

284

Although the effect of this variable was significant within the present study, the

exact mechanisms in terms of its functioning are speculative and further study is

needed to shed light on this area. It should also be mentioned here that this

finding is a particularly tentative one. The overall model from where this

significant effect emerged was not significantly better than full model in

accounting for variance explained in behaviour difficulties, for this reason and

those already mentioned in section 6.6.2, this finding should be interpreted with

considerable caution.

In conclusion these are interesting findings, although they have received little

attention within the published literature. The ‘short list’ of key protective factors

for general resilience functioning provided by Masten and colleagues (Masten

2006, Wright & Masten 2005, Masten & Reed 2002 and Masten & Coatsworth

1998), makes little mention of any school level factors suggesting they may not

be particularly important. In terms of behavioural outcomes a review of the most

salient protective factors does not acknowledge school level variables.

Evidence has suggested that protective factors may have a differing effect

within specific contexts, populations and outcomes (Fergus & Zimmerman

2005, Tiet et al. 2001 Rutter 2000); therefore the factors presented here may be

particularly important for children with SEND.

285

7.4 Limitations Psychological and educational research is often fraught with limitations, indeed

the very nature of research with human participants leads to a number of

constraints. No study can fully escape limitations in terms of its chosen

methodology and analysis of results. Addressing them however, enables their

relative importance to be assessed and results interpreted more reliably. Within

the present study limitations are discussed in terms of methodological and then

conceptual issues.

7.3.1. Methodological limitations

Sample representativeness There are potential limitations which are particularly pertinent concerning the

representativeness of the sample. The ten local authorities included within the

study were selected by the DfE (formally the DCSF) for the AfA project. These

local authorities are deemed representative of England (DCSF 2009b), although

the exact sampling criteria were not provided, so this claim is taken with some

caution. The schools selected from within each LA were also sampled to reliably

reflect the area in which they are located; some guidelines were given on school

selection, however, again the exact specifying criteria were not provided (DCSF

2009b). The AfA project had a particular focus on key outcomes such as

improving attendance or reducing bullying (see section 5.3 for more detail), it

could be surmised that LAs picked schools that were more likely to benefit from

the project as they had more significant problems in these areas. As such the

sample of selected schools may not be representative of the area in which they

are located. All children on the SEND register in years 1, 5, 7 or 10 within these

schools were the selected participants for the study. As SEND is fairly arbitrarily

defined (see section 1.3 for an overview) and is affected by a schools ability to

provide support for such pupils, the final sample may not be representative of

the picture found nationally.

286

Although there are potential concerns with the representativeness of the

sample, tables 5.5-5.8, provide an analysis to assess the deviation of the

present study’s sample in comparison to national averages. These analyses

were carried out across variables included within the study (at the individual and

school levels). Results showed that for all variables, only small effect size

differences existed between present study and national average, therefore

suggesting a representative sample. The only notable exception was school

level attendance which was poorer in the present study compared with national

averages.

Attrition rates Within any longitudinal study high levels of attrition are a common theme

(Cohen et al. 2011). Within the present study attrition rates were 58.3% for the

primary data set and 76.2% within the secondary data set. These figures

represent the percentage of participants who dropped out throughout the course

of the study due to non-completion of the WOST survey at either baseline or

follow-up. This is a clear limitation of the present study that a substantial

proportion of the optimum sample was not included within the present study,

and this may have led to biases within the sample. Nevertheless, the missing

data analysis (see appendix 8), showed missing data and attrition did not have

a significant impact on the results.

Despite sample characteristics being broadly comparable between those with

valid surveys present and those not, these comparisons were only made upon

measured variables within the study. It could be argued that some other

unmeasured variable could be attributable to the difference in those who

dropped out, and those who remained in the study. Nonetheless, as the present

study included a vast array of variables, it is difficult to conceive what these

potential unmeasured variables would be.

287

Achievement for all: An intervention? The data from the present study came from a wider project evaluating the

impact of AfA (Humphrey et al. 2011). One of the principle aims of the AfA

initiative was to improve a number of wider social outcomes for children with

SEND, one of these being behaviour difficulties. Schools that focused

specifically on improving behaviour difficulties could have experienced more

pronounced reductions in these problems compared to schools with a different

focus. The present study however, did not take into account which schools had

a specific AfA behaviour difficulties intervention. This decision was taken on a

number of grounds, firstly, all SEND children regardless of attending an AfA

school or not, will be subject to some kind of intervention. The vast majority of

participants in the present study received AfA as ‘the intervention’ but schools

outside of this project will have implemented some other plan or programme to

meet the specific needs for their pupils. In this regard AfA is not different from

any other intervention experienced by children outside of this project, and

results are generalisable to all children with SEND. Within the present study the

vast majority of children did attend schools which were implementing AfA,

(although not necessarily ones that were implementing a behaviour

intervention), however there was a minority of children which comprised the

control school sample for the AfA evaluation study. The inclusion of these

children within the present study limits the potential impact that AfA may have

had upon the present study’s findings.

Schools within the AfA sample did not always provide accurate information on

the two wider outcomes they were focusing upon, often being unsure of the

precise strategies being used to improve these wider outcomes. These

strategies varied widely between schools, with different amounts of time, energy

and resources being invested in them across the schools in the sample. Many

schools saw the wider outcomes as interrelated, whereby improving positive

relationships would have an effect on behaviour difficulties anyhow, e.g.

“several schools viewed the outcomes as mutually supportive and interrelated”

(Humphrey et al. 2011, page 65), and “schools drew clear links between the

different wider outcomes and strands of AfA, and many saw work in one area

directly impacting in others” (Humphrey et al. page 66). As these broad domains

288

were interdependent, some of the same strategies used by one school to

reduce behaviour difficulties may have been used in another to improve positive

relationships. As such it was deemed inappropriate to use a variable of whether

the schools’ explicit focus was on behaviour difficulties as a predictor.

The AfA intervention did reported a significant reduction in teacher reported

behaviour difficulties from baseline to follow up. This suggests that AfA appears

to be a successful intervention to reduce problem behaviour. However, on

closer inspection in comparison to non AfA schools the reduction in behaviour

difficulties from baseline to follow up, reflected a small change (mean = - 0.02)

and this equates to a small effect size. Furthermore, this finding found have

been influenced by sample sizes as there was considerable difference in the

AfA sample (4,426 pupils) compared with the control sample (194 pupils).

Humphrey et al. (2011) do acknowledge this issue and therefore suggest that

“given the relatively small number of respondents in the comparison school

sample, results should be treated as indicative rather than definitive” (page 78).

It is acknowledged within the present study that although the AfA intervention

may have had some effect on the data, these are likely to be very small.

Finally, the “emphasis throughout the project [AfA] was on the importance of

leadership and building upon existing good practice within schools to improve

outcomes for such children and young people” (Humphrey et al. 2011, page

18). The focus of AfA was hence upon existing good practice rather than

implementing something new and distinct for children within AfA schools. The

argument that the sample in the present study is not that much different from

any other sample of SEND children therefore holds, and enables the present

study’s findings to be generalised to all SEND children within schools in

England.

Additional variables The present study could be criticised for not including sufficient variables to fully

account for behaviour difficulties displayed. There is evidence (see chapter 2 for

more detailed discussion) that biological influences i.e., hormones (Book et al.

2001), neighbourhood influences i.e., exposure to violence (Patchin et al. 2006),

289

as well as family influences i.e., poor parenting (Pettit et al. 2001) and peer

influences i.e. being involved with anti-social peers (Ingoldsby et al. 2006) all

contribute to behaviour difficulties displayed. The present study acknowledges

that these are important variables; nonetheless it would be impossible within a

single study to include every potential influence on behaviour difficulties, and

attempting to do so would go beyond the scope of the study.

The target population was children with SEND, as little research has explicitly

utilised this population the variables included within the present study make an

important contribution to explaining these behaviours within a school context.

Research into behaviour difficulties, (a well-established albeit complex field)

results in a compromise between searching for new potentially important

predictor variables of these difficulties versus establishing existing ones within a

specific population i.e. children with SEND. In the case of the present study

twenty variables that all had a theoretical justification for inclusion within the

study as possible predictors of behaviour difficulties for typically developing

children were assessed in terms of a population of children with SEND.

Teacher effects There are potential limitations concerning the collection of data from the surveys

and particularly surrounding the teachers who completed these for their key

pupils. Teachers rated pupils on their behaviour difficulties, positive

relationships and bullying behaviour at baseline, and at follow-up, behaviour

difficulties were rated again. As follow-up was 18 months after baseline children

had moved year groups and were likely to have a different teacher who would

have rated them across this construct. It could be argued that change in

behaviour difficulties was therefore due to change in rater, rather than real

change in behaviour. This argument is however, countered with information on

the psychometric properties of the WOST, which have shown good inter rater

reliability between teachers and parents of 0.483. It is likely that inter rater

reliability between teachers would be even higher as they would be observing

behaviour within the same context.

290

A second potential limitation concerns the accuracy of the information provided

by teachers. Some teachers especially in secondary schools were the key

teachers for large numbers of children with SEND. Teachers, who completed

the survey on multiple occasions for multiple children, could have experienced a

fatigue effect, and possibly lacked the motivation to complete the surveys

accurately. Although this may have been an issue, realistically each survey was

very short and would take no more than 5 minutes to complete per child, so it is

unlikely that fatigue dramatically affected the results. Thirdly, teachers could

have experienced some conflict in completing the survey, being concerned that

their individual performance was being assessed (although the information

sheet they received clearly told them this was not the case). They could have

therefore rated pupils more favourably in order for their own performance as a

teacher to appear more positive. Teachers holding pre-existing ideas about the

AfA initiative or in some way having a vested interest in the outcome could have

rated their pupils in accordance with their personal view.

Using a teacher report of behaviour difficulties could be criticised as being less

accurate than a pupils own self report. Nonetheless, teacher reports were

chosen within the present study in order to maintain consistency, as younger

pupils (those in year 1) and those with the most complex SEND (i.e. PMLD)

would not be able to complete the task which would result in a significant loss of

information. Furthermore, if a self-report was used there could have been a

developmental effect, i.e. as children mature they become increasing more

accurate in their self-reporting. Therefore changes in behaviour difficulties may

reflect accuracy of the rater rather than any actual behaviour change. Teachers

are in the best position to reflect on behaviour difficulties which occur in their

classroom and around the school. They are also likely to be less biased in their

judgments, and as they see a wide spectrum of behaviours displayed they are

able to reliably assess degrees of behaviour difficulties displayed in relation to

other children.

291

Analytical Strategy The analytical strategy chosen for the present study was MLM, using such a

method allowed for variance to be explained at individual and school levels

respectively. A criticism concerns the fact no classroom level was

acknowledged. This ecological layer would fall between the individual and

schools levels. Students are clustered within their classrooms which in turn are

clustered within schools; consequently the effects observed at the school level,

may in fact reflect classroom level differences.

However, the decision to exclude this potential source of variance was taken as

to include a classroom level when already over 300 schools were included

within the study could be counterproductive and goes beyond the scope of the

study. Classrooms are variously defined and there are vast differences in

‘classrooms’ between primary and secondary schools (see section 5.9.2). Due

to the consistency between the primary and secondary models, and classrooms

being hard to define in secondary schools, this ecological level was excluded

within the present study. It is however, acknowledged as a potential source of

variation, which maybe particularly important within education based studies,

and is therefore an idea for future research. However, to account for a

classroom level for secondary school students who may be in around 10

different classrooms each of different compositions would require some kind of

cross classification multi-level model which are incredibly complex and require

very large sample sizes. As such this type of investigation may be out of reach

to many researchers.

Although not an explicit research question, the present study could be criticised

for not accounting for the relative importance of each of the significant risk

factors, and how they compare with one another. MLM outputs in MLwiN and

SPSS do not produce standardised coefficients and therefore it is not possible

to compare between variables that were measured on different scales. One way

to overcome this issue would be to standardise the variables before they are

added to the model, however a certain amount of meaning in the data is lost as

standardised coefficients are difficult to interpret in isolation.

292

Wider Outcome Survey for Teachers (WOST) A potential limitation of the present study concerns the measurement tool used

to assess behavioural difficulties. As previously discussed in section 5.6.2, the

WOST (Humphrey et al. 2011) was assessed for its psychometric properties,

with the conclusion that was deemed acceptable. Specifically it has been

argued there exists good content validity, adequate construct validity, good

internal consistency, respectable inter-rater reliability, good interpretability and it

acknowledges the impact of floor and ceiling effects (Humphrey et al. 2011).

Nonetheless, despite these findings there are a number of criticisms of these

analyses. Firstly, construct validity was measured by computing correlations

between the different constructs within the same survey. As they conformed to

the hypothesised expected direction, i.e. behaviour difficulties was positively

correlated with bullying and negatively correlated with positive relationships, this

was taken as a measure of adequate construct validity. Validating a survey

construct on other constructs within the same measure may not be a useful

exercise, and should be assessed against an external measurement. As such

the claim that this measure has construct validity must be interpreted in this light

and with considerable caution.

Secondly, in relation to internal consistency, although all Cronbach’s Alphas are

>0.7 which is deemed acceptable, the confirmatory factor analysis indicated that

only 2 of the most common 7 fit indices met the desired criteria. Whereas the

remaining 5 only approached the ideal fit criteria (see section 5.6.2). This could

have implications for the study in whether internal consistency actually exists

within this measure. Thirdly, there are other psychometric measures such as

criterion validity, test‐retest reliability and responsiveness which were not

carried out when the validation of the WOST was undertaken. Without such

evidence the psychometric properties of the WOST are weakened.

Further analyses needs to be undertaken with the WOST in order to overcome

some of these limitations. The collection of further data to validate the survey

against an external measure such as the SDQ would be particularly beneficial.

As such the conclusions drawn from this study need to be interpreted in the light

of the limited psychometric properties of this survey.

293

7.3.2. Conceptual limitations

Behaviour difficulties A key conceptual limitation within the present study concerns the major concept

under investigation i.e. behaviour difficulties. These types of behaviours can be

interpreted subjectively, and a similar issue identified by one person could be

seen as a major incident, whereas to a different person it is regarded as minor.

This point was expressed by Cooper, Smith & Upton (1994) who state “different

teachers (and schools) have different standards for, and expectations of, the

behaviour of their pupils whereby differences in reported behaviour may reflect

differences in the degree to which a behaviour is tolerated”, (page 17). This

could have been an issue within the present study, however as a reliable and

valid measurement tool was used (i.e. the WOST) it is hoped that subjective

interpretations were kept to a minimum.

It has also been noted that assessments of behaviour difficulties are affected by

the child’s age, gender and ethnic background (Papatheodorou 2005, Green et

al. 2005, Zeitlin & Refaat 1999), with certain behaviours judged as more

problematic depending on who is exhibiting them. Teachers may have rated an

individual’s behaviour objectively and not accounted for their age, gender or

ethnicity, or could have interpreted these behaviours in the light of these

individual characteristics. In either case there could be some bias in the

recording by either over or under reporting behaviour difficulties. Nonetheless,

the present study acknowledges using a teacher report will result in the best

possible measure of behaviour difficulties displayed. Teachers are in the best

position to know each child and observe their behaviour and can account for

appropriate age, gender and ethnic effects in behaviour difficulties displayed.

A limitation within the present study could be the use of the measurement tool

to assess behaviour difficulties. There are numerous assessment methods

available for researchers in this area, which include rating scales, observations,

interviews and socio-metric techniques (Merrell 2003). The method selected by

a piece of research to assess behaviour difficulties will affect who is defined as

294

having behaviour difficulties and also the degree to which their behaviour is

posing a problem. The present study utilised a rating scale, (i.e. the WOST),

and these types of assessments are the most commonly used (Merrell 2003).

Being aware of how behaviour difficulties have been assessed is important, not

only in order to gain a greater appreciation of this construct, but also to allow

meaningful comparisons between previous studies.

Special Educational Needs and Disabilities

As already mentioned in section 1.3, there are many issues surrounding the

concept of special educational needs and disabilities. Within the present study

the definition set out within the SEN code of practice was adopted and states “A

child has ‘special educational needs’ … if he has a learning difficulty which calls

for special educational provision to be made for him” (DfE 2001a page 6 section

1.3). There is an implicit assumption within this definition that if schools are able

to cope with the child and meet his/her needs, no additional support is required.

However, if they are unable to meet the child’s needs, he/she will be defined as

having SEND and extra support will be put in place. The definition also states

“Children are considered to have a learning difficulty if they a) have a

significantly greater difficulty in learning than the majority of children of the

same age”. This could be problematic as using the term ‘majority of children’,

could refer to those within the same school or children across the whole

country. As schools have vastly different numbers of children with SEND, what

is considered the majority of children will vary by school, which will then have an

influence on who is defined as having SEND.

The SEND definition is contentious and, as mentioned previously, is criticised

for allowing considerable variation across schools in who is defined as such

(Ofsted 2005). There does not appear to be a consistently applied approach to

identifying and assessing children with SEND, which include teacher

observations, standardised tests and tools, progress on National Literacy and

Numeracy Strategy Frameworks, performance in National Curriculum levels or

baseline screening tests (DfES 2001a). The lack of consistency here makes the

concept more of a school-based variable than an individual one, and becomes

295

more about whether the school has the resources available to support the

child’s needs, therefore reflecting an arbitrary process. The SEN code of

practice does give some more specific guidance that needs should fall within at

least one of four main domains: a) cognition and learning, b) behaviour,

emotional and social development c) communication and interaction or d)

sensory and /or physical needs. The sample within the study is consistent in at

least one regard, that they have experienced problems in a minimum of one of

these areas, and this makes them in some way a distinct population within the

school.

Cumulative Risk Score

The present study relied heavy on the cumulative risk score as a measure of an

individual’s degree of contextual risk experienced. There is concern that as risk

factors are not certainties but probabilities (Schoon 2006), there may be

considerable variability in what actual risk an individual experienced as a result

of having a certain factor present. However, by using a cumulative risk score,

variables are aggregated into one unified score. This has the effect of

diminishing the unique importance of any single factor, and therefore by adding

multiple risks together, allows a more accurate picture of the amount of risk an

individual is experiencing to be acknowledged.

Criticism has however, been levelled against the cumulative risk score

concerning the accuracy in how it is measured. Debate surrounds whether the

risk variables that make up the cumulative score should be noted as such on

empirical or theoretical grounds (Stouthamer-Loeber et al. 2002). Although the

most popular method is upon empirical grounds, issues remain whether risk is

conceptualised when the correlation between risk and the dependent variable is

significant (Lima et al. 2010) or when it is equal to or exceeds .25 (Atzaba-Poria

et al. 2004). Differences in how the score is computed could affect the reliability

when comparing between studies.

On ‘risk’ variables the top 25% of the sample are given a score of 1 denoting

‘risk’ and a score of 0 to denotes its absence. Total scores are added across

296

variables resulting in a cumulative risk score. There is criticism that

measurement techniques taking the top 25% to represent risk is an arbitrary

decision (Sullivan & Farrell 1999), and dichomising variables in this way results

in oversimplifying the data set, resulting in information loss (Pollard et al. 1999).

In response to some of the criticisms however, Farrington & Loeber (2000)

argue that splitting data by means of dichotomisation results in a minimal effect

on the data and does not affect the conclusions drawn from these studies. The

cumulative risk approach highlighted here is the most prolific found within the

literature and was therefore utilised within the present study.

297

7.5 Future research

Additional variables In the present study over twenty variables were measured as potential risk,

promotive or promotive factors for behaviour difficulties in a population of

children with SEND. There is scope however, for further investigations to

include more variables that are particularly pertinent to a SEND population.

More detailed information on ethnicity, and first language spoken and whether

children are “looked after” could be collected, as these are salient issues for the

SEND population (DfE 2010d). Additionally, variables such as self-esteem

(Schonberg & Shaw 2007), or certain personality traits (Viding et al. 2007) could

be collected. Furthermore school level variables might include the extent to

which a child with SEND participates in extracurricular activities at their school,

as this could be a promotive or protective factor (Mahoney 2000). The gender

composition of the school could also be measured, as at the individual level this

factor predicted behaviour difficulties, therefore investigating whether schools or

classes with a higher proportion of males could experience more behaviour

difficulties would be of benefit (Andersson et al. 2010, Hoxby 2002). Finally,

effective school characteristics such as having comprehensive policies, strong

leadership, or a climate of fairness and consistency, where pupils feel safe and

a have sense of belonging could all affect behaviour difficulties displayed, and

be important for a SEND population (Denny et al. 2011, Evans-Whipp et al.

2004).

There is also scope for future research to take into account the effects of

additional ecological levels. Indeed Bronfenbrenner (1979) has stated that

behaviour cannot be fully understood without the acknowledgement of multiple

ecological levels. Perhaps most pertinent with a sample of children with SEND

would be to include a classroom level. The school level influences already

measured could then be broken down to this level and the more proximal

influences of a classroom could be compared with the more distal effects of the

school level. Introducing a family level and variables that are nested within this

level would add to the research base. A number of potentially important

298

variables at this level have previously been acknowledged (see section 2.3) and

have included attachment relationships, mental health problems in parents,

parental discipline and family disruption. A particularly important factor for

children with SEND is the extent of parental engagement, and specifically how

involved parents are with their child’s education (Humphrey et al. 2011).

Including all these variables may add to the variance explained in behaviour

difficulties.

Further research could also pay greater attention to the influences of behaviour

difficulties at the neighbourhood level, as attempting to explain childhood and

adolescent behaviour difficulties without such effects cannot fully account for

the problem (Klaff et al. 2001). Factors that are important here include

neighbourhood disadvantage, (Drukker et al. 2003), residential instability;

(Brooks-Gunn et al. 1997), exposure to violence, (Patchin et al. 2006),

interaction with deviant peers in the neighbourhood; (Ingoldsby et al. 2006),

living in areas with higher crime and poverty levels (Beyers, Loeber, Wikstrom &

Stouthamer-Loeber 2001), more neighbourhood structural disadvantage (Mrug

& Windle 2009) and poorer access to facilities (Renzaho & Karantas 2010).

Children will have a least some exposure to the neighbourhood in which they

live, with these effects manifesting themselves directly or indirectly through

other variables. It may be important to investigate how these factors influence

behaviour whether directly or mediated through another variable (Vanfossen,

Brown, Kellam, Sokoloff & Doering 2010). Furthermore, knowledge could be

gained by assessing the relative strength of these factors, and it is hoped that

as research into MLM develops an effective way of comparing across variables

perhaps by using standardised coefficients would be a useful pursuit.

Processes of risk variables Although further research would benefit from establishing more variables across

additional ecological levels as risk, promotive or protective factors for behaviour

difficulties in children with SEND, it would also be productive to investigate

some of the reasons why these factors are significant. Kraemer et al. (2005)

have argued that there needs to be a ‘fundamental shift in risk research’ (page

299

69), as adding in new risk factors does not necessarily help to further account

for a particular disorder, as often these ‘new’ factors have little potency on the

outcome.

Kraemer et al. (2001) suggest in order to gain a better understanding of risk

factors a concentration on the processes of moderation and mediation need to

take place. Acknowledging how proximal and distal influences interact to

influence developmental problems is an important pursuit (Flouri et al. 2010).

Indeed, investigating the mechanisms by which these specific risk factors

function (Raviv et al. 2010) will lead to a greater understanding of the pathways

that lead to the display of behaviour difficulties and ultimately be an essential

precursor to any intervention designed to combat the problems (Sameroff et al.

2003). Evidence suggests that risk factors do not always exert their influence on

an outcome variable in isolation but there are potentially numerous ways that

risk factors work together, which often include complex relationships and

interactions between variables (Kraemer et al. 2001, Atzaba-Poria et al. 2004).

Therefore, what is needed is further research on how the risk, promotive or

protective factors interact to predict an outcome.

Cumulative Promotion A potential area for future research, to extend the findings within the present

study, would be to acknowledge the impact of multiple ‘positive’ factors that

promote behavioural competence and subsequently reduce the chance of

behaviour difficulties displayed. There is a high probability that promotive

factors like risk factors accumulate within individuals (Werner 2000). Some

studies have acknowledged these promotive factors within a cumulative

framework, although to a much lesser degree than studies using risk factors

within a cumulative framework (Beam, Gil-Rivas, Greenberger & Chen 2002,

Loeber et al. 2008, Van der Molen, Hipwell, Vermeiren & Loeber 2011).

Previous studies have shown a cumulative promotion effect in terms of reducing

delinquency. Specifically that a higher number of cumulative promotive factors

at the individual and family level predicted a decreased likelihood of being a

300

minor or serious delinquent, and an increased probability of being a non-

delinquent (Van der Laan, et al. 2010). Within a school context, evidence

suggests that an increasing number of promotive factors resulted in reductions

in the chance of disruptive behaviour (Van der Molen 2011). Further support for

this phenomenon also comes from Jessor, Van Den Bos, Vanderryn, Costa &

Turbin (1995), Sameroff et al. (1998) Herrenkohl et al. (2003) & Fergusson,

Vitaro, Wanner & Brendgen (2007) who all investigated the cumulative effects

of promotive factors finding that as these factors increased, the likelihood of

displaying problem behaviour was reduced. Research surrounding cumulative

promotion, specifically for children with SEND, investigating whether it reduces

the chance of displaying behaviour difficulties, would therefore add to the

literature base.

An analysis could be undertaken to assess the exact relationship between

number of promotive factors and outcome, as some have argued that this

relationship is linear (Stouthamer-Loeber et al. 2002, Loeber et al. 2008),

whereas others have suggested there could exist a curvilinear relationship

(Luthar & Cicchetti 2000). Future research could also assess whether the

effects of cumulative promotion or cumulative risk hold the greatest influence

over an outcome as, there is a particularly small literature base in terms of how

cumulative promotion relates to cumulative risk (Flouri 2008, Epstein, Botwin,

Griffin & Diaz 2001). It has been suggested that there may well be differences in

the strength of risk versus promotive effects in influencing behaviour problems

(Van der Put, Van der Laan, Stams, Dekovic & Hoeve 2011). Furthermore a

counterbalancing effect may exist whereby the specific number of risk factors,

or the number of domains considered denoting risk can be cancelled out with an

equal number of promotive factors, or domains denoting promotion

(Stouthamer-Loeber et al. 2002).

Ecological effects of cumulative risk In the present study it was not possible to assess the differing effects of

cumulative risk across distinct ecological levels. Cumulative risk was assessed

at the individual level, although as only one significant predictor was established

301

at the school level, for both the primary and secondary models, a cumulative

score could not be generated at this level. Future research could measure a

greater number of variables at the school level and the significant factors could

be calculated to form a cumulative risk score. Furthermore, if variables were

assessed at the neighbourhood and family level, a cumulative risk score could

also be generated for these levels. Deater-Deckard et al. (1998) have

suggested that cumulative scores across multiple ecological levels all contribute

to variance in explaining behaviour problems and are important in

understanding the aetiology of these behaviours.

If a cumulative score was to be established across multiple ecological levels

then an analysis could assess the relative importance of each of these levels

(Atzaba-Poria et al. (2004). Finally, testing the idea whether the same number

of risks within a single ecological context has similar effects on behaviour

difficulties compared to the same number of risks but spread across multiple

ecological contexts would be an interesting pursuit (Morales & Guerra 2006).

This idea for future research could test the hypothesis proposed by Call and

Mortimer (2001) that individuals need a ‘safe haven’ or an ecological context

where no risk is present, as this is essential to protect against risks experienced

in other ecological contexts.

Protective factors Further research could be conducted to assess the influences of protective

factors for behaviour difficulties. In order for a protective factor to be

conceptualised as such, there needs to be evidence of risk. There are a number

of ways that risk can be assessed which include being exposed to a single

significant adverse situation (i.e. death of a parent), or generating a score on a

checklist of multiple negative life events, or a cumulative risk score (Masten

2001, Luthar & Cushing 2002). The different ways in which risk has been

measured could be compared with one another, to assess the salience of any

identified protective factor. Within the present study, a gap within the literature

was noted that school level protective factors had not been explicitly highlighted

with the outcome behaviour difficulties. As such, the analyses presented here

302

maintained a focus on school level variables. It is known that protective factors

are relevant across numerous ecological levels (Wright & Masten 2005), and

therefore some acknowledgment of these specifically for a sample of children

with SEND is warranted.

It is not only important to uncover the protective factors that can moderate risk

experience and lead to better behavioural outcomes, but also to understand

exactly how these underlying processes or mechanisms of these factors work

(Vanderbilt-Adriance & Shaw 2008, Luthar, Sawyer, Brown 2006, Kaplan 2002).

Luthar et al. (2006) cite an example where if a high level of family support is

considered a prominent protective factor the question should be asked, what is

it about this variable that fosters resilience? 49

Could it be self-esteem, security,

or feelings of control? Where appropriate, future studies should look for these

underlying mechanisms and processes, not solely being focused if a factor

plays a role but how it does.

Within the present study a variable-focused approach was taken as a means of

highlighting the significant factors that protect against behaviour difficulties. An

alternative approach however, could be utilised within further study, such as a

person-focused approach. Within such analyses groups of individuals are

established on whether they are experiencing high degrees of risk or not, and

whether they are displaying behavioural competence or not. As such 4 distinct

groups are compared with one another, i.e. resilient individuals (high risk, high

competence) are compared with competent individual (low risk, high

competence) and again with maladaptive individuals (high risk, low

competence) (Masten & Obradovic 2006, Jaffee et al. 2007). It is anticipated

that similarities will emerge between resilient and competent individuals and

differences between resilient and maladaptive individuals. These analyses can

assess the percentage of individuals deemed resilient (Windle 2011), and they

can be effective when assessing multiple outcomes over substantial time

periods (Masten & Reed 2002).

49 Resilience is defined as “a dynamic process encompassing positive adaption within the context of significant adversity” (Luthar, et al. 2000, page 543). Individuals are therefore able to achieve a better outcome than would be expected from their risk experienced.

303

Comparing between individuals is an important issue as Luthar et al. (2006) cite

an example whereby for ‘high risk youth’ greater exposure to stringent limit

setting by parents is related to increases in prosocial behaviours, but for ‘low

risk youth’ it leads to decreases. In order to assess interaction effects however,

both groups need to be included together, and in this instance at least the

effects would counteract one another leading to non-significant main effect

findings. Therefore searching for interaction effects in this instance would

obscure the fact that limit setting was important for low-risk youth. It has been

argued that using a within group analytical approach of just those at high risk,

will be able to shed light on the relative importance of certain protective factors

in relation to the other variables, rather than comparisons between the relative

importance of a certain factor in high and low risk groups (Luthar et al. (2006).

304


• Summary of results: This section presented a brief summary of the

results, divided up according to the three broad research questions.

• Discussion of findings: This section discussed the findings of the

present study in relation to previous literature. Discussion was set out by

research question, with each of the risk, promotive and protective factors

discussed in turn.

• Limitations: This section focused on the limitations of the study,

beginning by discussing methodological concerns such as the sample

representativeness, attrition levels, whether AfA being an intervention

influenced the results, possible teacher reporting effects, and the

analytical strategy. Conceptual concerns were acknowledged around the

key concepts of the research behaviour difficulties and special

educational needs and disabilities and the cumulative risk score,

• Future research: In this section, further ideas for additional research

were discussed, including a) using previously unacknowledged variables

in accounting for behaviour difficulties for children with SEND, b)

accounting for the processes involved within the significant risk,

promotive and protective factors identified in the present study, c) using a

cumulative promotion metric to assess multiple positive influences on

behaviour difficulties, d) assessing risk across multiple ecological levels

and e) more in depth study of possible protective factors.

305

8. Conclusions

8.1. Introduction to chapter

In this chapter the final conclusions of this study are brought together. In section

8.2 the implications of the study are discussed. In section 8.3 the study’s

contributions to knowledge are outlined. Section 8.4 provides an overview and

summary of the study. Finally section 8.5 provides summary statements for the

chapter.

8.2 Implications

Given the present study utilised a nationally representative sample of children

with SEND; the resulting implications are extensive and wide reaching. For

purposes of clarity the implications of the present study are organized by

research question, from which they emerged.

8.2.1 Implications from Research Question 1

Acknowledging a school effect Interventions to reduce behaviour difficulties in children with SEND need to take

into consideration that the school in which they attend alongside factors within

them as individuals, make a significant contribution to their behaviour displayed.

The results of the present study showed that the individual and school

ecological levels, (within both the primary and secondary schools) significantly

contributed to a proportion of variance in behaviour difficulties. Therefore,

regardless of the type of school i.e. whether primary or secondary schools need

to take seriously their influence in the display of negative behaviour of their

children with SEND. They must not assume such behaviours are displayed

solely due to a child’s individual characteristics or their family background, as

evidence shows school level variables play a significant role in influencing these

problems. Interventions that adopt a whole school focus will be beneficial to all

306

pupils directly, as well as having a positive effect on the school culture or ethos.

Schools could adopt any of the numerous behaviour problem interventions that

are discussed and evaluated within two important reviews of the literature (e.g.

Greenberg Domitrovich & Bumbarger 2000, Maag & Katsiyannis 2010).

It is however, noted that the effects at the individual level are considerably

greater than school level influences. Although this does not dismiss the school

effects as unimportant, and whole school approaches should be adopted to

reduce behaviour difficulties, interventions that target specific individuals may

be the most effective in reducing behaviour difficulties displayed (Losel &

Beelman 2003). Indeed one study assessing the impact of the small group

aspects of primary SEAL (Social Emotional Aspects of Learning, DfES 2005),

showed evidence that targeted individuals did experience reductions in their

behaviour difficulties, as presumably they had more to gain than other pupils

(Humphrey et al. 2008).

Risk and promotive factors important for primary and secondary schools The findings from the present study show that for SEND children, being ‘male’ is

a risk factor for displaying behaviour difficulties across both primary and

secondary school. As this variable is a fixed marker (i.e. a risk factor that cannot

be changed), interventions should specifically target this group of individuals.

The aim of such interventions would be to enhance positive factors which

hopefully promote competence and protect against behaviour difficulties, such

as improving relationships or academic outcomes. Year group is another fixed

marker, and significant risk factor for both primary and secondary schools. The

older children in primary school and the younger children in secondary schools

are more likely to display behaviour difficulties. It is important to specifically

target this transition period and the end of childhood/beginning of adolescence.

Consideration needs to be taken that the type of behaviour difficulties displayed

may be different across primary and secondary schools, and targeting

aggressive behaviours particularly in primary schools and non-aggressive

behaviour difficulties in secondary schools may be the most effective approach.

FSM eligibility is also a key risk factor for behaviour difficulties, although this

307

variable could change, it is not within the power of the schools to do so. If

however, schools employed family support workers who work with at risk

families to diminish some of the negative influences often mediated through

lower SES, such as poorer parenting skills, this may improve the situation.

Perhaps the pertinent issue here for schools is that males, those of lower SES,

those within year 5 in primary and year 7 in secondary are the groups who

would benefit to the greatest extent from a targeted intervention programme.

A further significant risk factor across both primary and secondary schools is

being a bully (or a bystander to bullying incidents, specifically for secondary

school children). This variable is considered a variable marker (i.e. risk factors

that can be changed). Interventions that reduce bullying will therefore have a

subsequent positive influence on behaviour difficulties displayed. These

interventions, could be specifically targeted at those who have been termed

‘bullies’, as this will ultimately reduce behaviour difficulties. In secondary

schools pupils should be educated about the effects of being a bystander to

bullying and why this is considered problem behaviour. Programmes such as

Positive Behaviour Supports (PBIS) may be particularly effective in this regard

(Pugh & Chitiyo 2012).

Risk and promotive factors uniquely important for primary schools Within primary schools (and not within secondary schools) a further fixed

marker risk factor emerged; being born in the autumn. Those children born

within the autumn months are the oldest in their year and were more likely to

display behaviour difficulties than those born in the summer months and the

youngest in their school year. This variable cannot be changed, nonetheless

specially targeting the older children within a school year and educating them

around the issues of behaviour difficulties may help to reduce these problems.

Furthermore, separating children into classes or groups by season of birth,

particularly when developmental differences are most profound, in the earlier

years of primary school, may help to reduce problem behaviours displayed.

308

The quality of pupil relationships with peers and teachers within schools

(particularly primary schools), is a variable risk factor. Improving these

relationships will have an effect of reducing behaviour difficulties. Teachers

should be aware of their influence in this regard, and having better relationships

with the children in their class, and promoting effective peer-peer

communication and working will reduce behaviour difficulties displayed.

Teachers should be given opportunities to engage positively with their pupils,

and encourage positive peer relationships through effective interventions such

as Promoting Alternative Thinking Strategies (PATHS), (Kam, Greenberg &

Kusche 2004). Finally, at a school level, schools that improve academic

achievements in all their pupils regardless of their SEND status, and help them

to achieve their full potential, will have a positive influence on all children who

attend, and evidence suggest this will reduce behaviour difficulties displayed.

Risk and promotive factors uniquely important for secondary schools Within secondary schools those attending larger schools are considerably more

at risk than those within smaller schools. As this is a variable risk factor, where

possible, school size, (in terms of pupil numbers), should be kept to a minimum.

This is in order to reduce feelings amongst pupils of anonymity and promote

feelings of being valued and involved in the school. Having a key worker for

each child with SEND may prevent them getting ‘lost’ in the school system.

Encouraging teachers to take an active interest particularly of pupils with SEND

may reduce this effect.

Within secondary schools low attendance levels were a variable risk factor for

behaviour difficulties displayed. Promoting attendance will therefore have a

positive effect on children who are display behaviour difficulties. Employing

educational welfare officers within schools to help keep attendance to a

maximum will be beneficial, as well as counsellors to find the reasons why

children are not attending in the hope of improving the situation.

The final variable risk factor within secondary schools was poor academic

achievement. Schools should therefore provide interventions to support the

309

academic achievements of their children with SEND, such as implementing

Achievement for All 3A’s as improving their academic levels will have an effect

on reducing behaviour difficulties.

8.2.2 Implications from Research Question 2

Findings from the present study show that more risk in an individual’s

background, regardless of the specific nature of that risk, equates to more

severe behaviour difficulties. In order to reduce behaviour difficulties

interventions need to take into consideration multiple risk variables, attempting

to reduce any which are evident in a particular individual’s background. It is not

important exactly what risk variable is minimised, as the evidence suggests that

multiple risk factors work together, and when they accumulate within an

individual this will result in more severe behaviour difficulties displayed.

Therefore, to some extent, regardless of the specific type of risk, interventions

should reduce what they can. As some risk variables are fixed markers and

cannot be changed there is requirement to pay particular attention to those

variables which can be changed. Furthermore, when risks are unknown any

global approach that aims to reduce behaviour difficulties such as Social and

Emotional Learning (SEL) programs may be particularly effective (Durlak,

Weissberg, Dymnicki, Taylor & Schellinger 2011).

The relationship between cumulative risk and behaviour difficulties was non-

linear and behaviour difficulties increased disproportionally with increasing risk.

This is a worrying trend if interventions do not pay particular attention to it.

Indeed those individuals who have a large numbers of risk factors in their

background are at a proportionally greater disadvantage that those with fewer

risks. Tailoring any intervention towards those considered most at risk would

have the effect of reducing the disproportionate relationship between increasing

risk and behaviour difficulties. Indeed evidence suggests that higher risk youth

will benefit to the greatest extent from any intervention designed to reduce

aggressive behaviour (Wilson, Lipsey & Derzon 2003). It would be fairly easy in

schools to make a list of a child’s risk factors and those who have particularly

310

high levels of risk should be provided with interventions. Indeed Wilson & Lipsey

(2007) have argued that the most effective intervention programmes involve a

general or universal approach alongside a more targeted approached directed

at specific individuals considered at high risk. Nonetheless, it should be

acknowledged that severe levels of risk are experienced by relatively few

individuals. Therefore, tailoring interventions towards this group of children

alone, although beneficial to those directly involved, will make only a small

impact on these problems (Flouri 2008).

Despite the focus on reducing the risks present in an individual’s background

regardless of what they are, it is also important to acknowledge that specific

types of risk factors do hold differing degrees of influence over an outcome in

question. Behavioural improvements may be more effectively accomplished by

targeting the risks deemed the most powerful. However, where it is not obvious

which of the risk factors is holding the greatest influence just reducing the

number of risks as the cumulative model suggests would be appropriate and

beneficial for the individual concerned.

8.2.3 Implications from Research Question 3 Attending primary schools with high overall levels of academic achievement are

beneficial to all children with SEND; therefore global school level interventions

that aim to improve academic levels will be effective for every child in reducing

their behaviour difficulties. However, for those considered at high risk (in terms

of poor positive relationships), schools with higher overall academic

achievement can act as a salient protective factor, reducing the display of

behaviour difficulties. It is particularly important therefore that these high risk

children attend schools with high levels of academic achievement.

One option would be to adopt positive discrimination in this regard. Where

places at certain schools are limited, those considered at higher risk could be

given the preference to attend the higher academic achieving schools.

However, moving a child to a higher achieving school on the basis of their ‘at

311

risk’ status is not necessarily a feasible option. It would be more appropriate to

investigate the characteristics of higher achieving schools and implement some

of their policies and procedures in lower achieving schools. This would be

beneficial for all children and particularly for those considered at high risk. As

higher school level achievement is a significant protective factor, poorer

achieving schools have further incentive to increase their overall academic

achievement. This could result in significant reductions in behaviour difficulties

displayed by their children at high risk. Primary schools that have more

significant behaviour problems may be able to use interventions that aim to

improve academic attainment as an effective means of reducing behaviour

difficulties displayed by their pupils.

Within primary schools attending schools with a larger percentage of pupils with

SEND (particularly at the school action level of support) are beneficial for

children considered at high risk (in terms of poor positive relationships). This

variable acts as a protective factor, and has a positive effect in reducing the

display of behaviour difficulties. It is therefore beneficial for children considered

at high risk to attend a school with greater numbers of pupils with SEND. This is

an important finding within the present study, however, moving a child to a

school on the basis they have higher numbers of children on the SEN register at

SA because of their ‘at risk’ status is perhaps not a feasible option. Instead

looking at the characteristics of schools that have higher numbers of SEND

pupils, and applying these in the schools with lower percentage of SA pupils

would be beneficial for high risk children. It may be within such schools there

are greater levels of inclusion whereby an ethos is created within the school that

celebrates diversity and difference and supports children with their individual

needs. It is likely that in such inclusive schools more energy and resources are

spent on interventions that are designed to reduce behaviour difficulties.

Adopting a more inclusive approach in schools and implementing programmes

such as PATHS, which is designed to help reduced behaviour difficulties and

has been noted as being effective with in the UK context (Curtis & Norgate

2007) would be particularly beneficial.

312

Within secondary schools, attending schools within urban locations are

beneficial for children considered at high risk (in terms of poor academic

achievement). This variable acts as a protective factor, and has a positive effect

in reducing the display of behaviour difficulties. It is therefore beneficial for

children considered at high risk to attend a school in an urban area. Although

this is a key finding within the present study, moving a child to an urban school

on the basis of their ‘at risk’ status is not a feasible option. Instead investigating

the characteristics of urban schools which are distinct from rural schools and

implementing these in the rural schools would be beneficial for high risk

children. It may be within such urban schools there are greater levels of

resources available where high risk individuals can be supported best. In such

schools more energy and resources may be spent on interventions that are

designed to reduced behaviour difficulties. Rural secondary schools should pay

particular attention to this finding and increase the resources available on

intervention programmes such as PATHS, which is designed to help reduce

behaviour difficulties as this would help to reduce the difficulties displayed.

8.2.4 Summary statements of key implications

Research questions 1

1. Schools need to be aware that their characteristics have an influence

on pupil level behaviour difficulties. School level interventions should

therefore be implemented.

2. Interventions that specifically target children who are male, eligible for

FSM and in years 5 in primary and 7 in secondary will be the most

effective. In primary schools those children born in the autumn and

who are older in the school year, would benefit most from these

interventions.

313

3. Schools should provide interventions to all children that aim to reduce

bullying, as this will have positive influences on pupil behaviour.

4. Specific interventions at the primary school level should include

improving pupils’ relationships with peers and teachers and targeting

overall school level academic achievement.

5. In secondary schools a focus should be placed on keeping school

size to a minimum and individual attendance and academic

achievement to a maximum.


6. Interventions need to account for multiple risks and reduce risk

whenever it can be changed.

7. Targeting those at high levels of risk, will diminish the disproportional

effect that increasing risk has upon behaviour difficulties displayed

8. When it is known which risks have the most powerful influence on

behaviour difficulties these should be targeted first as this will have

the biggest impact on reducing behaviour difficulties.


9. In primary schools overall academic achievement in schools should

be promoted in all schools, and inclusivity encouraged, as these will

have a particularly positive effect on pupil behaviour.

10. Rural secondary schools should spend more on interventions

particularly tailored at high risk children to reduce their behaviour

difficulties.

314

8.3 Contribution to Knowledge

The present study has provided a unique and significant contribution to

knowledge advancing the evidence base surrounding risk, promotive and

protective factors for behaviour difficulties in children with SEND. Firstly the

present study has set out at its inception, a name for the concept under

investigation i.e. behaviour difficulties, justifying why this was used over and

above other terms within the literature. Furthermore, a clear and unambiguous,

definition of behaviour difficulties has been provided: “A behaviour difficulty

refers to an individual’s behaviour that is persistently aggressive, destructive,

dishonest or disobedient, and is deemed inappropriate for their age or cultural

background. These behaviours will have negative affects upon themselves,

others or property, and may occur solely within a unique or across multiple

contexts”. In providing such a definition the present study has clearly set out

what this term signifies and how it is conceptualised, and in offering a definition

at the outset, misconceptions and any ambiguity that this term carries can be

discarded. Theoretical advances can be made within this field when

researchers clearly and consistency state the definition of their concept.

A major contribution to knowledge within the present study was its focus on

children with SEND. Despite representing a significant proportion of the school

population, (approximately one fifth of all school pupils, DfE 2011), there

remains a dearth of study with this under researched group. SEND pupils all

have at least one thing in common, that additional support is being offered to

them, and in utilising such a population in research, the resulting implications

are relevant not only to large numbers of children but to a range of

professionals who are currently working with them. To my knowledge no study

has specifically investigated risk, promotive and protective factors for behaviour

difficulties with a SEND population within England. The present study therefore

offers a significant contribution to knowledge.

The present study has also been able to distinguish between risk, promotive

and protective effects within the same study which offers an important

contribution to knowledge. Often within a single study only the influences of

315

either risk factors or promotive/protective factors are noted. It is unusual for a

study to acknowledge all types of influences within the same piece of research,

so the present study is able to add to the literature in this regard.

The present study has also been able to contribute to the field of explaining risk,

promotive and protective factors for behaviour difficulties across multiple

ecological levels (i.e. individual and school levels). In adopting Bronfenbrenner’s

(1979) Ecological Systems Theory as a framework for organising variables

across the individual and school level this is consistent with Ribeaud & Eisner

(2010) who argue multiple factors across distinct ecological levels should be

accounted for when attempting to explain behaviour difficulties in children and

adolescents. Relatively few studies have acknowledged multiple ecological

levels within the same study, and in this regard the present study adds to the

literature. The present study also commented on the relative importance of the

two ecological levels within the study (individual and school) assessing which

holds the greatest influence in accounting for variance in behaviour difficulties.

Few studies have explicitly acknowledged the relative influence of broad

ecological levels, so the present study contributes to the literature base in this

regard.

The present study has contributed to the field in terms of methodological

advancements by utilising the WOST, a new measure of behaviour difficulties

(Humphrey et al. 2011)50

50 Wider Outcome Survey for Teachers (Humphrey et al. 2011).

. This was a bespoke survey that was specifically

designed for children with SEND. In utilising such a measure the key issues

relevant to children with SEND can be captured. This survey has been validated

psychometrically with promising findings (Wiglesworth, Oldfield & Humphrey

2012), suggesting it is an appropriate tool to use when assessing behaviour

difficulties in children with SEND. The present study is one of the first to use the

measure. As such it is able to provide a significant contribution to the field of

behaviour difficulties in children with SEND as provides the first

psychometrically sound measure of behaviour difficulties that is specifically

tailored to this group.

316

In adopting a longitudinal design the present study was able to establish a

number of variables as true risk factors. Such a design is deemed important as

a risk factor can only be defined as such when it is significantly related to

outcome as well as preceding it (Offord & Kraemer 2000). A longitudinal design

is therefore essential to truly classify a variable as a risk factor (Murray et al.

2009). As a number of studies investigating ‘risk factors’ have adopted a cross

sectional design they are only really highlighting correlates of behaviour

difficulties. The present study has contributed to this field by using an

appropriate design to establish true risk factors. The design of this study also

used an analytical technique called multi-level modelling (see section 5.9 for an

overview). Using such a technique has enabled researchers to accurately

proportion variance across different ecological levels and is essential in

analysing data within educational studies where pupils are found nested within

schools. This is a fairly new technique and as such the present study is one of

the first to use it to highlight risk, promotive and protective factors for behaviour

difficulties and the first to do so with a SEND population.

Evidence clearly shows that risk factors for behaviour difficulties do not work in

isolation (Flouri & Kallis 2007). These factors accumulate within individuals to

impact the behaviour displayed. Failing to acknowledge how risk factors may

work together will therefore offer an incomplete account of how risk factors

influence behaviour difficulties. The present study has been able to contribute to

knowledge here in assessing the cumulative risk hypothesis, and that increased

risk leads to increased behaviour difficulties. This is the first study using a

SEND population to do this. Only a handful of research studies have assessed

whether increasing risk within an individual’s background leads to a linear or

nonlinear increase in problem behaviours, and no study has conducted such an

analysis with a SEND population. In assessing these effects the present study

found that a quadratic effect best fits the data and that as risk increases there is

a disproportional increase in behaviour difficulties displayed. This has therefore

provided a significant contribution to knowledge in the field.

317

A further analysis within the present study was to discover school protective

factors, these are influences that affect all pupils attending the school, although

are particularly beneficial for those considered at high risk, in reducing their

display of behaviour difficulties. In conducting such an analysis an important

contribution to knowledge is offered as very few studies have explicitly

acknowledged school level protective factors and no study has done so with a

SEND population.

The present study not only had a particular focus upon school level effects but

also distinguished between primary and secondary schools. This decision was

taken on a number of grounds due to differences in school size (building and

pupil numbers), teachers and mix of pupils (see section 5.9.2 for an overview).

The variables that influence behaviour difficulties at the school level are

therefore likely to be different across school type, and this was in fact what the

present study found. Previous research when acknowledging a school effect in

accounting for behaviour difficulties has often made no comparison between

primary and secondary schools, and generally focuses on a single school type.

By acknowledging differences between school types and how variables within

this ecological level affect behaviour difficulties a significant contribution to

knowledge has been made.

The present study is also unique in the literature in utilising a large and

representative sample of SEND pupils across England. Participants were drawn

from schools within 10 local authorities deemed representative of England in

terms of geographical location, proportion of ethnic minorities and population

density (DCSF 2009b). Within each local authority primary and secondary

schools were selected which again were representative of the area including

those in affluent as well as deprived areas. The total number of schools

included within the present study was 305 (248 primary and 57 secondary).

Pupils in years 1, 5, 7, and 10 from these schools who were on the register as

having SEND became the participants for the study. The final sample can be

deemed representative of SEND pupils found nationally. The study was also

fairly large with a final sample of 4288 pupils, (2660 in primary schools, and

1628 in secondary schools), which is one of the largest studies of its kind.

318

A further contribution to knowledge offered by the present study is its focus on a

wide range of variables within the same study. A number of these have been

previously highlighted as important factors in accounting for behaviour

difficulties (i.e. gender, academic achievement, attendance etc.), although

others are new and distinct such as positive relationships (as measured by the

WOST). Assessing the influence of established predictors of behaviour

difficulties from the general population as well as potentially ‘new’ predictors

within the same study, and particularly for children with SEND has not been

conducted previously.

The present study has offered a further significant contribution by maintaining a

particular focus on school level effects. There has been significantly less

research investigating school as an important ecological level compared with

family influences on child development (Bronfenbrenner 1994), therefore the

present study may be useful to the field. Acknowledging the importance of a

school level has uncovered risk and promotive factors that are present at this

level, and have a significant influence on an individual’s behaviour.

As highlighted above the present study has provided a significant contribution to

knowledge and has furthered our understanding of why children and

adolescents develop behaviour difficulties. This study was unique in its explicit

focus on children with SEND which represent a substantial proportion of the

school population. With a sampling methodology deemed representative of

England there is real power in the study’s ability to generalise its findings. A

particular focus was placed upon the school and individual ecological levels,

and splitting up school in to primary and secondary models to uncover

significant variables was a unique feature of the research. Assessing the

importance of a number of established as well as ‘new’ variables in accounting

for behaviour problems has added to the literature base. Utilising a longitudinal

design enabled true risk factors to emerge, and analysing the data using MLM

allowed the school level effect to be appropriately acknowledged. Testing risk,

promotive and protective factors within the same study for children with SEND

is unique in the literature. The present study has therefore informed the debate

around risk, promotive and protective factors for behaviour difficulties providing

319

up-to-date research into this field, and suggests interventions that could be put

in place to combat these problems.

In summary the present study has contributed to knowledge in theoretical,

methodological and practical terms:

Theoretical

1. Clearly defining and justifying the concept of ‘behaviour difficulties’,

clearing up any misconceptions or ambiguity within the field.

2. Focusing specifically on children with SEND: these children represent a

large but often under researched group of children

3. Being the first study to investigate risk, promotive and protective factors

for behaviour difficulties in children with SEND within England.

4. Distinguishing between risk, promotive and protective effects of predictor

variables within the same study.

5. Accounting for risk, promotive and protective factors for behaviour

difficulties across multiple ecological levels, and assessing the relative

importance of the individual and school ecological levels.

Methodological

6. Using a new psychometrically sound measurement of behaviour

difficulties (i.e. WOST), specially designed for children with SEND,

therefore encouraging methodological advancements within this field.

7. Adopting a longitudinal design, allowing true risk factors to emerge.

8. Using multi-level modelling to analyse the data, and therefore accounting

for the inherent clustering of data which are present in many educational

based studies.

9. Assessing the effects of cumulative risk within a SEND population, and

whether increasing risk within an individual’s background leads to a

linear or disproportional increase in problem behaviours.

10. Testing for interaction effects at the school level, in order to highlight

salient school level protective factors for different school types

320

Practical

11. Being one of the biggest studies of its kind, incorporating a large and

representative sample of SEND pupils across England, including children

in different years and at different school types, increasing the

generalisablity of the study.

12. Incorporating a wide range of variables within the same study, including

‘established’ predictors of behaviour difficulties from the general

population, as well as potentially ‘new’ predictors particularly relevant for

children with SEND.

13. Having a particular focus on school level effects, uncovering school level

protective factors, that are particularly beneficial for those considered at

high risk.

14. Offering an important contribution to the where intervention programmes

should be directed and who would benefit to the greatest degree from

these.

321

8.4 Summary of study

In summary the present study aimed firstly to highlight the salient risk and

promotive factors for behaviour difficulties in children with SEND across the

individual and school ecological levels. Secondly, the study aimed to assess the

cumulative effects of risk factors, in terms of whether increasing risk is

predictive of increased behaviour problems and whether this relationship

conforms to a linear or nonlinear pattern. Finally, the study aimed to highlight

any school level protective factors that are particularly beneficial to children

considered at high risk in reducing their behaviour difficulties.

These aims were achieved by collecting baseline data on 20 distinct variables,

(9 at the school level and 11 at the individual level), as well as a measure of

behaviour difficulties assessed by teachers and measured at baseline and

follow-up 18 months later. Analysis was carried out using MLM, with the

outcome being ‘follow-up’ behaviour difficulties, controlling for ‘baseline’

behaviour difficulties. Significant risk and promotive factors at the individual and

school level were acknowledged. A cumulative risk score was generated on the

basis of the significant contextual risks assessed. Finally an analysis involving

school level factors in interacting with risk was taken to establish some school

level protective factors.

Results revealed that risk factors for behaviour difficulties across all schools

were, being male, eligible for FSM and being a bully. In primary schools

additional significant risk factors included having poorer positive relationships,

being born in the autumn months, being in year 5 and attending a school with

low levels of academic achievement. In secondary schools additional risk

factors included having low overall attendance, lower individual level academic

achievement, being in year 7, being a bystander to bullying and attending a

school with a greater number of children on role. School level factors accounted

for 15% of total variance in behaviour difficulties in primary schools and 13% in

secondary schools, with the remainder being at the individual level.

322

The cumulative risk model for both primary and secondary school revealed

increasing risk regardless of its exact nature resulted in increased behaviour

difficulties. Type of risk was however, still important, as when risks were added

independently to a statistical model they accounted for more variance then the

cumulative risk model. Nonetheless the importance of the cumulative risk model

was in showing that the relationship between increasing cumulative risk and

behaviour difficulties was not proportional and those with higher levels of risks

present are at a disproportional disadvantage.

Finally, protective factors were assessed at the school level revealing that in

primary schools, attending a school with high academic achievement is

particularly beneficial for children considered at high risk (in terms of having

poor positive relationships). Attending schools with high levels of SEND children

on role is also beneficial for children considered at high risk (in terms of having

poor positive relationships). In secondary schools pupils attending urban

schools were at an advantage in terms of reducing their behaviour difficulties

when considered at high risk (i.e. having poor academic achievement)

compared with their rural counterparts.

Despite a number of limitations within this study, including the large attrition

rates, results being potentially influenced by the AfA intervention, and some

concerns about the true representativeness of sample, the findings are

generalisable to all children with SEND in England. It can be concluded

therefore that this study offers a significant contribution to knowledge in terms of

understanding to a greater extent the risk, promotive and protective factors that

influence behaviour difficulties of children with SEND.

Behaviour difficulties are ultimately affected by multiple factors, therefore

addressing any one of these potential influences will be beneficial in reducing

the problems displayed (Rutter 2000). It is important that future studies continue

to move beyond a uniquely individual variable focus and acknowledge

environmental influences within families schools and communities that promote

positive adaption, (Olsson et al. 2003). Interventions designed to reduce such

problems should not only attempt to reduce risk, but where it is not possible to

323

prevent certain adversities or when risks are unknown, also encourage

protection (Masten & Reed 2002, Van der Laan, et al. 2010), breaking chain

reactions of negative life events, and promoting positive adaption, (Olsson et al.

2003, Colman & Hagell 2007, Murray 2003, Rutter 1999). These interventions

need to be both context and outcome specific (Rutter 2000), helping those

currently struggling as well as others to follow adaptive pathways that lead away

from these problems. Interventions will be of the greatest benefit if they promote

positive adaption, and focus on protective factors, rather than treating problems

once they have already occurred, (Luthar 2006).


• Implications: This section highlighted the key implications that have

emerged from the study, within research question 1, 2 and 3.

• Contribution to knowledge: This section has set out multiple reasons why

the present study has offered a significant contribution to knowledge

• Summary of study: This is the final section of the thesis and gives a

broad overview of the study.

324

REFERENCES

Achenbach, T., McConaughy, S., & Howell, C. (1987). Child/adolescent behavioral and emotional problems: Implications of cross‐informant correlations for situational specificity. Psychological Bulletin, 101, 213‐232.

Achenbach, T. M. (1991a). Manual for the Child Behaviour Checklist/4-18 and 1991, profile. Burlington, VT, USA: University of Vermont, Department of Psychiatry.

Achenbach, T. M. (1991b). Manual for the Teacher's Report Form and 1991 Profile. Burlington, VT, USA: University of Vermont Department of Psychiatry.

Achenbach, T., M., Dumenci, L., & Rescorla, L., A. (2002). Ten-year comparisons of problems and competencies for nation samples of youth: self, parent and teacher reports. Journal of Emotional and Behavioural Disorders, 10, 194-203.

Achenbach, T., M., Dumenci, L., & Rescorla, L., A. (2003). Are American children’s problems still getting worse? A 23-year comparison. Journal of Abnormal Child Psychology, 31, 1-11.

Ackerman, B. P. Izard, C. E., Schoff, K., Youngstrom, E. A., & Kogos, J. (1999). Contextual risk, caregiver emotionality, and problem behaviors of six and seven year old children from economically disadvantaged families. Child Development. 70, 1415-1427.

Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpreting interactions. London: Sage.

American Psychiatric Association. (2000). Diagnostic and statistical manual of mental disorders (4th ed., text revision). Washington, DC: American Psychiatric Association.

Andersson, H. W., Bjørngaard, J. H., Kaspersen, S. L., Wang, C. E. A, Skre, I., & Dahl, T. (2010). The effects of individual factors and school environment on mental health and prejudiced attitudes among Norwegian adolescents. Social Psychiatry & Psychiatric Epidemiology, 45, 569–77.

Appleyard, K., Egeland, B., van Dulmen, M. H. M., & Sroufe, L. A. (2005). When more is not better: the role of cumulative risk in child behavior outcomes. Journal of Child Psychology & Psychiatry, 46, 235–45.

Armstrong, D. & Squires, G. (2012). Contemporary Issues in Special Educational Needs: considering the whole child. Maidenhead: Open University Press.

Arseneault, L., Bowes, L., Shakoor, S. (2010) Bullying victimization in youths and mental health problems: ’Much ado about nothing?’ Psychological Medicine, 40, 717-729.

Arsenault, L., Moffitt, T.E., Caspi, A., Taylor, A., Rijsdijk, F.V., Jaffee, S.R., Ablow, J.C., & Measelle, J.R. (2003). Strong genetic effects on cross-situational antisocial behaviour among 5- year-old children according to mothers, teachers, examiner-observers, and twins’ self-reports. Journal of Child Psychology & Psychiatry, 44, 832-848.

325

Atzaba-Poria, N., Pike, A., & Deater-Deckard, K. (2004). Do risk factors for problem behaviour act in a cumulative manner? An examination of ethnic minority and majority children through an ecological perspective. Journal of Child Psychology & Psychiatry, 45, 707–18.

Aveyard, P., Markham, W. a, & Cheng, K. K. (2004). A methodological and substantive review of the evidence that schools cause pupils to smoke. Social Science & Medicine, 58, 2253–65.

Ayers, H., Clarke, D., & Murray, A. (2000). Perspectives on behaviour: A practical guide to effective interventions for teachers. London: David Fulton Publishers.

Baillargeon, R. H., Zoccolillo, M., Keenan, K., Côté, S., Pérusse, D., Wu, H.-X., Boivin, M., et al. (2007). Gender differences in physical aggression: A prospective population-based survey of children before and after 2 years of age. Developmental Psychology, 43, 13–26.

Baker, J. A., Grant, S., & Morlock, L. (2008). The teacher-student relationship as a developmental context for children with internalizing or externalizing behavior problems. School Psychology Quarterly, 23, 3–15.

Barnes, J., Belsky, J., Broomfield, K. A., & Melhuish, E. (2006). Neighbourhood deprivation, school disorder and academic achievement in primary schools in deprived communities in England. International Journal of Behavioral Development, 30, 127–136.

Barnow, S., Lucht, M., Freyberger, H. J. (2005). Correlates of aggressive and delinquent conduct problems in adolescence. Aggressive Behavior, 1, 24-39.

Barth, J., Dunlap, S. T., Dane, H., Lochman, J. E., & Wells, K. C. (2004). Classroom environment influences on aggression, peer relations, and academic focus. Journal of School Psychology, 42, 115–133.

Beam, M. R., Gil-rivas, V., Greenberger, E., & Chen, C. (2002). Adolescent Problem Behavior and Depressed Mood : Risk and Protection Within and Across Social Contexts. Journal of Youth & Adolescence, 31, 343–357.

Berger, K. (2011). The developing person through the lifespan. New York: Worth Publishers.

Berk, L. E. (2009). Child Development. London: Pearson Education

Beyers, J. M., Loeber, R., Wikström, P. O., & Stouthamer-Loeber, M. (2001). What predicts adolescent violence in better-off neighborhoods? Journal of Abnormal Child Psychology, 29, 369–81.

Bierderman, J., Milberger, S., Faraone, S. V., Kiely, K., Guite, J., Mick, E., Ablon, S., Warburton, R., & Reed, E. (1995). Family-environment risk factors for attention deficit hyperactivity disorder: A test of Rutter’s indicators of adversity. Archives of General Psychiatry, 52, 464-470.

Bisset, S., Markham, W. A., & Aveyard, P. (2007). School culture as an influencing factor on youth substance use. Journal of Epidemiology & Community Health, 7, 485–490.

326

Bond, L., Butler, H., Thomas, L., Carlin, J., Glover, S., Bowes, G., & Patton, G. (2007). Social and school connectedness in early secondary school as predictors of late teenage substance use, mental health, and academic outcomes. The Journal of Adolescent Health: Official Publication of the Society for Adolescent Medicine, 40, 357.e9–18.

Bongers, I. L., Koot, H. M., van der Ende, J., & Verhulst, F. C. (2004). Developmental trajectories of externalizing behaviors in childhood and adolescence. Child Development, 75, 1523–37.

Book, A., Starzyk, K., Quinsey, V., 2001. The relationship between testosterone and aggression: a meta-analysis. Aggression & Violent Behavior, 6, 579–599.

Bowes, L., Arseneault, L., Maughan, B., Taylor, A., Caspi, A., & Moffitt, T. E. (2009). School, neighborhood, and family factors are associated with children’s bullying involvement: a nationally representative longitudinal study. Journal of the American Academy of Child & Adolescent Psychiatry, 48, 545–53.

Boyd, D., & Bee H. (2009). Lifespan development. London: Pearson Education.

Bradley, R. H., & Corwyn, R. F. (2002). Socioeconomic status and child development. Annual Review of Psychology, 53, 371–99.

Breslau, N., Breslau, J., Miller, E., & Raykov, T. (2011) . Behavior problems at ages 6 and 11 and high school academic achievement: Longitudinal latent variable modelling. Psychiatry Research, 185, 433–437.

British Educational Research Association. (2011). Ethical guidelines for educational research. London: British Educational Research Association.

British Psychological Society. (2009). Code of ethics and conduct: Guidance published by the Ethics Committee of the British Psychological Society. Leicester: The British Psychological Society.

Bronfenbrenner, U. (1979). The ecology of human development: experiments by nature and design. London: Harvard University Press.

Bronfenbrenner, U. (1994). Ecological models of human development. In International Encyclopaedia of Education, Vol 3, 2nd edition. Oxford: Elsevier.

Bronfenbrenner, U. (2005). Making human beings human: biological perspectives on human development. London: Sage.

Brooks-Gunn, J., & Duncan, G. J. (1997). The Effects of Poverty on Children. Future of Children, 7, 55-71.

Brooks-Gunn, J., Duncan, G. J., Leventhal, T., & Aber, J. L. (1997). Lessons Learned and Future Directions for Research on the Neighborhoods in which Children Live. In J. Brooks-Gunn, G. J. Duncan, & J. L. Aber (Eds.), Neighborhood Poverty, Vol 1. Context and Consequences for Children (pp279-297). New York: Russell Sage Foundation.

Brown, W. (1982). Classroom climate: possible effects of special needs on the mainstream. Journal for Special Educators, 19, 20–27.

Brown, M. & Schoon, I. (2008). Child Behaviour. In K. Hansen, E. Jones, H. & Joshi (Eds.), Millennium Cohort Study Fourth Survey : A User’s Guide to Initial Findings (p165-176). London: Centre for Longitudinal Studies.

327

Bryman, A. (2012). Social Research Methods. Oxford: Oxford University Press.

Bulman, W. (2012). Exploring associations between classroom relationships and learning for children with Autism Spectrum Disorders and Behavioural, Emotional and Social Difficulties. Unpublished thesis. University of Manchester: School of Education.

Buyse, E., Verschueren, K., Doumen, S., Damme, J. V., & Maes, F. (2008). Classroom problem behavior and teacher-child relationships in kindergarten : The moderating role of classroom climate. Journal of School Psychology, 46, 367 – 391.

Calkins, S. D., Blandon, A. Y., Williford, A. P., & Keane, S. P. (2007). Biological, behavioural, and relational levels of resilience in the context of risk for early childhood behaviour problems. Development & Psychopathology, 19, 675-700.

Call, K. T., & Mortimer, J. T., (2001). Arenas of comfort in adolescence: A study of adjustment in context. London: Lawrence Erlbaum Associates

Campbell, S. B., Shaw, D. S., & Gilliom, M. (2000). Early externalizing behavior problems: toddlers and preschoolers at risk for later maladjustment. Development & Psychopathology, 12, 467-488.

Ceballo, R., Ramirez, C., Hearn, K. D., & Maltese, K. L. (2003). Community violence and children’s psychological well-being: does parental monitoring matter? Journal of Clinical Child & Adolescent Psychology, 32, 586-592.

Chang, L., Schwartz, D., Dodge, K., & McBride-Chang, C. (2003). Harsh parenting in relation to child emotion regulation and aggression. Journal of Family Psychology, 17, 598–606.

Chaplain, R. (2003). Teaching without disruption in the primary school: A model for managing pupil behaviour. London: RoutledgeFalmer.

Chou, L.-C., Ho, C.-Y., Chen, C.-Y., & Chen, W. J. (2006). Truancy and illicit drug use among adolescents surveyed via street outreach. Addictive Behaviors, 31, 149–54.

Claes, E., Hooghe, M., & Reeskens, T. (2009). Truancy as a contextual and school‐related problem: a comparative multilevel analysis of country and school characteristics on civic knowledge among 14 year olds. Educational Studies, 35, 123–142.

Cohen, J. (1992). A power primer. Psychological Bulletin, 112, 155-159.

Cohen, L., Manion, L., & Morrison, K. (2011). Research methods in education. London: Routledge.

Coie, J. D., Watt, N. F., West, S. G., Hawkins, J. D., Asarnow, J. R., Markman, H. J., Ramey, S. L., et al. (1993). The science of prevention. A conceptual framework and some directions for a national research program. The American Psychologist, 48, 1013–22.

Coleman, J., & Hagell, A. (2007). The nature of risk and resilience in adolescence. In J. Coleman, & A. Hagell (Eds). Adolescence risk and resilience: Against the odds (pp2-16). Chichester: John Wiley & Sons.

328

Colman, I., Murray, J., Abbott, R., Maughan, B., Kuh, D., Croudace, T., & Jones, P. (2009). Outcomes of conduct problems in adolescence: 40 year follow-up of national cohort. British Medical Journal, 338, 1-10.

Cole, T. (2003) ‘The future: new name – same aims’, SEBDA News. Penrith: The Social Emotional and Behavioural Difficulties Association.

Collishaw, S., Maughan, B., Goodman, R., & Pickles, A. (2004). Time trends in adolescent mental health. Journal of Child Psychology & Psychiatry, 45, 1350-1362.

Collishaw, S., Goodman, R., Pickles, A., & Maughan, B. (2007) Modelling the contribution of changes in family life to time trends in adolescent conduct problems. Social Science & Medicine, 65, 2576-2587.

Connor, D. F. (2004). Aggression and antisocial behaviour in children and adolescents: Research and treatment. New York: The Guilford Press.

Cooper, P., Smith, G., & Upton, C. J., (1994). Emotional and behavioural difficulties: theory to practice. London: Routledge.

Cooper, P. (1996). Giving it a name: The value of descriptive categories in educational approaches to emotional and behavioural difficulties. Support for Learning, 11, 146-150.

Cooper, P. (1999a). What do we mean by emotional behavioural difficulties? In P. Cooper (Ed). Understanding and supporting children with emotional and behavioural difficulties (pp9-12). London: Jessica Kingsley Publishers.

Cooper, P. (1999b). Changing perceptions of EBD: Maladjustment, EBD and beyond. Emotional & Behavioural Difficulties, 4, 3-11.

Crick, N. R., & Zahn-Waxler, C. (2003). The development of psychopathology in females and males: Current progress and future challenges. Development & Psychopathology, 15, 719–742.

Criss, M. M., Pettit, G. S., Bates, J. E., Dodge, K. A., & Lapp, A. L. (2002). Family Adversity, Positive Peer Relationships, and Children’s Externalizing Behavior: A Longitudinal Perspective on Risk and Resilience. Child Development, 73, 1220–1237.

Crotty, M. (1998). The Foundations of Social Research: Meaning and Perspective in the Research Process. London: Sage.

Curtis, C., & Norgate, R. (2007). An Evaluation of the Promoting Alternative Thinking Strategies Curriculum at Key Stage 1. Educational Psychology in Practice, 23, 33–44.

Daniels, H., Visser, J., Cole, T., De Reybekill, N. (1999). Emotional and Behavioural Difficulties in Mainstream Schools. Research Report No 90. London: DfEE.

Daniel, L. G. & King, D. A. (1997) Impact of inclusion education on academic achievement, student behaviour and self-esteem, and parental attitudes, Journal of Educational Research, 91, 67–80.

Darke, S., Ross, J., & Lynskey, M. (2003). The relationship of conduct disorder to attempted suicide and drug use history among methadone maintenance patients. Drug & Alcohol Review, 22, 21-25.

329

DCSF, (2008a). The Education of Children and Young People with Behavioural, Emotional and Social Difficulties as a Special Educational Need. Nottingham: DCSF Publications.

DCSF, (2008b). Bullying involving children with special educational needs and disabilities. Safe to learn: embedding anti-bullying work in schools. Nottingham: DCSF Publications.

DCSF. (2009a). Achievement for All: Guidance for Schools. Nottingham: DCSF Publications.

DCSF. (2009b). Achievement for All: Local Authority Prospectus. Nottingham: DCSF Publications.

Deater-Deckard, K., Dodge, K. A, Bates, J. E., & Pettit, G. S. (1998). Multiple risk factors in the development of externalizing behavior problems: group and individual differences. Development & Psychopathology, 10, 469–93.

Dekovic, M. (1999). Risk and Protective Factors in the Development of Problem Behavior During Adolescence. Journal of Youth & Adolescence, 28, 667–685.

Deković, M., Wissink, I. B., & Marie Meijer, A. (2004). The role of family and peer relations in adolescent antisocial behaviour: comparison of four ethnic groups. Journal of Adolescence, 27, 497–514.

Delfos, M. F. (2004). Children and behavioural problems: Anxiety, aggression, depression and ADHD, a biopsychogical model with guidelines for diagnostics and treatment. London: Jessica Kingsley Publishers.

Denny, S. J., Robinson, E. M., Utter, J., Fleming, T. M., Grant, S., Milfont, T. L., Crengle, S., et al. (2011). Do schools influence student risk-taking behaviors and emotional health symptoms? The Journal of Adolescent Health, 48, 259–67.

DES, (1978). Special Educational Needs (The Warnock Report). London: HMSO.

DES, (1989). Discipline in Schools: Report of the Committee of Enquiry Chaired by Lord Elton, ‘The Elton Report’. London: HMSO.

DeWit, D. J., Offord, D. R., Sanford, M., Rye, B. J., Shain, M., & Wright, R. (2000). The effect of school culture on adolescent behavioural problems: self-esteem, attachment to learning, and peer approval of deviance as mediating mechanisms. Canadian Journal of School Psychology, 16, 15–38.

DfE, (1994). Circular 9/94, Pupils with Problems: the Education of Children with Emotional and Behavioural Difficulties. London: HMSO.

DfES (2001a). Special Educational Needs: code of practice. Nottingham: DfES Publications.

DfES (2001b). Special Education Needs and Disabilities Act (SENDA). London: HMSO

DfES (2003). Data collection by type of Special Educational Need. Nottingham: DfES Publications.

DfES (2005). Excellence and enjoyment: social emotional aspects of learning. Nottingham: DfES Publications.

330

DfE. (2010a). Schools, pupils and their characteristics. Nottingham. DfE Publications.

DfE. (2010b). National curriculum assessments at Key Stage 2 in England, 2009/10 (revised). Nottingham. DfE Publications.

DfE. (2010c). Pupil absence in schools in England, autumn term 2009 and spring term 2010. London: Department for Education Publications.

DfE. (2010d). Children with special needs 2010: an analysis. Nottingham. DfE Publications.

DfE (2011a). Special Educational Needs Information Act: An Analysis 2011. Nottingham. DfE Publications.

DfE. (2011b). Permanent and Fixed Period Exclusions from Schools in England 2009/10. London: Department for Education Publications.

DfE (2011c). Support and aspiration: A new approach to special educational needs and disability. London: Department for Education Publications

DfE (2012). Permanent and Fixed Period Exclusions from Schools and Exclusion Appeals in England 2010/11. London: Department for Education.

Dodge, K. A, Lansford, J. E., Burks, V. S., Bates, J. E., Pettit, G. S., Fontaine, R., & Price, J. M. (2003). Peer rejection and social information-processing factors in the development of aggressive behavior problems in children. Child Development, 74, 374–393

Dodge, K. A., & Pettit, G. S. (2003). A biopsychosocial model of the development of chronic conduct problems in adolescence. Developmental Psychology, 39, 349–371.

Domina, T. (2005). Leveling the Home Advantage: Assessing the Effectiveness of Parental Involvement in Elementary School. Sociology of Education, 78, 233–249.

Donnellan, M. B., Trzesniewski, K. H., Robins, R. W., Moffitt, T. E., & Caspi, A. (2005). Self-Esteem Antisocial Behavior, Aggression and Delinquency. Psychological Science, 16, 328–335.

D’Onofrio, B. M., Goodnight, J. A.,Van Hulle, C. A., Rodgers, J. L., Rathouz,P. J., Walderman, I. D., & Lahey, B. B. (2009). A Quasi-Experimental Analysis of the Association Between Family Income and Offspring Conduct Problems. Journal of Abnormal Child Psychology, 37, 415-429.

Doumen, S., Verschueren, K., Buyse, E., Germeijs, V., Luyckx, K., & Soenens, B. (2008). Reciprocal Relations Between Teacher – Child Conflict and Aggressive Behavior in Kindergarten : A Three- Wave Longitudinal Study. Journal of Clinical Child & Adolescent Psychology, 37, 588–599.

Drukker, M., Kaplan, C., Feron, F., & Os, J. V. (2003). Children’ s health-related quality of life, neighbourhood socio-economic deprivation and social capital . A contextual analysis. Social Science & Medicine, 57, 825–841.

Durlak, J. A., Weissberg, R. P., Dymnicki, A. B., Taylor, R. D., & Schellinger, K. B. (2011).The Impact of Enhancing Students’ Social and Emotional Learning: A Meta-Analysis of School-Based Universal Interventions. Child Development, 82, 405–432.

331

Eccles, J. S., Barber, B. L., Stone, M., & Hunt, J. (2003). Extra-curricular activities and adolescent development. Journal of Social Issues, 59, 865-889.

Education Act (1981). London: HMSO.

Education Act (1996). London: HMSO.

Edwards, B., & Bromfield, L. M. (2010). Neighbourhood influences on young children’s emotional and behavioural problems. Family Matters, 84, 7–19.

Ehrensaft, M. K. (2005). Interpersonal relationships and sex differences in the development of conduct problems. Clinical Child & Family Psychology Review, 8, 39–63.

Eisenberg, N., Cumberland, A., Spinrad, T. L., Fabes, R., Shepard, S., Reiser, M., Murphy, B. C., Losoya, S. H., & Guthrie, I. K., (2001). The relations of regulation and emotionality to children's externalizing and internalizing problem behavior. Child Development, 72, 1112-1134.

Epstein, J. a., Botvin, G. J., Griffin, K. W., & Diaz, T. (2001). Protective factors buffer effects of risk factors on alcohol use among inner-city youth. Journal of Child & Adolescent Substance Abuse, 11, 77–90.

Eriksson, I., Cater, Å., Andershed, A., & Andershed, H. (2011). What protects youths from externalising and internalising problems ? A critical review of research findings and implications for practice. Australian Journal of Guidance and Counselling, 21, 113–125.

Evans, G. W. (2003). A multimethodological analysis of cumulative risk and allostatic load among rural children. Developmental Psychology, 39, 924–933.

Evans, G. W., & Gary, W. (2004). The environment of childhood poverty. American Psychologist, 59, 77–92.

Evans, G. W., Kim, P., Ting, A. H., Tesher, H. B., & Shannis, D. (2007). Cumulative risk, maternal responsiveness, and allostatic load among young adolescents. Developmental Psychology, 43, 341–51.

Evans, R., & Pinnock, K. (2007). Promoting resilience and protective factors in the children’s fund. Journal of Children & Poverty, 13, 21–36.

Evans-Whipp, T., Beyers, J. M., Lloyd, S., Lafazia, A. N., Toumbourou, J. W., Arthur, M. W., & Catalano, R. F. (2004). A review of school drug policies and their impact on youth substance use. Health Promotion International, 19, 227–234.

Farrell, M. (2004). Special Educational Needs: A Resource for Practitioners. London: Sage.

Farrington, D. P., & Loeber, R. (2000). Some benefits of dichotomization in psychiatric and criminological research. Criminal Behavior and Mental Health, 10, 100–122.

Farris, J. R., Smith, L. E., Weed, K. (2007). Resilience and vulnerability in the context of multiple risks. In J. G. Borkowski, J. R. Farris, T. L. Whitman, S. S. Carothers, K. Weed & D. A. Keogh, (Eds.). Risk and resilience: adolescent mothers and their children grow up (pp179-204). London: Lawrence Erlbaum Associates.

332

Felson, R. B., Liska, A. E., South, S. J., & McNulty, T. L. (1994). The subculture of violence and delinquency: individual vs. school context effects. Social Forces, 73, 155-173.

Fergus, S., & Zimmerman, M. A., (2005). Adolescent resilience: A framework for understanding healthy development in the face of risk. Annual Review of Public Health, 26, 399-419.

Fergusson, M. F., & Horwood, L., J. (2003). Resilience to childhood adversity: results of a 21 year study. In S. S. Luthar (Ed). Resilience and Vulnerability, (pp130-155). Cambridge: Cambridge University Press.

Fergusson, D. M., Horwood, L.J., Ridder, E. (2005). Show me the child at seven: The consequences of conduct problems in childhood for psychosocial functioning in adulthood. Journal of Child Psychology & Psychiatry, 46, 837-849.

Fergusson, D. M., Horwood, L. J., & Ridder, E. M. (2007). Conduct and attentional problems in childhood and adolescence and later substance use, abuse and dependence: Results of a 25-year longitudinal study. Drug & Alcohol Dependence 88S, S14–S26.

Fergusson , D. M., Vitaro, F., Wanner, B., & Brendgen, M. (2007). Protective and compensatory factors mitigating the influence of deviant friends on delinquent behaviours during early adolescence. Journal of Adolescence, 30, 33-50

Field, A. (2009). Discovering statistics using SPSS. London: Sage.

Fife‐Schaw, C. (2006) Introduction to Structural Equation Modelling. In G. M. Brakewell, C. Hammond, C. Fife-Schaw, & J. Smith. (2006). Research Methods in Psychology (pp444-465). London: Sage.

Flory, K., Milich, R., Lynam, D., R., Leukefeld, C., & Clayton, R. (2003). Relation between childhood disruptive behaviour disorders and substance use and dependence symptoms in young adulthood: Individuals with symptoms of attention deficit/hyperactivity disorder and conduct disorder are uniquely at risk. Psychology of Addictive Behaviors, 17, 151-158.

Flouri, E., & Kallis, C. (2007). Adverse life events and psychopathology and prosocial behavior in late adolescence: testing the timing, specificity, accumulation, gradient, and moderation of contextual risk. Journal of the American Academy of Child and Adolescent Psychiatry, 46, 1651–9.

Flouri, E. (2008). Contextual risk and child psychopathology. Child Abuse & Neglect, 32, 913-7.

Flouri, E., Tzavidis, N., & Kallis, C. (2010). Area and family effects on the psychopathology of the Millennium Cohort Study children and their older siblings. Journal of Child Psychology & Psychiatry, 51, 152–61.

Ford, T., Goodman, R., & Meltzer, H., (2003). The British child and adolescent mental health survey 1999: The prevalence of DSM-IV disorders. Journal of the American Academy of Child & Adolescent Psychiatry, 42, 1203-1211.

Forehand, R., Biggar, H., & Kotchick, B. A. (1998). Cumulative risk across family stressors: short- and long-term effects for adolescents. Journal of Abnormal Child Psychology, 26, 119–28.

333

Fraser, M. W., & Terzian, M. A. (2005). Risk and Resilience in Child Development: Principles and Strategies of Practice. In G. P. Mallon & P. McCartt-Hess (Eds). Child Welfare for the 21st Century: A handbook of Practices, Policies and Programs, (p55-71). Chichester: Columbia University Press.

Frederickson, N., & Cline, T. (2009). Special educational needs, inclusion and diversity. Maidenhead: Open University Press.

Garmezy, N., Masten, A. S., & Tellegen, A. (1984). The study of stress and competence in children: a building block for developmental psychopathology. Child Development, 55, 97-111.

Garmezy, N. (1996). Reflections and commentary on risk, resilience and development. In R. J. Haggerty, L. R. Sherrod, N., Garmezy, & M. Rutter. (Eds.). Stress, risk and resilience in children and adolescents: Processes, mechanisms and interventions. Cambridge: Cambridge University Press.

George, R., & Thomas, G. (2000). Victimization in Middle and High School students: A Multilevel Analysis. Educational Research, 84, 48–57.

Gerard, J. M., & Buehler, C. (1999). Multiple risk factors in the family environment and youth problem behaviors. Family Relations, 61, 343–361.

Gerard, J. M., & Buehler, C. (2004a). Cumulative Environmental Risk and Youth Problem Behavior. Journal of Marriage and Family, 66, 702–720.

Gerard, J. M., & Buehler, C. (2004b). Cumulative environmental risk and youth maladjustment: the role of youth attributes. Child Development, 75, 1832–1849.

Gilliom, M., & Shaw, D. (2004). Co-development of externalizing and internalizing problems in early childhood. Development & Psychopathology, 16, 313-333.

Gini, G. (2008). Associations between bullying behaviour, psychosomatic complaints, and emotional and behavioural problems. Paediatrics & Child Health, 44, 492–497.

Goodman R (2001) Psychometric properties of the Strengths and Difficulties Questionnaire (SDQ). Journal of the American Academy of Child & Adolescent Psychiatry, 40, 1337-1345.

Goodman, R., Gledhill, J., & Ford, T. (2003). Child psychiatric disorder and relative age within school year: cross sectional survey of large population sample. British Medical Journal, 327, 1-4

Gottfredson, D. C., & DiPietro, S. M. (2011). School size, social capital, and student victimization. Sociology of Education, 84, 69–89.

Gottfredson, D. C. (2001). Schools and delinquency. Cambridge: Cambridge University Press.

Green, H., McGinnity, A., Meltzer, H., Ford, T., & Goodman, R. (2005). Mental health of children and young people in Great Britain (2004). London: Office for national statistics.

334

Greenberg, M., Domitrovich, C., & Bumbarger, B. (2000). Preventing mental disorders in school-aged children: a review of the effectiveness of prevention programs. Pennsylvania State University: Prevention Research Centre for the Promotion of Human Development.

Greenberg, M. T., Speltz, M. L., DeKlyen, M., & Jones, K. (2001). Correlates of clinic referral for early conduct problems: variable- and person-oriented approaches. Development & Psychopathology, 13, 255–76.

Greenberg, M. T. (2006). Promoting resilience in children and youth: prevention interventions and their interface with neuroscience. Annals New York Academy of Science, 1094, 139-150.

Greenberger, E., Chen, C., Beam, M., Whang, S-M., & Dong, Q. (2000). The perceived social contexts of adolescents' misconduct: A comparative study of youths in three cultures. Journal of Research on Adolescence, 10, 369-392.

Gutman, L. M., Sameroff, A. J., & Cole, R. (2003). Academic growth curve trajectories from 1st grade to 12th grade: Effects of multiple social risk factors and preschool child factors. Developmental Psychology, 39, 777–790.

Gutman, L. M., Midgley, C. (2000). The role of protective factors in supporting the academic achievement of poor African American students during the middle school transition. Journal of Youth & Adolescence, 29, 223-248

Guttmannova, K., Szanyi, J. M., & Cali, P. W. (2007). Internalizing and externalizing behavior problem scores: Cross-ethnic and longitudinal measurement invariance of the behavior problem index. Educational & Psychological Measurement, 68, 676–694.

Hagell, A. (2007). Antisocial behaviour. In J. Colman & A. Hagell (Eds.). Adolescence, risk and resilience: Against the odds, (pp125-142). Chichester: John Wiley & Sons.

Hall, J. E., Sammons, P., Sylva, K., Melhuish, E., Taggart, B., Siraj-Blatchford, I., & Smees, R. (2010). Measuring the combined risk to young children’s cognitive development: an alternative to cumulative indices. British Journal of Developmental Psychology, 28, 219-238.

Haller, E. J. (1992). High school size and student indiscipline: Another aspect of the school consolidation issue? Educational Evaluation & Policy Analysis, 14, 145.

Halliwell, M. (2003). Supporting children with special educational needs: a guide for assistants in schools and pre-schools. London: David Fulton.

Hampel, P., Manhal, S., & Hayer, T. (2009). Direct and relational bullying among children and adolescents: Coping and psychological adjustment. School Psychology International, 30, 474–490.

Hamre, B. K., & Pianta, R. C. (2001). Early teacher – child relationships and the trajectory of children’s school outcomes through Eighth Grade. Child Development, 72, 625–638.

Hanewald, R. (2011). Reviewing the literature on “at-risk” and resilient children and young people. Australian Journal of Teacher Education, 36, 16-29.

Harris, N., Eden, K., & Blair, A. (2000). Challenges to School Exclusion: Exclusion, Appeals and the Law. London: RoutledgeFalmer.

335

Harris, M., & Butterworth, G. (2002). Developmental Psychology: A student handbook. Hove: Psychology Press Limited.

Hastings, R, P. (2002). Parental stress and behaviour problems of children with developmental disability. Journal of Intellectual & Developmental Disability. 27, 149-160.

Hawes, D. J., & Dadds, M.R., (2005). Oppositional and conduct problems. In J. Hudson & R. Rapee (Eds.), Current thinking on psychopathology and the family. New York: Elsevier.

Healey, A., Knapp, M., & Farrington, D. (2004) Adult labour market implications of antisocial behaviour in childhood and adolescence: findings from a UK longitudinal study. Applied Economics, 36, 93-105.

Hebron, J. S. (2012). Bullying among children and young people with autism spectrum disorders: an investigation into prevalence, victim role, risk and protective factors. Unpublished thesis. University of Manchester: School of Education.

Heck, R. H., Thomas, S. L. & Tabata, L. N. (2010). Multilevel and longitudinal modelling with IMB SPSS. London: Routledge.

Henry, K. L. (2007). Who’s skipping school: characteristics of truants in 8th and 10th grade. The Journal of School Health, 77, 29–35.

Henry, K. L., & Huizinga, D. H. (2007). Truancy’s effect on the onset of drug use among urban adolescents placed at risk. The Journal of Adolescent Health, 40, 358e9–358e17.

Henry, K. L., & Thornberry, T. P. (2010). Truancy and escalation of substance use during adolescence. Journal of Studies on Alcohol & Drugs, 71, 115–24.

Herrenkohl, T. I., Hill, K. G., Chung, I., Guo, J., Abbott, R. d., & Hawkins, D., (2003). Protective factors against serious violent behavior in adolescence: a prospective study of aggressive children. Social Work Research, 27, 179-191.

Hill, N. E., Castellino, D. R., Lansford, J. E., Nowlin, P., Dodge, K. A, Bates, J. E., & Pettit, G. S. (2004). Parent academic involvement as related to school behavior, achievement, and aspirations: demographic variations across adolescence. Child Development, 75, 1491–509.

Hille, E. T., den Ouden, a L., Saigal, S., Wolke, D., Lambert, M., Whitaker, A, Pinto-Martin, J. A, Hoult, L., Meyer, R., Feldman, J. F., Verloove-Vanhorick, S. P., & Paneth, N. (2001). Behavioural problems in children who weigh 1000 g or less at birth in four countries. Lancet, 357, 1641–16433.

Hinshaw, S. P. (1992). Externalising behaviour problems and academic underachievement in childhood and adolescence: causal relationships and underlying mechanisms. Psychological Bulletin, 111, 127-155.

Hobbs, G., & Vignoles, A. (2007). Is free school meal status a valid proxy for socio-economic status (in schools research)? London: Centre for the Economics of Education.

Hodkinson, A & Vickerman, P (2009). Key Issues in Special Educational Needs and Inclusion. London: Sage.

336

Hoglund, W. L., & Leadbeater, B. J. (2004). The effects of family, school, and classroom ecologies on changes in children’s social competence and emotional and behavioral problems in first grade. Developmental Psychology, 40, 533–44.

Home Office (2004) Defining and Measuring Anti-Social Behaviour. Development & Practice Report 26. London: Home Office.

Home Office (2011). Downloaded from http://www.homeoffice.gov.uk/crime/anti-social-behaviour/(1st June 2011)

Hooper, S. R., Burchinal, M. R., Erwick-Roberts, J., Zeisel, s., & Neebe, E. C. (1998). Social and family risk factors for infant development at one year: an application of the cumulative risk model. Journal of Applied Developmental Psychology, 19, 85-96.

Hope, T. L., & Bierman, K. L. (1998). Patterns of Home and School Behavior Problems in Rural and Urban Settings. Journal of School Psychology, 36, 45–58.

House of Commons (2006). Education and Skills Committee: Special Educational Needs. Third Report of Session 2005–06, Volume I. London: The Stationery Office Limited.

Howarth, R., & Fisher, P. E. (2005). Emotional and behaviour difficulties. London: Continuum.

Howitt, D., & Cramer, D. (2011). Introduction to Research Methods in Psychology. Harlow: Pearson Education.

Hoxby, C. M. (2002). The power of peers: How does the makeup of the classroom influence achievement? Education Next, 2, 57–63.

Humphrey, N., Kalambouka, A., Bolton, J., Lendrum, A., Wigelsworth, M., Lennie C., et al. (2008). Primary social and emotional aspects of learning: evaluation of small group work. Nottingham: DCSF Publications.

Humphrey, N., Squires, G., Barlow, A., Bulman, W. F. L., Hebron, J. S., Oldfield, J., et al. (2010). Achievement for All national evaluation: Interim report (RR028). London: DfE.

Humphrey, N., Squires, G., Barlow, A., Bulman, W. F. L., Hebron, J. S., Oldfield, J., et

al. (2011). Achievement for All National Evaluation: Final Report (DfE-RR176). London: DfE.

Humphrey, N, Wigelsworth, M, Barlow, A., & Squires, G. (2012). The role of school

and individual differences in the academic attainment of learners with special educational needs and disabilities: a multi-level analysis. International Journal of Inclusive Education, 1, 1–23,

Ingoldsby, E. M., Shaw, D. S., Winslow, E., Schonberg, M., Gilliom, M., & Criss, M. M. (2006). Neighborhood disadvantage, parent-child conflict, neighborhood peer relationships, and early antisocial behavior problem trajectories. Journal of Abnormal Child Psychology, 34, 303–19.

337

Jaffee, S. R., Caspi, A., Moffitt, T. E., Polo-Tomás, M., & Taylor, A. (2007). Individual, family, and neighborhood factors distinguish resilient from non-resilient maltreated children: a cumulative stressors model. Child Abuse & Neglect, 31, 231–53.

Jensen, E. (2009). Teaching with poverty in mind: What being poor does to kids' brains and what schools can do about it. Alexandria, Virginia: ASCD.

Jenson, J. M., & Fraser, M. W., (2011). A risk and resilience framework for child, youth and family policy. In J. M. Jenson & M. W. Fraser (Eds.), Social policy for children and families: a risk and resilience perspective. London: Sage.

Jessor, R., Van Den Bos, J., Vanderryn, J., Costa, F. M., & Turbin, M. S. (1995). Protective factors in adolescent problem behavior: Moderator effects and developmental change. Developmental Psychology, 31, 923-933.

Jones, D. J., Forehand, R., Brody, G., & Armistead, L. (2002). Psychosocial adjustment of African American children in single-mother families: a test of three risk models. Journal of Marriage & Family, 64, 105-115.

Jones, S., & Forshaw, M. (2012). Research Methods in Psychology. Harlow: Pearson Education.

Kalambouka, A., Farrell, P., Dyson, A. & Kaplan, I. (2005). The impact of population inclusivity in schools on student outcomes. London: University of London Institute of Education, EPPI Centre, Social Science Research Unit.

Kam, C., Greenberg, M. T., & Kusche, C. A. (2004). Sustained effects of the PATHS curriculum on the social and psychological adjustment of children in special education Journal of Emotional and Behavioral Disorders, 12, 66-78.

Kaplan, H. B. (2002). Toward an Understanding of Resilience: A critical review of definitions and models. In M. D. Glantz, & J. L. Johnson (Eds.), Resilience and development: Positive life adaptations, (pp 17-83). London: Kluwer Academic Publishers.

Kavale, K. A., Forness, S. R., & Alper, A. E. (1986). Research in behavioural disoders/emotional disturbance: A survey of subject identification criteria. Behavioral Disorders, 11, 159-167.

Kellam, S. G., Ling, X., Merisca, R., Brown, C. H., & Ialongo, N. (1998). The effect of the level of aggression in the first grade classroom on the course and malleability of aggressive behavior into middle school. Development & Psychopathology, 10, 165–85.

Kenny, D. A. (1979). Correlation and Causality. Wiley, New York.

Kerr, M. A, Black, M. M., & Krishnakumar, A. (2000). Failure-to-thrive, maltreatment and the behavior and development of 6-year-old children from low-income, urban families: a cumulative risk model. Child Abuse & Neglect, 24, 587–98.

Keyes, C. L. M. (2004). Risk and resilience in human development: an introduction. Research in Human Development, 1, 223-227.

Khoury-Kassabri, M., Benbenishty, R., Astor, R. A., & Zeira, A. (2004). The contributions of community, family, and school variables to student victimization. American Journal of Community Psychology, 34, 187–204.

338

Kim, Y. S., Leventhal, B. L., Koh,Y. J., Hubbard, A., & Boyce, W. T. (2006). School bullying and youth violence: causes or consequences of psychopathologic behavior? Archives of General Psychiatry, 63, 1035-41.

Kim-Cohen, J., Moffitt, T. E., Taylor, A., Pawlby,S. J., & Caspi, A. (2005). Maternal depression and children’s antisocial behavior: nature and nurture effects. Archives in General Psychiatry, 62, 173-181.

King, S. M., Iacono, W. G., & McGue, M. (2004). Childhood externalizing and internalizing psychopathology in the prediction of early substance use. Addiction, 99, 1548-59.

Klaff, A. C., Kroes, M., Vles, J. S., Hendriksen, J. G. M., Feron, F. J. M., Steyaert, J., van Zeben, T. M. C. B., Jolles, J., & van Os, J. (2001). Neighbourhood level and Individual level SES effects on child problem behaviour: A multilevel analysis. Journal of Epidemiology & Community Health, 55, 246-250.

Kliewer, W., Ramirez, M., Obando, P., Sandi, L., & Karenkeris, C. (2006). Violence exposure and drug use in Central American youth: Family cohesion and parental monitoring as protective factors. Adolescence, 16, 455–478.

Kohen, D., Oliver, L., & Pierre, F. (2009). Examining the effects of schools and neighbourhoods on the outcomes of kindergarten children in Canada. International Journal of Speech-Language Pathology, 11, 404–418.

Kraemer, H., Stice, E., Kazdin, A., Offord, D., & Kupfer, D. (2001). How do risk factors work together? Mediators, moderators and independent, overlapping and proxy risk factors. American Journal of Psychiatry, 158, 848-856.

Kraemer, H., Lowe, K., & Kupfer, D. J. (2005). To your health: how to understand what research tells us about risk. Oxford: Oxford University Press.

Kraemer, H., Kazdin, A. E., Offord, D. R., Kessler, R. C., Jensen, P. S., & Kupfer, D., J. (1997). Coming to Terms with the Terms of Risk. Archives of General Psychiatry, 54, 337-343.

Kuperminc, G., Leadbeater, B. J., Blatt, S. J. (2001). School Social Climate and Individual Differences in Vulnerability to Psychopathology among Middle School Students. Journal of School Psychology, 39, 141–159.

Ladd, G. W. & Burgess, K. B. (1999). Charting the relationship trajectories of aggressive, withdrawn, and aggressive/withdrawn children during early grade school. Child Development, 70, 910-929.

Lahey, B. B., Schwab-Stone, M., Goodman, S. H., Waldman, I. D., Canino, G., Rathouz, P. J., Miller, T. L., et al. (2000). Age and gender differences in oppositional behavior and conduct problems: A cross-sectional household study of middle childhood and adolescence. Journal of Abnormal Psychology, 109, 488–503.

Lahey, B. B., & Waldman, I. D. (2003). A Developmental Propensity Model of the Origins of Conduct Problems During Childhood and Adolescence. In B. B. Lahey, T. E. Moffitt, & A. Caspi, (Eds.). Causes of Conduct Disorder and Juvenile Delinquency, (pp76-117). New York: Guildford Press.

339

Laird, R. D., Jordan, K. Y., Dodge, K. A., Pettit, G. S. & Bates, J. E. (2001). Peer rejection in childhood, involvement with antisocial peers in early adolescence, and the development of externalizing behaviour problems. Development & Psychopathology, 13, 337–354.

Larsson, B., & Frisk, M. (1999). Social competence and emotional/behaviour problems in 6-16 year-old Swedish school children. European Child & Adolescent Psychiatry, 33, 24–33.

Laukkanen, E., Shemeikka, S., Notkola, I., Koivumaa-Honkanen, H., & Nissinen, A. (2002). Externalizing and internalizing problems at school as signs of health-damaging behaviour and incipient marginalization. Health Promotion International, 2, 139-146.

Leshner , A. I., (2002). Introduction. In M. D., Glantz, & J . L. Johnson (Eds.), Resilience and development: positive life adaptations (p1-3). New York: Kluwer Academic Publishers.

Leventhal, T., & Brooks-Gunn, J. (2000). The neighborhoods they live in: the effects of neighborhood residence on child and adolescent outcomes. Psychological Bulletin, 126, 309–37.

Lien, L., Tambs, K., Oppedal, B., Heyerdahl, S., & Bjertness, E. (2005). Is relatively young age within a school year a risk factor for mental health problems and poor school performance? A population-based cross-sectional study of adolescents in Oslo, Norway. BMC public health, 5, 102

Lima, J., Caughy, M., Nettles, S. M., & O’Campo, P. J. (2010). Effects of cumulative risk on behavioral and psychological well-being in first grade: moderation by neighborhood context. Social Science & Medicine, 71, 1447–54.

Little, R. J. A. (1988). A test of completely missing at random for multivariate data with missing values. Journal of the American Statistical Association, 83, 1198-1202.

Liu, J. (2004). Childhood Externalising Behavior: theory and implications. Journal of Child and Adolescent Psychiatric Nursing, 17, 99-103.

Loeber, R., Slot, W., & Stouthamer-Loeber, M., (2008). A cumulative developmental model of risk and promotive factors. In R. Loeber, W. Slot, P. van der Laan & M. Hoeve (Eds.) Tomorrow’s criminals (p133-161). Farnham: Ashgate Publishing Limited.

Long, M., Wood, C., Littleton, K., Passsenger, T., & Sheehy, K. (2011). The psychology of education. London: Routledge.

Losel, F., & Beelman, A. (2003). Effects of child skills training in preventing antisocial behavior: A systematic review of randomized evaluations. Annals of the American Academy of Political & Social Science, 587, 84–109.

Loukas, A., Prelow, H. M., Suizzo, M., & Allua, S. Mothering and peer associations mediate cumulative risk effects for Latino youth. Journal of Marriage & Family, 70, 76-85.

Luthar, S. S. (1993). Annotation: methodological and conceptual issues in research on childhood resilience. Journal of Child Psychology & Psychiatry, 34, 441–53.

340

Luthar, S. S. & Cicchetti, D. (2000). The construct of resilience: Implications for interventions and social policies. Development & Psychopathology, 12, 857-885.

Luthar, S. S., Cicchetti, D., & Becker, B. (2000a). The construct of resilience: a critical evaluation and guidelines for future work. Child Development, 71, 543–562.

Luthar, S. S, Cicchetti, D., & Becker, B. (2000b). Research on Resilience: Response to Commentaries. Child Development, 71, 573-575.

Luthar, S. S., & Cushing, G. (2002). Measurement Issues in the Empirical Study of Resilience: An Overview. In M. D. Glantz, & J. L. Johnson (Eds). Resilience and Development: Positive Life Adaptations, (pp129-160). London: Kluwer Academic Publishers.

Luthar, S. & Zelzo, L. (2003). Research on Resilience: An Integrative Review. In S. S. Luthar (Ed). Resilience and Vulnerability, (pp510-550). Cambridge: Cambridge University Press

Luthar, S. S. (2006). Resilience in development: A synthesis of research across five decades. In D. Cicchetti & D. J. Cohen (Eds.), Developmental Psychopathology, Vol III, Risk, Disorder, and Adaptation, (pp739-795). Hoboken, NJ: John Wiley & Sons

Luthar, S. S., Sawyer, J. A., & Brown, P. J., (2006). Conceptual Issues in Studies of Resilience: Past, Present and Future Research. Annuals of the New York Academy of Sciences, 1094, 105-115.

Ma, S., Truong, K., & Strurm, R. (2007). School characteristics and behavior Problems of U.S. Fifth-Graders. Psychiatric Services, 58, 610.

Maag, J. W., & Katsiyannis, A. (2010). Early intervention programs for children with behavior problems and at risk for developing antisocial behaviors: Evidence and research-based practices. Remedial and Special Education, 31,464–475

MacCallum, R. C., Zhang, S., Preacher, K. J., & Rucker, D. D., (2002). On the practice of dichotomization of quantitative variables. Psychological Methods, 7, 19-40.

MacKenzie, M. J., Kotch, J. B., Lee, L.-C., Augsberger, A., & Hutto, N. (2011). A cumulative ecological-transactional risk model of child maltreatment and behavioral outcomes: Reconceptualizing early maltreatment report as risk factor. Children & Youth Services Review, 33, 2392–2398.

Maes, L., & Lievens, J. (2003). Can the school make a difference? A multilevel analysis of adolescent risk and health behaviour. Social Science & Medicine, 56, 517–29.

Maggs, J. L., Frome, P. M., Eccles, J. S., & Barber, B. L. (1997). Psychosocial resources, adolescent risk behaviour and young adult adjustment: is risk taking more dangerous for some than others? Journal of Adolescence, 20, 103-119.

Mahoney, J. L. (2000). School extracurricular activity participation as a moderator in the development of antisocial patterns. Child Development, 71, 502–16.

Malcolm, H., Wilson, V., Davidson, J., & Kirk, S. (2003). Absence from School : A study of its causes and effects in seven LEAs. Nottingham: DfES Publications

341

Martin, R. P., Foels, P., Clanton, G., & Moon, K. (2004). Season of birth is related to child retention rates, achievement, and rate of diagnosis of specific LD. Journal of Learning Disabilities, 37, 307–317.

Masten, A. S., Garmezy, N., Tellegen, a, Pellegrini, D. S., Larkin, K., & Larsen, A. (1988). Competence and stress in school children: the moderating effects of individual and family qualities. Journal of Child Psychology & Psychiatry, 29, 745–64.

Masten, A. S., Best, K. M., & Garmezy, N. (1990). Resilience and development: contributions from the study of children who overcome adversity. Development & Psychopathology, 2, 425-444.

Masten, A. S., & Wright, M. (1998). Cumulative risk and protection models of child maltreatment. Journal of Aggression, Maltreatment & Trauma, 2, 7-30.

Masten, A. S., & Coatsworth, J. D. (1998). The development of competence in favorable and unfavorable environments. Lessons from research on successful children. The American Psychologist, 53, 205–20.

Masten, A. S. (2001). Ordinary magic: Resilience processes in development. American Psychologist, 56, 227–238.

Masten, A. S. (2002). Resilience Comes of Age: Reflections on the Past and Outlooks for the Next Generation of Research. In M. D. Glantz, & J. L. Johnson (Eds.). Resilience and Development: Positive Life Adaptations, (pp281-296). London: Kluwer Academic Publishers.

Masten, A. S., & Reed, M. G. (2002). Resilience in development. In S. R. Snyder & S. J. Lopez (Eds.), The handbook of positive psychology, (pp74-88). Oxford: Oxford University Press.

Masten, A. S., & Powell. J. L. (2003). A resilience framework for research, policy and practice. In S. S. Luthar (Ed). Resilience and Vulnerability, (pp1-28). Cambridge: Cambridge University Press.

Masten, A. S., & Obradovic, J. (2006). Competence and resilience in development. Annals of the New York Academy of Sciences, 1094, 13–27.

Masten, A. S. (2006). Promoting Resilience in development: A general framework for systems of care. In R. J. Flynn, P. M. Dudding, & J. g. Barber. (Eds.), Promoting Resilience in Child Welfare, (pp3-17). Ottawa: University of Ottawa Press.

Maughan, B., Rowe, R., Messer, J., Goodman, R., & Meltzer, H. (2004). Conduct disorder and oppositional defiant disorder in a national sample: developmental epidemiology. Journal of Child Psychology & Psychiatry, 45, 609–21.

Maughan, B., Collishaw, S., Meltzer, H., & Goodman, R. (2008). Recent trends in UK and child adolescent mental health. Social Psychiatry & Psychiatric Epidemiology, 43, 305-310.

McAra, L (2004). Truancy, School Exclusion and Substance Misuse. The Edinburgh Study of Youth Transitions and Crime. Edinburgh: Centre for Law and Society, The University of Edinburgh.

342

McCrae, J. S., & Barth, R. P. (2008). Using a cumulative risk to screen for mental health problems in child welfare. Research on Social Work Practice, 18, 144-159.

McLeod, J. D., & Kaiser, K. (2004). Childhood Emotional and Behavioral Problems and Educational Attainment. American Sociological Review, 69, 636–658.

McEvoy, a., & Welker, R. (2000). Antisocial behavior, academic failure, and school climate: A critical review. Journal of Emotional & Behavioral Disorders, 8, 130–140.

Mcintosh, K., Flannery, K. B., Sugai, G., Braun, D. H., & Cochrane, K. L. (2008). Relationships between academics and problem behavior in the transition from Middle school to High school. Journal of Positive Behavior Interventions, 10, 243–255.

Mcloyd, V. C., & Smith, J. (2002). Physical discipline & behavior problems in African American European American and Hispanic children: Emotional support as a moderator. Family Relations, 64, 40–53.

Menet, F., Eakin, J., Stuart, M., & Rafferty, H. (2000). Month of birth and effect on literacy, behaviour and referral to psychological service. Educational Psychology in Practice, 16, 225–234.

Mercer, S. H., McMillen, J. S., & DeRosier, M. E. (2009). Predicting change in children’s aggression and victimization using classroom-level descriptive norms of aggression and pro-social behavior. Journal of School Psychology, 47, 267–89.

Merrell, K, W. (2003). Behavioral, social, and emotional assessment of children and adolescents. London: Lawrence Erlbaum Associates Publishers.

Mertens, D. M. (2005). Research and evaluation in education and psychology: Integrating diversity with quantitative, qualitative, and mixed methods. London: Sage.

Messer, J., Goodman, R., Rowe, R., Meltzer, H., & Maughan, B. (2006). Preadolescent conduct problems in girls and boys. Journal of the American Academy of Child & Adolescent Psychiatry, 45, 184–91.

Miller, P., & Plant, M. (1999). Truancy and perceived school performance: an alcohol and drug study of UK teenagers. Alcohol & Alcoholism, 34, 886–893.

Millie, A., Jacobson, J, McDonald, E & Hough, M. (2005). Antisocial behaviour strategies: finding a balance. Bristol: Policy Press.

Mooij, T. (1998). Pupil-class determinants of aggressive and victim behaviour in pupils. British Journal of Educational Psychology, 68, 373–85.

Morales, J. R., & Guerra, N. G. (2006). Effects of multiple context and cumulative stress on urban children’s adjustment in elementary school. Child Development, 77, 907–23.

Morgan, D., L. (2007). Paradigms lost and paradigms regained: Methodological implications for combining qualitative and quantitative designs. Journal of Mixed Methods Research, 1, 48-76.

343

Morgan, P. L., Farkas, G., Tufis, P. A., & Sperling, R. A. (2008). Are reading and behaviour problems risk factors for each other? Journal of Learning Disabilities, 41, 417-436.

Moss, E., Smolla, N.,, Cyr, C.,, Dubois-Comtois, K., Mazzarello, T., & Berthiaume, C. (2006). Attachment and behavior problems in middle childhood as reported by adult and child informants. Development & Psychopathology, 18, 425-44.

Mrug, S., & Windle, M. (2009). Mediators of neighborhood influences on externalizing behavior in preadolescent children. Journal of Abnormal Child Psychology, 37, 265–80.

Muijs, D. (2011). Doing quantitative research in education with SPSS. London: Sage.

Munn,P & Lloyd,G. (1998). Discipline In Schools. A Review of extent, causes and cures. A Literature Review for the Scottish Office. Moray House Publications.

Murray, C. (2003). Risk factors, protective factors, vulnerability, and resilience: A framework for understanding and supporting the adult transitions of youth with high-incidence disabilities. Remedial and Special Education, 24, 16–26

Murray, C., & Greenberg, M. T. (2006). Examining the Importance of Social Relationships and Social Contexts in the Lives of Children With High-Incidence Disabilities. The Journal of Special Education, 39, 220–233.

Murray, J., Farrington, D. P., & Eisner, M. P. (2009). Drawing conclusions about causes from systematic reviews of risk factors: The Cambridge Quality Checklists. Journal of Experimental Criminology, 5, 1–23.

Naglieri, J. A., & LeBuffe, P. A. (2005). Measuring Resilience in children: From theory to practice. In S., Goldstein, & S. B., Brooks. (Eds.) Handbook of resilience in children (p107-121). New York: Springer.

Nazroo, J. (2011). Using Longitudinal Survey Data, In J. Mason, & A. Dale (Eds.). Understanding Social Research: Thinking Creatively about Method (pp 225-242). London: Sage.

Offord, D. R., & Kraemer, H. (2000). Risk factors and prevention. Evidenced Based Mental Health, 3, 70-71.

Ofsted (2005). Managing Challenging Behaviour. London: Ofsted.

Olson, S. L., Ceballo, R., & Park, C. (2002). Early problem behavior among children from low-income, mother- headed families : A multiple risk perspective. Journal of Clinical Child & Adolescent Psychology, 31, 419-430.

Olsson, C. a, Bond, L., Burns, J. M., Vella-Brodrick, D. A, & Sawyer, S. M. (2003). Adolescent resilience: a concept analysis. Journal of Adolescence, 26, 1–11.

Olweus, D. (1993). Bullying at school: what we know and what we can do. Oxford: Blackwell.

O'Regan, F. J. (2005). Surviving and succeeding in special educational needs. London: Continuum.

Pallant, J. (2010). SPSS Survival Manual. Maidenhead: Open University Press

344

Papatheodorou, T. (2005). Behaviour problems in the early years: A guide for understanding and support. London: RoutledgeFalmer.

Patchin, J. W., Huebner, B. M., McCluskey, J. D., Varano, S. P., & Bynum, T., S. (2006). Exposure to community violence and childhood delinquency. Crime & Delinquency, 52, 307-332.

Payne, A. A. (2008). A multilevel analysis of the relationships among communal school organization, student bonding, and delinquency. Journal of Research in Crime & Delinquency, 45, 429–455.

Peer, L. & Reid, G. (2012). Special educational needs: a guide for inclusive practice. London: Sage.

Pettit, G. S., Laird, R. D., Dodge, K. A, Bates, J. E., & Criss, M. M. (2001). Antecedents and behavior-problem outcomes of parental monitoring and psychological control in early adolescence. Child Development, 72, 583–98.

Peugh, J.L., & C.K. Enders (2004), Missing data in educational research: A review of reporting practices and suggestions for improvement. Review of Educational Research, 74, 525-556.

Pollard, J. A., Hawkins, D. & Arthur, M. W., (1999). Risk and protection: are both necessary to understand diverse behavioural outcomes in adolescence? Social Work Research, 23, 145-158.

Polizzi, N., Martin, R. P., & Dombrowski, S. C. (2007). Season of birth of students receiving special education services under a diagnosis of emotional and behavioral disorder. School Psychology Quarterly, 22, 44–57.

Price, M. N., & Hyde, J. S. (2009). When two isn’t better than one: predictors of early sexual activity in adolescence using a cumulative risk model. Journal of Youth & Adolescence, 38, 1059–71.

Propper, C., & Rigg, J. (2007). Socio-Economic Status and Child Behaviour: Evidence from a contemporary UK cohort. Centre for Analysis of Social Exclusion (CASE), 125, 1-16.

Pugh, R. & Chitiyo, M. (2012).The problem of bullying in schools and the promise of positive behaviour supports. Journal of Research in Special Educational Needs, 12, 47–53.

Qi, C. H., & Kaiser, A. P. (2003). Behavior problems of preschool children from low-income families: Review of the literature. Topics in Early Childhood Special Education, 23, 188-216

Rasbash, J., Steele, F., Browne, W.J. and Goldstein, H. (2009) A User’s Guide to MLwiN, v2.10. Centre for Multilevel Modelling. Bristol: University of Bristol.

Raviv, T., Taussig, H. N., Culhane, S. E., & Garrido, E. F. (2010). Cumulative risk exposure and mental health symptoms among maltreated youth placed in out-of-home care. Child Abuse & Neglect, 34, 742–51.

Reber, A. S., & Reber, E. S. (2001). The penguin dictionary of psychology. London: Penguin Books.

345

Reef, J., Diamantopoulou, S., van Meurs, I., Verhulst, F., & van der Ende, J. (2010). Predicting adult emotional and behavioral problems from externalizing problems trajectories in a 24 year longitudinal study. European Child & Adolescent Psychiatry, 19, 577-585.

Reid, K. (2010). Improving attendance and behaviour in Wales: the action plan. Educational Studies, 36, 233–247.

Reid, R., Gonzalez, J. E., Nordness, P. D., Trout, a., & Epstein, M. H. (2004). A Meta-Analysis of the Academic Status of Students with Emotional/Behavioral Disturbance. The Journal of Special Education, 38, 130–143.

Reinke, W. M., & Herman, K. C. (2002). Creating school environments that deter antisocial behaviors in youth. Psychology in the Schools, 39, 549–559.

Reis, J., Trockel, M., & Mulhall, P. (2007). Individual and School Predictors of Middle School Aggression. Youth & Society, 38, 322–347.

Renzaho, A. M. N., & Karantzas, G. (2010). Effects of parental perception of neighbourhood deprivation and family environment characteristics on pro-social behaviours among 4-12 year old children. Australian & New Zealand Journal of Public Health, 34, 405–11.

Ribeaud, D., & Eisner, M. (2010). Risk factors for aggression in pre-adolescence: Risk domains, cumulative risk and gender differences - Results from a prospective longitudinal study in a multi-ethnic urban sample. European Journal of Criminology, 7, 460–498.

Richman, J. M., & Fraser, M. W., (2001). Resilience in childhood: the role of risk and protection. In J. M. Richman, & M. W. Fraser (Eds.), The context of youth violence: Resilience, Risk and Protection (pp1-12). West point, CT: Greenwood.

Riddell, S., Weedon, E., & Harris, N. (2012). Special and additional support needs in England and Scotland: current dilemmas and solutions. In L. Peer & G. Reid (Eds). Special Educational Needs: A guide for inclusive practice, (pp9-23). London: Sage.

Roberts, R. E., Attkisson, C. C., & Rosenblatt, A. (1998). Prevalence of Psychopathology among children and adolescents. American Journal of Psychiatry, 115, 715-725.

Robson, C. (2011). Real world research. Chichester: Wiley.

Rose, R., Howley, M., Fergusson, A., & Jament, J. (2009) Mental health and special educational needs: exploring a complex relationship. British Journal of Special Education, 36, 3-8.

Rouse, H. L., & Fantuzzo, J. W. (2009). Multiple risks and educational wellbeing: A population-based investigation of threats to early school success. Early Childhood Research Quarterly, 24, 1–14.

Russell, R.J.H. and Startup, M.J. (1986). Month of birth and academic achievement. Personality & Individual Differences, 7, 839-846

Rutter, M., Maughan, B., Mortimore, P., Ouston., J., & Smith, A. (1979). Fifteen thousand hours: Secondary schools and their effects on children. Boston, MA: Harvard University Press.

346

Rutter, M. (1979). Protective factors in children’s response to stress and disadvantage. In M. W. Kent & J. E. Rolf (Eds.), Primary prevention of psychopathology: Vol 3, Promoting social competence and coping in children (pp49-74). Hanover, NH: University Press of New England

Rutter, M. (1985). Resilience in the Face of Adversity. British Journal of Psychiatry, 147, 598-611.

Rutter, M. (1993). Resilience: Some conceptual considerations. Journal of Adolescent Health, 14, 626–631

Rutter, M. (1999). Resilience concepts and findings: implications for family therapy. Journal of Family Therapy, 21, 119–144.

Rutter, M. (2000). Resilience reconsidered: conceptual considerations, empirical findings, and policy implications. In J. P. Shonkoff & S. J. Meisels (Eds.). Handbook of early childhood intervention, (pp651-682). Cambridge: Cambridge University Press.

Rutter, M. (2007). Resilience, competence, and coping. Child Abuse & Neglect, 31, 205–209.

Sameroff, A. J. Seifer, R., Baldwin, A., & Baldwin, C. (1993). Stability of intelligence from preschool to adolescence: the influence of social and family risk factors. Child Development, 64, 80-97.

Sameroff, A., Bartko, W. T., Baldwin, A., Baldwin, C., & Seifer, R. (1998). Family and social influences on development of child competence. In M. Lewis, & C. Feiring (Eds). Families, risk and competence. (pp161-185). London: Lawrence Erlbaum Associates.

Sameroff, A., Gutman, L., & Peck, S. C. (2003). Adaptation among youth facing multiple risks: protective research findings. In S. S. Luthar (Ed). Resilience and Vulnerability. (pp130-155). Cambridge: Cambridge University Press.

Schlomer, G.L., Bauman, S.,, Card, N. A. (2010). Best practices for missing data management in counseling psychology. Journal of Counseling Psychology, 57, 1-10.

Schonberg, M. a, & Shaw, D. S. (2007). Do the predictors of child conduct problems vary by high- and low-levels of socioeconomic and neighborhood risk? Clinical Child & Family Psychology Review, 10, 101–36.

Schoon, I. (2006). Risk and resilience: adaptations in changing times. Cambridge: Cambridge University Press.

Scott, S., Knapp, M., Henderson, J., & Maughan, B. (2001). Financial cost of social exclusion: follow up study of antisocial children into adulthood. British Medical Journal, 323, 191.

Sellström, E., & Bremberg, S. (2006). Is there a “school effect” on pupil outcomes? A review of multilevel studies. Journal of Epidemiology & Community Health, 60, 149–55.

Sharpe, M. N., York, J. L. & Knight, J. (1994) Effects of inclusion on the academic performance of classmates without disabilities: a preliminary study. Remedial & Special Education, 15, 281–287.

347

Silver, R. B., Measelle, J. R., Armstrong, J. M., & Essex, M. J. (2005). Trajectories of classroom externalizing behavior : Contributions of child characteristics, family characteristics , and the teacher – child relationship during the school transition. Journal of School Psychology, 43, 39 – 60.

Smith, P. K. (2004). Bullying: Recent developments. Child & Adolescent Mental Health, 9, 98–103.

Smith, D. J., &Rutter, M. (1995). Time trends in psychosocial disorders of youth. In M. Rutter & D. J. Smith (Eds.). Psychosocial disorder in young people: Time trends and their causes. Chichester: John Wiley & Sons.

Soles, T., Bloom, E. L., Heath, N. L., & Karagiannakis, A. (2008). An exploration of teachers’ current perceptions of children with emotional and behavioural difficulties. Emotional & Behavioural Difficulties, 13, 275-290.

Sourander, A., Santalahti, P., Haavisto, A., Piha, J., Ikaheimo, K., & Helenius, H. (2004). Have there been changes in children’s psychiatric symptoms and mental health service use? A 10 year comparison from Finland. Journal of the American Academy of Child & Adolescent Psychiatry, 43, 1134-1145.

Sourander, A., Multimaki, P., Nikolakaros, G., Haavisto, A., Ristkari, T., Helenius, H., Parkkola, K., Piha, J., Tamminen, T., Moilanen, I., Kumpulainen, K., & Almqvist, (2005). Childhood predictors of psychiatric disorders among boys: a prospective community based follow up study from aged 8 years to early adulthood. Journal of the American Academy of Child & Adolescent Psychiatry, 44, 756-767.

Spooner, W. (2006). The SEN handbook for trainee teachers, NQTs and TAs. London: David Fulton.

Squires, G. (2012). Historical and socio-political agendas around defining and including children with special educational needs. In D. Armstrong, & G. Squires (Eds.). Contemporary Issues in Special Educational Needs: considering the whole child (pp9-24). Maidenhead: Open University Press

Stewart, E. A., Simons, R. L., & Conger, R. D., (2002). Assessing Neighbourhood and Social Psychological Influences on Childhood Violence in an African-American Sample. Criminology, 40, 801-830.

Stewart, E. A. (2003). School social bonds, school climate, and school misbehavior a multilevel analysis. Justice Quarterly, 20, 575-604.

Storvoll, E. E., & Wichstrøm, L. (2002). Do the risk factors associated with conduct problems in adolescents vary according to gender? Journal of Adolescence, 25, 183–202.

Stouthamer-Loeber, M., Loeber, R., Farrington, D. P., Zhang, Q., Van Kammen, W., & Maguin, E. (1993). The double edge of protective and risk factors for delinquency: Interrelations and developmental patterns. Development & Psychopathology, 5, 683-701.

Stouthamer-Loeber, M., Loeber, R., Wei, E., Farrington, D. P., & Wikström, P.-O. H. (2002). Risk and promotive effects in the explanation of persistent serious delinquency in boys. Journal of Consulting & Clinical Psychology, 70, 111–123.

348

Sullivan, T. N., Kung, E. M., & Farrell, A. D. (2004). Relation between witnessing violence and drug use initiation among rural adolescents: parental monitoring and family support as protective factors. Journal of Clinical Child & Adolescent Psychology, 33, 488-498.

Sullivan, T. N., & Farrell, A. D. (1999). Identification and Impact of Risk and Protective Factors for Drug Use Among Urban African American Adolescents. Journal of Clinical Child Psychology, 28, 122-136.

Swahn, M. H., Gaylor, E., & Bossarte, R. M., (2010). Co-occurring suicide attempts and physical fighting : a comparison between urban, suburban, and rural high school students. Vulnerable Children & Youth Studies, 5, 353-362.

Tabacnick, B. G, & Fidell, L. S. (2007). Using multivariate statistics. London: Pearson Education.

Tapasak, R. C. & Walther-Thomas, C. S. (1999). Evaluation of a first-year inclusion programme: student perceptions and classroom performance. Remedial & Special Education, 20, 216– 225.

Terwee, C., Bot, S., de Boer, M., van der Windt, D., Knol, D, Dekker, J., Bouter, L. et al. (2007). Quality criteria were proposed for measurement properties of health status questionnaires. Journal of Clinical Epidemiology, 60, 34‐42.

Theriot, M. T., Craun, S. W., & Dupper, D. R. (2010). Multilevel evaluation of factors predicting school exclusion among middle and high school students. Children & Youth Services Review, 32, 13–19.

Thomas, D. E., & Bierman, K. L. (2006). The impact of classroom aggression on the development of aggressive behavior problems in children. Development & Psychopathology, 18, 471–87.

Thorell, L. B., & Rydell, A.M. (2008). Behaviour problems and social competence deficits associated with symptoms of attention-deficit/hyperactivity disorder: Effects of age and gender. Child Care, Health & Development, 34, 584–95.

Tick, N. T., van de Ende, J., & Verhulst, F. C. (2007). Twenty-year trends in emotional and behavioral problems in Dutch children in a changing society. Acta Psychiatrica Scandinavica, 116, 473-482.

Tiet, Q. Q., Bird, H. R., Davies, M., Hoven, C., Cohen, P., Jensen, P. S., & Goodman, S. (1998). Adverse life events and resilience. Journal of the American Academy of Child & Adolescent Psychiatry, 37, 1191-1200.

Tiet, Q. Q., Bird, H. R., Hoven, C. W., Wu, P., Moore, R., & Davies, M. (2001). Resilience in the Face of Maternal Psychopathology and Adverse Life Events. Journal of Child & Family Studies, 10, 347–365.

Tremblay, R. E. (2000). The development of aggressive behaviour during childhood : What have we learned in the past century. International Journal of Behavioral Development, 24, 129–141.

Trentacosta, C. J., Hyde, L. W., Shaw, D. S., Dishion, T. J., Gardner, F., & Wilson, M. (2008). The relations among cumulative risk, parenting, and behavior problems during early childhood. Journal of Child Psychology & Psychiatry, 49, 1211–1219.

349

Twisk, J. W. R. (2006). Applied multilevel analysis: a practical guide. Cambridge: Cambridge University Press.

Twisk, J. & de Vente, W. (2002). Attrition in longitudinal studies. How to deal with missing data. Journal of Clinical Epidemiology, 55, 329-37

Vanderbilt-Adriance, E., & Shaw, D. S. (2008). Conceptualizing and re-evaluating resilience across levels of risk, time, and domains of competence. Clinical Child & Family Psychology Review, 11, 30–58.

Van der Laan, A. M., Veenstra, R., Bogaerts, S., Verhulst, F. C., & Ormel, J. (2010). Serious, minor, and non-delinquents in early adolescence: the impact of cumulative risk and promotive factors. The TRAILS study. Journal of Abnormal Child Psychology, 38, 339–51.

Van der Molen, E., Hipwell, A. E., Vermeiren, R., & Loeber, R. (2011). Cumulative effects of mothers’ risk and promotive factors on daughters' disruptive behavior. Journal of Abnormal Child Psychology, 40, 727-739.

Van der Put, C., Van der Laan, P., Stams, G, Dekovic, M., & Hoeve, M. (2011). Promotive factors during adolescence: are there changes in impact and prevalence during adolescence and how does this related to risk factors? International Journal of Child, Youth and Family Studies, 1 & 2, 119-141.

Vanfossen, B., Brown, C. H., Kellam, S., Sokoloff, N., & Doering, S. (2010). Neighbourhood context and the development of aggression in boys and girls, Journal of Community Psychology, 38, 329–349.

Viding, E., Frick, P. J., & Plomin, R. (2007). Aetiology of the relationship between callous-unemotional traits and conduct problems in childhood. The British Journal of Psychiatry, Supplement, 49, 33–8

Visser, J. (2002). The David Wills lecture 2001. Eternal verities: the strongest links. Emotional & Behavioural Difficulties, 7, 68-84.

Visser, J. (2003). A study of children and young people who present challenging behaviour. Birmingham: University of Birmingham.

Visser, J., & Stokes, S. (2003). Is education ready for the inclusion of pupils with emotional and behavioural difficulties: a rights perspective? Educational Review, 55, 65-75.

Warner, B. S., Weist, M. D., & Krulak, A. (1999). Risk factors for school violence. Urban Education, 34, 52–68.

Warnock, M., (2010). Special Educational Needs: A new look. In M. Warnock, B. Norwich & L. Terzi (Eds.), Special Educational Needs: A New Look: Key Debates in Educational Policy (pp11-46). London: Continuum.

Warnock, M., Norwich, B., & Terzi, L. (2010). Special Educational Needs: A New Look: Key Debates in Educational Policy. London: Continuum.

Welsh, W. N. (2000). The effects of school climate on school disorder. The Annals of the American Academy of Political and Social Science, 567, 88-107.

Werner, E. E., & Smith, R. S., (1992). Overcoming the odds: High-risk children from birth to adulthood. Ithaca, NY: Cornell University Press.

350

Werner, E. E. (1993). Risk, resilience, and recovery: perspectives from the Kauai longitudinal study. Development & Psychopathology, 5, 503-515

Werner, E. E., (2000). Protective Factors and Individual Resilience. In J. P. Shonkoff & S. J. Meisels (Eds.), Handbook of early childhood intervention (2nd edition) (pp115-132). Cambridge: Cambridge University Press.

Wiener, J. (2004). Do Peer Relationships Foster Behavioural Adjustment in Children with Learning Disabilities? Learning Disability Quarterly, 27, 21-30

Wiglesworth, M., Oldfield, J., & Humphrey, N. (2012). Validation of the Wider Outcomes Survey for Teachers (WOST): A measure for assessing the behaviour, relationships and exposure to bullying of children and young people with Special Educational Needs or Disabilities (SEND). Manuscript submitted for publication

Wilson, S. J.,& Lipsey, M. W., & Derzon, J. H. (2003). The effects of school-based intervention programs on aggressive behavior: A meta-analysis. Journal of Consulting and Clinical Psychology, 71, 136–149.

Wilson, S. J.,& Lipsey, M. W (2007). School-based interventions for aggressive and disruptive behavior: update of a meta-analysis. American Journal of Preventive Medicine, 33, S130–S143

Windle, M. (2002) Critical conceptual and measurement issues in the study of resilience. In M. D. Glantz, & J. L. Johnson (Eds.). Resilience and development: Positive life adaptations (pp161-178). London: Kluwer Academic Publishers

Windle, G. (2011). What is resilience? A review and concept analysis. Reviews in Clinical Gerontology, 21, 152-169.

Wilcox-Rountree, P., & Clayton, R. R. (1999). A Contextual Model of Adolescent Alcohol Use across the Rural-Urban Continuum. Substance Use & Misuse, 34, 495-519.

Wilcox, P., & Clayton, R. R. (2001). A multilevel analysis of school- based weapon possession. Substance Abuse, 18, 509-541.

Wilson, D. (2004). The Interface of school climate and school connectedness and relationships with aggression and victimization. Journal of School Health, 74, 293–299.

Wilson, V., Malcolm, H., Edward, S., & Davidson, J. (2008). “Bunking off”: the impact of truancy on pupils and teachers. British Educational Research Journal, 34, 1–17.

Wilkinson, L. (1999). Statistical methods in psychology journals: Guidelines and explanations. American Psychologist, 54, 594-604.

Whipple, S. S., Evans, G. W., Barry, R. L., & Maxwell, L. E. (2010). An ecological perspective on cumulative school and neighborhood risk factors related to achievement. Journal of Applied Developmental Psychology, 31, 422–427.

Wolke, D., Woods, S., Bloomfield, L., & Karstadt, L. (2000). The association between direct and relational bullying and behaviour problems among primary school children. Journal of Child Psychology & Psychiatry, 41, 989–1002.

351

Woodward, L. J., Fergusson, D. M., & Horwood, L. J. (2002). Romantic relationships of young people with childhood and adolescent onset antisocial behavior problems. Journal of Abnormal Child Psychology, 30, 231-243.

World Health Organisation (1992). ICD-10 Classification of mental and behaviour disorders: clinical descriptors and diagnostic guidelines. Geneva: WHO.

Wright, M., & Masten, A., (2005). Resilience process in development: fostering positive adaptation in the context of adversity. In S., Goldstein, & S. B., Brooks (Eds.). Handbook of resilience in children (p17-37). New York: Springer.

Yeide, M., & Kobrin, M. (2009). Truancy Literature Review, (Prepared for U.S. Department of Justice). Washington, DC: Office of Juvenile Justice & Delinquency Prevention.

Zeitlin, H., & Refaat, R. (1999). Cultural issues in child and adolescent psychiatry. In P. Cooper (Ed.), Understanding and supporting children with emotional and behavioural difficulties. London: Jessica Kingsley Publishers

Zimmerman, M. A., & Arunkumar, R. (1994). Resiliency research: Implications for schools and policy. Social Policy Report: Society for Research in Child Development, 8, 1-20.

Zwirs, B., Burger, H., Schulpen, T., Vermulst, a. a., HiraSing, R. a., & Buitelaar, J. (2010). Teacher ratings of children’s behavior problems and functional impairment across gender and ethnicity: Construct equivalence of the SDQ Journal of Cross-Cultural Psychology, 42, 466–481.

352

APPENDICES

Appendix 1: a) Original WOST questionnaire 353

b) Retained items after psychometric analysis 356

Appendix 2: School information regarding Achievement for All 357

Appendix 3: Parent information and consent form 360

Appendix 4: Teacher information and consent form 364

Appendix 5: Survey instructions for teachers 368

Appendix 6: Multilevel model outputs 371

Appendix 7: Data assumptions and requirements 394

Appendix 8: Missing data analysis 396

353

Appendix 1a: Original WOST questionnaire

Achievement for All Survey for Teachers of Children and Young People with Special Educational Needs and

Disabilities (SEND)

This survey is being conducted with teachers of children and young people with a range of SEND. It is for use in the national evaluation of a government initiative called Achievement for All. The first few questions in the survey are designed to allow us to match your responses to data about your pupils that are held within our secure database. Please note that you will not be asked to provide your pupil name. All responses will be completely anonymous and treated as confidential.

Section 1 – ABOUT THE PUPIL

1. Please select your Local Authority

2. Please tell us the name of your pupil’s school

3. Please tell us the year group that your pupil is currently in

4. Please tell us your pupil’s date of birth 5. If you know, please indicate your pupil’s

primary SEND (special educational need or disability)

6. If you know, what level of support is your pupil getting at school? School Action (my pupil is supported by teachers and teaching assistants in school) School Action Plus (my pupil is supported by teachers and teaching assistants in school AND at least one other specialist visits the school to provide support)

Statement of SEND (my pupil has a Statement of Special Educational Needs issued by the Local Authority) No longer on the special needs register Don't know

7. Please select your pupil’s gender

Bexley Coventry Essex Nottinghamshire Redcar & Cleveland

Camden East Sussex Gloucestershire Oldham Sheffield

School name

Year 1 Year 5 Year 7 Year 10

Year 2 Year 6 Year 8 Year 11

Date of birth d d / m m / y y

Specific Learning Difficulty (e.g. dyslexia) Autistic Spectrum Disorder

Moderate Learning Difficulties Visual Impairment

Severe Learning Difficulties Hearing Impairment

Profound and Multiple Learning Difficulties Multi‐Sensory Impairment

Behavioural, Emotional and Social Difficulties Physical Disability

Speech, Language and Communication Difficulties Don't know

Male Female

http://www.manchester.ac.uk/index.ph�

354

8. Please tell us your pupil’s UPN (13 character unique pupil number)

Section 2 – ABOUT YOUR PUPIL’S BEHAVIOUR

Please read each of the following statements about your pupil’s behaviour and indicate how often each occurs.

Section 3 – ABOUT YOUR PUPIL’S RELATIONSHIPS WITH OTHER PEOPLE

Please read each of the following statements about your pupil’s relationships with other people and indicate the extent to which you agree.

Strongly Disagree

Disagree Agree Strongly Agree

10 The pupil has at least one good friend 11 The pupil can compromise with other children (e.g.

take turns)

12 The pupil is helpful towards others 13 The pupil is popular with other children 14 The pupil has a good relationship with at least one

teacher

15 The pupil can compromise with teachers (e.g. will complete a difficult task before moving on to a preferred activity)

16 The pupil is kind towards others 17 The pupil makes friends easily 18 The pupil will approach groups of children 19 The pupil can join in other children’s activities

Never Rarely Sometimes Often 1 The pupil does as he/she is asked 2 The pupil cheats and tells lies 3 The pupil takes things that do not belong to

him/her

4 The pupil breaks or spoils things on purpose 5 The pupil behaves well when unsupervised 6 The pupil gets angry and has tantrums 7 The pupil gets in fights with other children 8 The pupil says nasty things to other children 9 The pupil takes responsibility for his/her actions

355

Section 4 – ABOUT YOUR PUPIL’S EXPERIENCE OF BULLYING

By bullying we mean “behaviour by an individual or group, usually repeated over time, that intentionally hurts another individual or group either physically or emotionally” (DCSF, 2008, p.1). Please estimate how frequently your pupil is involved in incidents of bullying.

Please indicate his/her typical role in such incidents.

Please read each of the following statements about your pupil’s experiences of bullying and indicate how often each example occurs.

Never Rarely Sometimes Often

36 The pupil is picked on by other children

37 The pupil is hurt by other children (e.g. gets pushed or kicked)

38 The pupil is called names or teased by other children.

39 Other children spread unkind gossip about the pupil.

40 The pupil is bullied over the internet and/or by text message

41 Other children stop the pupil from joining in their games and activities at break‐times.

42 The pupil is actively disliked by other children

43 Other children stop the pupil from joining in during class activities

44 The pupil is picked on because of his/her special needs

Thank you very much for completing the survey.

Your responses are completely anonymous and will be treated as confidential. Please return the completed survey to: FREEPOST RLYU-KAAB-AXRC University of Manchester Dr. Alexandra Barlow, School of Education, Oxford Road, Manchester, M13 9PL This is the final Achievement for All pupil survey. Your input has been vital to the robust evaluation of this initiative and we greatly appreciate you cooperation. If you have any queries about this research project, please contact Dr. Alexandra Barlow on 0161 273 5304.

Daily Weekly Termly Not involved

Victim Bully Bully and victim Bystander Not involved

356

Appendix 1b: Items on WOST questionnaire retained after psychometric analysis

Behaviour

Positive relationships Strongly

Disagree Disagree Agree Strongly Agree

1. The pupil can compromise with other children (e.g. take turns)

2. The pupil is helpful towards others 3. The pupil is popular with other

children

4. The pupil can compromise with teachers (e.g. will complete a difficult task before moving on to a preferred activity)

5. The pupil is kind towards others 6. The pupil makes friends easily 7. The pupil can join in other children’s

activities

Bullying

Never Rarely Sometimes Often 1. The pupil cheats and tells lies 2. The pupil takes things that do not

belong to him/her

3. The pupil breaks or spoils things on purpose

4. The pupil gets angry and has tantrums

5. The pupil gets in fights with other children

6. The pupil says nasty things to other children

Never Rarely Sometimes Often 1. The pupil is picked on by other children 2. The pupil is hurt by other children (e.g. gets

pushed or kicked)

3. The pupil is called names or teased by other children.

4. Other children spread unkind gossip about the pupil.

5. Other children stop the pupil from joining in their games and activities at break-times.

6. The pupil is actively disliked by other children 7. Other children stop the pupil from joining in

during class activities

357

Appendix 2: School information letter about Achievement for All

Achievement for All National Evaluation

28th October 2009

Dear colleague

I am writing to invite your school to participate in the national evaluation of Achievement for All (AfA). This research project is being led by the University of Manchester and funded by the Department for Children, Schools and Families. It began in September 2009 and will conclude in August 2011. You may remember our brief presentation about the project at the recent NCSL launch event in Coventry. If you are not the AfA lead in your school, please pass this letter on to the appropriate person and contact us using the email address provided below so that we can amend our records.

Our records show that your school is receiving funding from Coventry to implement AfA. The LA Project lead for AfA, Susan Briggs, has endorsed this research project and is keen for all AfA schools to participate. In this letter we describe what participation in the project will entail and the benefits of your school becoming involved. This will enable you to make a fully informed decision regarding participation.

What is the aim of the research?

Our main aim is to examine the impact of AfA on a variety of outcomes (e.g. academic attainment, behaviour, positive relationships, parental engagement and confidence) for children and young people with special educational needs and disabilities (SEND). We also aim to find out what processes and practices in schools are most effective in improving these outcomes.

What will participation entail for my school?

We will be conducting surveys focusing upon outcomes related to Strands 2 and 3 of AfA51

51 Outcome data relating to Strand 1 of AFA – pupil’s academic attainment scores – will be collected by the National Strategies. They will contact you independently to discuss this matter.

. The surveys will be completed three times – in January 2010, November 2010 and June 2011. We will be asking the designated AfA key teachers and parents of pupils with SEND to complete the surveys online on a secure, password-protected website. The key teacher surveys will take approximately 5 minutes per pupil and will focus upon behaviour, positive relationships and bullying. The parental surveys will take approximately 10 minutes per pupil and will focus upon behaviour, positive relationships, bullying, extended services participation and parental engagement and confidence. In line with the AfA guidance, our target sample are all pupils with SEND in Years 1, 5, 7 and 10.

358

If you agree to your school’s participation in the study, you will be sent further details covering the practicalities of notifying parents and completing the surveys in the near future. Briefly, survey access will be facilitated by a unique school password that will be sent prior to each wave of data collection. The password will allow each designated AfA key teacher to log into the website and view a list of pupils with SEND in the target year groups in your school (this is made possible because we have agreement from DCSF to use the National Pupil Database to populate our survey). They will then be able to quickly and easily complete a survey on each of their designated pupils before logging off.

In relation to parents, we will provide information sheets and opt-out consent forms prior the first wave of data collection. Parents will also receive a unique school password to allow them to access the survey site. However, in the interests of privacy and data protection they will not be able to view a list of pupils with SEND in the target year groups in your school. Instead, they will be asked to provide some key pieces of information (e.g. their child’s gender, year group, and date of birth) that will allow us to later match their survey responses to background information about their child.

In addition to the above outcomes surveys, we will also be conducting two surveys focusing on AfA implementation – one in March 2010 and one in March 2011. This survey needs to be completed by the AfA project lead in your school, and will take around 30 minutes to complete. It will focus upon activity in each of the three strands of AfA (assessment, tracking and intervention; structured conversations with parents; provision for developing wider outcomes), in addition to broader issues such as the climate of the school (e.g. relationships between pupils and staff).

Alongside our survey research, we will also be conducting case studies of a small number of schools (2 per Local Authority). We are currently working with the AfA Project Lead in each Local Authority to identify potential case study schools, and will contact you separately should your school be nominated for this strand of the research.

What happens to the data collected?

The data will be analysed by our research team at the University of Manchester. We will write a report based on our analyses for the Department for Children, Schools and Families. It is also likely that we will write articles for academic journals based on the project findings. Finally, it is possible that we will write a book about the research.

How is confidentiality maintained?

All data provided will be treated as confidential and will be completely anonymous. Identifying information (e.g. pupils’ names) will only be used in order to match responses about the same individual from different respondents (e.g. parents and teachers) and across different times (e.g. January 2010, November 2010, June 2011). After this matching process is complete all identifying information will be destroyed.

The website that houses the survey will be completely secure and password protected. All survey data will be stored on a secure, password protected drive to which only senior members of the research team have access.

359

Criminal Records Check

Every member of our research team has undergone a Criminal Records Bureau check at the Enhanced Disclosure level.

What are the benefits of participation?

There are many good reasons to get involved in this evaluation. These include:

• Minimal data collection burden – each survey is designed to be as brief and user friendly as possible, with a window of 1 month for completion of surveys at each wave of data collection

• Your school will receive bespoke, aggregated feedback following each wave of data collection, comparing your findings to the average within your LA and the national sample as a whole; this feedback should be very useful to help in your future planning and can be used in your school’s Self Evaluation Form.

• Data will be available on outcomes pertaining to all three strands of AfA – allowing you to judge its impact in your school.

• We will make every effort to include your ‘hard to reach’ parents - including translated information sheets and surveys for parents whose first language is not English, paper versions for parents without access to the internet, and telephone surveys for parents with literacy difficulties

• The opportunity to take part in what we think will be the largest study of its kind ever conducted in England

• All data will be treated in the strictest confidence and will be completely anonymised prior to analysis and reporting.

What happens next?

We would like you to take some time to think about your school’s participation in this project and perhaps discuss it with colleagues. One of our research team will contact you by telephone and/or email within the next couple of weeks to follow up this invitation. If you decide to participate, we will give you information about the next steps involved.

Alternatively, you can contact Jeremy Oldfield by telephone on 0161 275 3522 or by email at [email protected] to let us know about your decision.

Many thanks for taking the time to read this invitation. We sincerely hope your school will participate in the research and wish you the best of luck in implementing Achievement for All.

Yours sincerely

On behalf of the Achievement for All National Evaluation team

360

Appendix 3: Parent information and consent letters

ACHIEVEMENT FOR ALL – NATIONAL EVALUATION

INFORMATION SHEET FOR PARENTS

Your child’s school is involved in an exciting project called Achievement for All. This project hopes to improve children’s learning and experience of school. The government has asked us to evaluate how well this project works.

We are writing to you because your child is involved in the project and we would like to know what you think about it. We will collect your views in a 10 minute survey in February/March 2010, November 2010 and June 2011 (see below).

Please take time to read the following information carefully and decide whether or not you would like to take part.

If you would like any more information or have any questions about the research project, please telephone Jeremy Oldfield on 0161-275-3522 or email him at [email protected].

Who will conduct the research?

The research will be conducted by Dr. Neil Humphrey and other staff in the School of Education, University of Manchester, Oxford Road, Manchester M13 9PL.

Title of the research

Achievement for All – National Evaluation

What is the aim of the research?

Our main aim is to find out what impact Achievement for All has on outcomes for children and young people with SEND. We also aim to find out what processes and practices in schools are most effective in improving these outcomes.

Where will the research be conducted?

450 schools across 10 Local Authorities in England are involved.

What is the duration of the research?

The project itself runs from September 2009 until August 2011.

361

Why have I been chosen?

We are writing to you because your child’s school is taking part in the AfA initiative, he/she is on the school’s SEND register and also is in one of the AfA target year groups (Years 1 and 5 in primary schools; Years 7 and 10 in secondary schools).

What would I be asked to if I took part?

You will be asked to complete a brief online survey about:

• your child’s behaviour, • positive relationships, • bullying, • participation in wider activities, and • your engagement and confidence in the school.

This survey will be completed three times – in February/March 2010, November 2010 and June 2011. It will take approximately 10 minutes to complete.

In consenting to take part you are also giving your permission for a key teacher at your child’s school to complete a similar but briefer survey (that only covers behaviour, bullying and positive relationships) at the times noted above.

There will be no direct contact between any of our research team and your child for this part of the research project.

The survey will be available in the following additional languages: Arabic, Bengal, Chinese traditional and simplified, French, Gujarati, Polish, Somali and Urdu. If you do not have access to the internet we will be happy to either provide a paper copy or complete it over the telephone with you at an agreed time. If you would like to do this please contact Jeremy Oldfield (details above) and he will arrange this for you.

There is another part of the project which involves case studies that will involve direct contact – if your child’s school becomes involved in this element you will receive a separate information sheet and consent form.

What happens to the data collected?

The data will be analysed by our research team at the University of Manchester. We will write a report based on our analyses for the Department for Children, Schools and Families. It is also likely that we will write articles for academic journals based on what we find out in the project. Finally, it is possible that we will write a book about the research. Your child’s name will not be used in any of the reports that we write.

How is confidentiality maintained?

All data provided will be treated as confidential and will be completely anonymous. Identifying information (e.g. your child’s name) will only be used in order to match responses about the same individual from different respondents (e.g. parents and teachers) and across different times (e.g. February/March 2010, November 2010, June 2011). After this matching process is complete all identifying information will be destroyed.

The website that houses the survey will be completely secure and password protected. All survey data will be stored on a secure, password protected drive to which only senior members of the research team have access.

362

What happens if I do not want to take part or I change my mind?

It is up to you if you want to take part. If you decide to take part you do not need to do anything – you will be sent further details about when and how to complete the survey in the near future.

If you decide not to take part then you need to either complete the opt-out consent form enclosed and return it to our research team at the address above or contact Jeremy Oldfield by telephone or email (details above) by 12th March 2010.

If you decide to take part and then change your mind, you are free to withdraw at any time without needing to give a reason. If you do this please rest assured that we will destroy any data generated in relation to your child as part of the study.

Will I be paid for participating in the research?

We are not able to offer any payment or incentive for participating in this study.

Criminal Records Check

Every member of our research team has undergone a Criminal Records Bureau check at the Enhanced Disclosure level.

Contact for further information

Jeremy Oldfield Educational Support and Inclusion School of Education University of Manchester Oxford Road Manchester M13 9PL Tel: 0161 275 3522 Email: [email protected]

What if something goes wrong?

If completing the survey makes you worry about your child’s wellbeing then you should contact his/her school in the first instance and ask to speak to the Achievement for All co-ordinator.

If you ever wish to make a formal complaint about the conduct of the research you should contact the Head of the Research Office, Christie Building, University of Manchester, Oxford Road, Manchester M13 9PL.

363


PARENTAL CONSENT FORM

An information sheet is attached to this form. Please read it carefully before making a decision about taking part in the study.

If you are willing to take part then you do not need to do anything at the moment. You will be sent further details about when and how to complete the survey in the near future. In consenting to take part, you are also giving your permission for a teacher at your child’s school to complete a survey about your child.

If you decide not to take part, then you need to complete the opt-out consent form below and return it to Jeremy Oldfield, Educational Support and Inclusion, School of Education, University of Manchester, Oxford road, Manchester, M13 9PL. Alternatively, Jeremy can be contacted by telephone on 0161 275 3522 or email at [email protected]. If you do not wish to participate please let us know by 12th March 2010.

Finally, please also remember that if you do decide to take part, you are free to change your mind at any point in the study. Just let us know and we will destroy any data generated in relation to your child.

------------------------------------------------------------------------------------------------------------------------

I do not wish to participate in the Achievement for All national evaluation. Furthermore, I do not give consent for a key teacher at my child’s school to complete a survey about him/her in relation to this study.

Name of child

Sex of child

Year group

Name of school

Local Authority (if known)

Signed: __________________________________ (parent/guardian) Date: __________

Please return this form to Jeremy Oldfield, Educational Support and Inclusion, School of Education, University of Manchester, Manchester M13 9PL by 12th March 2010.

http://www.manchester.ac.uk/index.php�

364

Appendix 4: Teacher information letter and consent form ACHIEVEMENT FOR ALL – NATIONAL EVALUATION INFORMATION SHEET FOR AfA KEY TEACHERS Your school is involved in the national evaluation of Achievement for All (AfA). This research project is being led by the University of Manchester and funded by the Department for Children, Schools and Families. We are writing to you because your school’s AfA co-ordinator has indicated that you are the designated key teacher for one or more pupils with special educational needs and disabilities (SEND). We would like to collect your views about the behaviour, positive relationships and bullying of each of the pupils with SEND for whom you are the key teacher. We will do this via a 5 minute survey in January 2010, November 2010 and June 2011 (see below). Please take time to read the following information carefully and decide whether or not you would like to take part. If you would like any more information or have any questions about the research project, please telephone Jeremy Oldfield on 0161 275 3522 or email him at [email protected]. Who will conduct the research? The research will be conducted by Dr. Neil Humphrey and other staff in the School of Education, University of Manchester, Oxford Road, Manchester M13 9PL. Title of the research Achievement for All – National Evaluation What is the aim of the research? Our main aim is to find out what impact Achievement for All has on outcomes for children and young people with SEND. We also aim to find out what processes and practices in schools are most effective in improving these outcomes. Where will the research be conducted? 450 schools across 10 Local Authorities in England are involved. What is the duration of the research? The project itself runs from September 2009 until August 2011.

mailto:[email protected]�


365

Why have I been chosen? We are writing to you because your school’s AfA co-ordinator has indicated that you are the designated key teacher for one or more pupils with special educational needs and disabilities (SEND). What would I be asked to if I took part? You will be asked to complete a brief online survey about each pupil with SEND for whom you are the key teacher, covering: behaviour, positive relationships, and bullying This survey will be completed three times – in January 2010, November 2010 and June 2011. It will take approximately 5 minutes to complete per pupil. What happens to the data collected? The data will be analysed by our research team at the University of Manchester. We will write a report based on our analyses for the Department for Children, Schools and Families. It is also likely that we will write articles for academic journals based on the project findings. Finally, it is possible that we will write a book about the research. In all publications and reports data will be presented anonymously. How is confidentiality maintained? All data provided will be treated as confidential and will be completely anonymous. Identifying information (e.g. pupils’ names) will only be used in order to match responses about the same individual from different respondents (e.g. parents and teachers) and across different times (e.g. January 2010, November 2010, June 2011). After this matching process is complete all identifying information will be destroyed. The website that houses the survey will be completely secure and password protected. All survey data will be stored on a secure, password protected drive to which only senior members of the research team have access. Criminal Records Check Every member of our research team has undergone a Criminal Records Bureau check at the Enhanced Disclosure level. What happens if I do not want to take part or I change my mind? It is up to you if you want to take part. If you decide to take part you do not need to do anything – you will be sent further details about when and how to complete the survey in the near future.

366

If you decide not to take part then you need to either complete the opt-out consent form enclosed and return it to our research team at the address above or contact Jeremy Oldfield by telephone or email (details above) by (date). If you decide to take part and then change your mind, you are free to withdraw at any time without needing to give a reason. Will I be paid for participating in the research? We are not able to offer any payment or incentive for participating in this study. Criminal Records Check Every member of our research team has undergone a Criminal Records Bureau check at the Enhanced Disclosure level. Contact for further information Jeremy Oldfield Educational Support and Inclusion School of Education University of Manchester Oxford Road Manchester M13 9PL Tel: 0161 275 3522 Email: [email protected] What if something goes wrong? If completing the survey makes you worry about any of your pupils’ wellbeing then you should speak to your school’s AfA co-ordinator in the first instance. If you ever wish to make a formal complaint about the conduct of the research you should contact the Head of the Research Office, Christie Building, University of Manchester, Oxford Road, Manchester M13 9PL.


367


TEACHER CONSENT FORM An information sheet is attached to this form. Please read it carefully before making a decision about taking part in the study. If you are willing to take part then you do not need to do anything at the moment. You will be sent further details about when and how to complete the survey in the near future. If you decide not to take part, then you need to complete the opt-out consent form below and return it to Jeremy Oldfield, Educational Support and Inclusion, School of Education, University of Manchester, Oxford road, Manchester, M13 9PL. Alternatively, Jeremy Oldfield can be contacted by telephone on 0161 275 3522 or email at [email protected]. If you do not wish to participate please let us know by (date). Finally, please also remember that if you do decide to take part, you are free to change your mind at any point in the study. --------------------------------------------------------------------------------------------------------------------------- I do not wish to participate in the Achievement for All national evaluation. My details are as follows:

My name Name of my school Local Authority I am AfA key teacher for the following pupil(s):

1.

2. 3. 4. 5.

Signed: __________________________________ Date: __________ Please return this form to Jeremy Oldfield, Educational Support and Inclusion, School of Education, University of Manchester, Manchester M13 9PL by (date).



368

Appendix 5: Survey instructions for teachers


On-line Teacher Survey Instructions Dear Colleague, This letter tells you how to complete our on-line teacher survey. These surveys are to be completed for all students receiving SEN provision at your school in year groups 1, 5, 7 and 10. The survey should be filled in by the key teacher for each student. Each survey should take approximately 2 to 3 minutes to complete per student. This survey will go live from Monday 18th January 2010 to Sunday 14th February, and can be accessed 24 hours a day, 7 days a week during this period. Completing the survey Follow the steps below. If you have any problems or queries please contact Jeremy Oldfield (Tel: 0161 275 3522, email: [email protected]) or for technical help contact Lawrence Wo (Tel: 0161 275 3415, email: [email protected]). Step 1 Please go to the following website http://www.afa.uk.net/ Click on the ‘teacher’ image as shown below

https://webmail.manchester.ac.uk/horde/imp/compose.php?to=lawrence.wo%40manchester.ac.uk�

http://www.manchester.ac.uk/index.php�

369

Step 2 When you arrive at the survey website you will be asked for your school's password. Enter your school’s password and then click ‘Next’. The password is: XXXXX

Step 3 The name of your school should now be displayed. If this is correct click 'Next'. If the school name is incorrect, click on the blue ‘click here’ hyperlink and this will take you back to the log in screen where you can re-enter your school password. If you still have problems finding your school please contact Lawrence Wo (Tel: 0161 275 3415, email: [email protected]) for technical support. Step 4 You will then see a list of student names. These are students in Years 1, 5, 7 and 10 who have been identified as receiving SEN provision at your school.

You will complete the survey for one student at a time. To select a student to survey click the radio button next to their name. Click 'Next'. If there are students receiving SEN provision who are not on the list, you will have the option to add these students. This option (‘Add a student’) will appear at the bottom of the list of

https://webmail.manchester.ac.uk/horde/imp/compose.php?to=lawrence.wo%40manchester.ac.uk�

370

student names. If you select this option and then click ‘Next’ you will be asked to enter the student's first name, surname and UPN (single letter and 12 digit unique pupil number) before clicking ‘Submit’. The survey for the additional student will then start. Step 5 This will take you to the survey about this student. Click to select your answers.

When you have completed the survey for that student, click ‘Submit’. You will be asked if you want to complete a survey for another student. Click ‘Return to login’ and you will be taken back to the login screen, where you will be asked to re-enter your password. This step is to improve security. Please follow the above instructions to complete the survey again for another student.

Once you have completed all the surveys about students for whom you are the key teacher, please close your browser window. Thank you very much for completing these surveys. All responses are completely anonymous and will be treated as confidential. There will be two more surveys that you will be asked to complete, one in November 2010 and June 2011. We will send you a reminder about these surveys closer to the time.

371

Appendix 6ai: Empty Primary Model

Estimates of Fixed Effectsa

Parameter Estimate Std. Error df t Sig.

95% Confidence Interval

Lower Bound Upper Bound

Intercept .196935 .019255 364.701 10.228 .000 .159071 .234799

XBeh_mean_T1 .586988 .014313 2653.771 41.011 .000 .558922 .615054

a. Dependent Variable: Behaviour mean T3 (items: 2, 3, 4, 6, 7, 8).

Information Criteriaa

-2 Log Likelihood 3973.122

Akaike's Information

Criterion (AIC) 3981.122

Hurvich and Tsai's Criterion

(AICC) 3981.137

Bozdogan's Criterion (CAIC) 4008.667

Schwarz's Bayesian Criterion

(BIC) 4004.667

Estimates of Covariance Parametersa

Parameter Estimate Std. Error Wald Z Sig.



Residual .237428 .006812 34.856 .000 .224445 .251161

Intercept [subject =

School_No]

Variance .043416 .006257 6.939 .000 .032733 .057586


372

Appendix 6aii: Empty Secondary Model





Intercept .327105 .038018 56.625 8.604 .000 .250965 .403246

XBeh_mean_T1 .530592 .020389 1612.339 26.024 .000 .490601 .570584





Criterion (AIC) 2934.001


(AICC) 2934.025



(BIC) 2955.581





Residual .336411 .012007 28.018 .000 .313682 .360787


School_No]

Variance .049595 .013749 3.607 .000 .028805 .085390


373

Appendix 6bi: Full Primary Model

Estimates of Fixed Effectsb


95% Confidence Interval Lower Bound

Upper Bound

Intercept 1.072477 .268573 552.624 3.993 .000 .544928 1.600025 [Season_Of_Birth=1] .072057 .030992 1895.297 2.325 .020 .011276 .132838 [Season_Of_Birth=2] .026272 .028333 1880.619 .927 .354 -.029295 .081840 [Season_Of_Birth=3] .028033 .028925 1863.624 .969 .333 -.028696 .084762 [Season_Of_Birth=4] 0a 0 . . . . . [SENDsupport=3] .048283 .061110 1923.431 .790 .430 -.071567 .168132 [SENDsupport=4] .041012 .026570 1947.215 1.544 .123 -.011096 .093121 [SENDsupport=5] 0a 0 . . . . . [SEND4cat=1.00] .095214 .064215 1765.637 1.483 .138 -.030730 .221159 [SEND4cat=2.00] .268767 .036121 1941.534 7.441 .000 .197928 .339606 [SEND4cat=3.00] -.013717 .032956 1947.343 -.416 .677 -.078350 .050916 [SEND4cat=4.00] -.009095 .099510 1886.473 -.091 .927 -.204256 .186065 [SEND4cat=5.00] 0a 0 . . . . . [Bully_role_T=1] -.075666 .047764 1930.402 -1.584 .113 -.169340 .018009 [Bully_role_T=2] .221312 .053207 1930.613 4.159 .000 .116963 .325661 [Bully_role_T=3] .035401 .044742 1947.485 .791 .429 -.052347 .123149 [Bully_role_T=4] .046474 .056293 1943.187 .826 .409 -.063927 .156874 [Bully_role_T=6] 0a 0 . . . . . XBeh_mean_T1 .419778 .026406 1912.339 15.897 .000 .367992 .471565 Year_group .097220 .025476 1902.037 3.816 .000 .047257 .147184 Gender -.081052 .022954 1886.447 -3.531 .000 -.126069 -.036035 Ethnicity -.014742 .033336 1906.737 -.442 .658 -.080121 .050636 FSM_eligible .070156 .024564 1896.484 2.856 .004 .021981 .118331 EngbaselineZscore -.027629 .012195 1946.785 -2.266 .024 -.051545 -.003712 Attendance_Year1 -.001136 .001854 1895.321 -.612 .540 -.004773 .002501 Rel_mean_T -.095518 .025341 1940.774 -3.769 .000 -.145216 -.045820 Bully_mean_T .031760 .030342 1947.162 1.047 .295 -.027746 .091265 Sch_urbanicity .001147 .052310 204.591 .022 .983 -.101988 .104282 NewSchoolSize .008174 .017411 177.233 .469 .639 -.026185 .042533 Sch_FMS .000672 .001697 206.320 .396 .692 -.002673 .004018 Sch_EAL -.000282 .000861 229.434 -.328 .743 -.001978 .001414 Sch_SA_proportion -.004250 .002777 167.717 -1.530 .128 -.009733 .001233 Sch_SAP_ST_proportion -.000740 .003437 181.040 -.215 .830 -.007522 .006042 Sch_Pri_EngMaths4 -.005830 .001456 182.885 -4.005 .000 -.008702 -.002958 Sch_Overall_absence -.034791 .018044 166.613 -1.928 .056 -.070415 .000833 Exclsuions2 .002470 .020115 150.742 .123 .902 -.037274 .042214 a. This parameter is set to zero because it is redundant.

374

b. Dependent Variable: Behaviour mean T3 (items: 2, 3, 4, 6, 7, 8).





Residual .202147 .006835 29.577 .000 .189186 .215997


School_No]

Variance .031980 .005780 5.533 .000 .022441 .045574




Akaike's Information Criterion

(AIC) 2656.105


(AICC) 2657.349



(BIC) 2845.640

375

Appendix 6bii: Full Secondary Model





Intercept .299458 .539120 55.066 .555 .581 -.780934 1.379850 XBeh_mean_T1 .410950 .039664 1214.830 10.361 .000 .333132 .488768 [Season_Of_Birth=1] .007955 .044685 1183.690 .178 .859 -.079716 .095626 [Season_Of_Birth=2] -.020081 .045111 1184.174 -.445 .656 -.108588 .068425 [Season_Of_Birth=3] -.002710 .047845 1180.594 -.057 .955 -.096580 .091161 [Season_Of_Birth=4] 0a 0 . . . . . [SENDsupport=3] -.058095 .061337 1212.635 -.947 .344 -.178434 .062243 [SENDsupport=4] -.015656 .038952 1215.480 -.402 .688 -.092076 .060765 [SENDsupport=5] 0a 0 . . . . . [SEN4cat=1.00] -.052630 .100665 1214.467 -.523 .601 -.250126 .144866 [SEN4cat=2.00] .075143 .042900 1210.688 1.752 .080 -.009024 .159310 [SEN4cat=3.00] -.004694 .060033 1204.431 -.078 .938 -.122476 .113087 [SEN4cat=4.00] -.048734 .093058 1183.006 -.524 .601 -.231312 .133844 [SEN4cat=5.00] 0a 0 . . . . . [Bully_role_T=1] -.051567 .062081 1204.409 -.831 .406 -.173367 .070233 [Bully_role_T=2] .198703 .072671 1193.706 2.734 .006 .056125 .341281 [Bully_role_T=3] .127806 .074102 1203.630 1.725 .085 -.017577 .273188 [Bully_role_T=4] .194932 .085660 1216.991 2.276 .023 .026874 .362990 [Bully_role_T=6] 0a 0 . . . . . Year_group -.086559 .034098 1210.603 -2.539 .011 -.153456 -.019662 Gender -.091377 .035791 1212.275 -2.553 .011 -.161596 -.021158 Ethnicity .032240 .052514 1201.914 .614 .539 -.070790 .135269 FSM_eligible .076179 .037407 1194.616 2.036 .042 .002788 .149570 EngbaselineZscore1 -.051437 .018953 1216.865 -2.714 .007 -.088621 -.014253 Attendance_Year1 -.009512 .002300 1186.147 -4.136 .000 -.014024 -.005000 Rel_mean_T -.067256 .037985 1211.998 -1.771 .077 -.141779 .007267 Bully_mean_T -.065245 .041443 1216.703 -1.574 .116 -.146552 .016063 Sch_urbanicity .110492 .128896 31.321 .857 .398 -.152285 .373268 NewSchoolSize .027236 .012535 40.879 2.173 .036 .001919 .052553 Sch_FMS .009469 .008082 42.415 1.172 .248 -.006836 .025775 Sch_EAL -.004678 .003418 42.863 -1.369 .178 -.011572 .002216 Sch_SA_proportion -.001529 .005976 42.153 -.256 .799 -.013587 .010529 Sch_SAP_ST_proportion -.006798 .008694 38.665 -.782 .439 -.024387 .010792 Sch_Sec_Achievement .007641 .004011 35.239 1.905 .065 -.000499 .015782 Sch_Overall_absence .062284 .039389 34.338 1.581 .123 -.017734 .142302 Ex2 -.003318 .010710 36.958 -.310 .758 -.025020 .018384 a. This parameter is set to zero because it is redundant. b. Dependent Variable: Behaviour mean T3 (items: 2, 3, 4, 6, 7, 8).

376





Residual .290556 .012084 24.046 .000 .267812 .315231


School_No]

Variance .030266 .011332 2.671 .008 .014530 .063044




Akaike's Information Criterion (AIC) 2070.602

Hurvich and Tsai's Criterion (AICC) 2072.616


Schwarz's Bayesian Criterion (BIC) 2244.143

377

Appendix 6ci: Full Primary Model (SEND category removed)




Upper Bound

Intercept 1.137313 .281081 506.437 4.046 .000 .585085 1.689541 [Season_Of_Birth=1] .077688 .030715 1964.285 2.529 .012 .017450 .137925 [Season_Of_Birth=2] .030363 .028211 1953.978 1.076 .282 -.024965 .085690 [Season_Of_Birth=3] .041673 .028708 1932.975 1.452 .147 -.014629 .097974 [Season_Of_Birth=4] 0a 0 . . . . . [SENDsupport=3] .028793 .055682 1981.073 .517 .605 -.080407 .137994 [SENDsupport=4] .047069 .024979 2015.001 1.884 .060 -.001919 .096057 [SENDsupport=5] 0a 0 . . . . . [Bully_role_T=1] -.068953 .047573 2000.062 -1.449 .147 -.162251 .024345 [Bully_role_T=2] .241766 .052826 1997.856 4.577 .000 .138167 .345365 [Bully_role_T=3] .048644 .044380 2016.897 1.096 .273 -.038392 .135680 [Bully_role_T=4] .036876 .056220 2010.483 .656 .512 -.073378 .147131 [Bully_role_T=6] 0a 0 . . . . . XBeh_mean_T1 .465424 .025376 1985.495 18.341 .000 .415658 .515190 Year_group .090921 .023917 1984.170 3.802 .000 .044016 .137827 Gender -.091599 .022793 1956.120 -4.019 .000 -.136300 -.046899 Ethnicity -.025030 .033466 1977.150 -.748 .455 -.090663 .040602 FSM_eligible .084278 .024386 1964.932 3.456 .001 .036453 .132103 EngbaselineZscore -.008633 .011738 2014.129 -.736 .462 -.031653 .014387 Attendance_Year1 -.000774 .001851 1966.491 -.418 .676 -.004404 .002855 Rel_mean_T -.115081 .025027 2010.144 -4.598 .000 -.164163 -.065999 Bully_mean_T .016619 .030040 2014.670 .553 .580 -.042294 .075532 Sch_urbanicity -.004190 .052083 202.240 -.080 .936 -.106886 .098506 NewSchoolSize .007088 .017368 176.629 .408 .684 -.027187 .041363 Sch_FMS .000706 .001697 210.619 .416 .678 -.002640 .004051 Sch_EAL -.000326 .000857 231.275 -.380 .704 -.002016 .001363 Sch_SA_proportion -.003831 .002782 171.758 -1.377 .170 -.009322 .001660 Sch_SAP_ST_proportion .000690 .003437 184.520 .201 .841 -.006091 .007471 Sch_Pri_EngMaths4 -.005745 .001458 187.977 -3.941 .000 -.008621 -.002869 Sch_Overall_absence -.029592 .018056 170.677 -1.639 .103 -.065234 .006049 Exclsuions2 .003016 .020178 155.180 .149 .881 -.036842 .042875 a. This parameter is set to zero because it is redundant. b. Dependent Variable: Behaviour mean T3 (items: 2, 3, 4, 6, 7, 8).

378





Residual .207814 .006884 30.186 .000 .194749 .221754


School_No]

Variance .032141 .005726 5.613 .000 .022668 .045572





Criterion (AIC)

2790.444


(AICC)

2791.380


Schwarz's Bayesian

Criterion (BIC)

2958.725

The information criteria are displayed in

smaller-is-better forms.

a. Dependent Variable: Behaviour mean

T3 (items: 2, 3, 4, 6, 7, 8).

379

Appendix 6cii: Full Secondary Model (SEND category removed)


Parameter Estimate Std. Error df t Sig. 95% Confidence Interval

Lower Bound Upper Bound Intercept .237578 .601094 48.599 .395 .694 -.970618 1.445774 [Season_Of_Birth=1] .001125 .044711 1192.272 .025 .980 -.086597 .088846 [Season_Of_Birth=2] -.027406 .045095 1193.270 -.608 .543 -.115881 .061069 [Season_Of_Birth=3] -.011285 .048004 1191.046 -.235 .814 -.105467 .082897 [Season_Of_Birth=4] 0a 0 . . . . . [SENDsupport=3] -.065896 .059171 1226.976 -1.114 .266 -.181985 .050192 [SENDsupport=4] -.008828 .038034 1229.178 -.232 .816 -.083448 .065791 [SENDsupport=5] 0a 0 . . . . . [Bully_role_T=1] -.068641 .062317 1213.299 -1.101 .271 -.190902 .053621 [Bully_role_T=2] .189396 .072822 1204.943 2.601 .009 .046524 .332268 [Bully_role_T=3] .111359 .073494 1216.889 1.515 .130 -.032830 .255549 [Bully_role_T=4] .209205 .085398 1229.723 2.450 .014 .041662 .376747 [Bully_role_T=6] 0a 0 . . . . . XBeh_mean_T1 .430008 .038968 1226.339 11.035 .000 .353557 .506460 Year_group -.092143 .034151 1225.034 -2.698 .007 -.159144 -.025141 Gender -.091602 .035684 1228.127 -2.567 .010 -.161612 -.021593 Ethnicity .045803 .052588 1212.939 .871 .384 -.057371 .148978 FSM_eligible .081785 .037421 1203.592 2.186 .029 .008368 .155202 EngbaselineZscore1 -.044511 .018586 1229.714 -2.395 .017 -.080975 -.008047 Attendance_Year1 -.009652 .002275 1202.896 -4.243 .000 -.014116 -.005189 Rel_mean_T -.071807 .038058 1222.353 -1.887 .059 -.146473 .002859 Bully_mean_T -.049484 .041408 1228.367 -1.195 .232 -.130722 .031754 Sch_urbanicity .116272 .133802 30.619 .869 .392 -.156758 .389301 NewSchoolSize .031404 .012769 39.418 2.459 .018 .005585 .057222 Sch_FMS .014290 .008276 40.353 1.727 .092 -.002431 .031011 Sch_EAL -.006469 .003504 40.671 -1.846 .072 -.013547 .000610 Sch_SA_proportion -.003443 .006160 40.333 -.559 .579 -.015890 .009004 Sch_SAP_ST_proportion -.008700 .008971 36.904 -.970 .338 -.026878 .009478 Sch_Sec_Achievement .007931 .004146 33.420 1.913 .064 -.000500 .016362 Sch_Overall_absence .056311 .040521 32.714 1.390 .174 -.026156 .138778 Ex2 .004125 .010933 34.243 .377 .708 -.018088 .026338 a. This parameter is set to zero because it is redundant. b. Dependent Variable: Behaviour mean T3 (items: 2, 3, 4, 6, 7, 8).

380




Lower

Bound

Upper

Bound

Residual .295601 .012246 24.138 .000 .272547 .320605


School_No]

Variance .033710 .012538 2.689 .007 .016261 .069880





Criterion (AIC)

2107.946


(AICC)

2109.498


Schwarz's Bayesian

Criterion (BIC)

2261.389




T3 (items: 2, 3, 4, 6, 7, 8).

381

Appendix 6di: Primary Cumulative Model





Intercept .278254 .039622 2193.324 7.023 .000 .200554 .355955 XBeh_mean_T1 .533372 .016319 2640.739 32.684 .000 .501372 .565372 CENTREDCONTEXTUALRISK .089410 .015969 2646.262 5.599 .000 .058096 .120724 Year_group .090479 .021359 2522.390 4.236 .000 .048596 .132363 Gender -.090972 .020565 2574.875 -4.424 .000 -.131297 -.050647 [Season_Of_Birth=1] .049212 .027786 2578.228 1.771 .077 -.005273 .103697 [Season_Of_Birth=2] .027421 .025847 2578.994 1.061 .289 -.023262 .078104 [Season_Of_Birth=3] .031691 .026321 2553.309 1.204 .229 -.019922 .083304 [Season_Of_Birth=4] 0a 0 . . . . . a. This parameter is set to zero because it is redundant. b. Dependent Variable: Behaviour mean T3 (items: 2, 3, 4, 6, 7, 8).





Residual .231543 .006645 34.847 .000 .218880 .244940


School_No]

Variance .041831 .006059 6.904 .000 .031493 .055563





Criterion (AIC)

3923.131


(AICC)

3923.214


Schwarz's Bayesian

Criterion (BIC)

3981.988




T3 (items: 2, 3, 4, 6, 7, 8).

382

Appendix 6dii: Secondary Cumulative Model





Intercept .532441 .060216 356.225 8.842 .000 .414018 .650865

XBeh_mean_T1 .476403 .021342 1617.613 22.323 .000 .434543 .518263

Year_group -.065698 .029706 1627.993 -2.212 .027 -.123964 -.007432

Gender -.099387 .031807 1616.383 -3.125 .002 -.161774 -.037001

centerRisk .124458 .017756 1626.480 7.010 .000 .089631 .159284






Residual .324921 .011594 28.024 .000 .302973 .348460


School_No]

Variance .041775 .011928 3.502 .000 .023871 .073107





Criterion (AIC)

2878.192


(AICC)

2878.261


Schwarz's Bayesian

Criterion (BIC)

2915.958




T3 (items: 2, 3, 4, 6, 7, 8).

383

Appendix 6ei: Primary Quadratic Model




Upper Bound

Intercept .262553 .040258 2236.832 6.522 .000 .183607 .341499 XBeh_mean_T1 .528253 .016474 2643.521 32.065 .000 .495949 .560556 CENTREDCONTEXTUALRISK .066729 .019046 2645.600 3.504 .000 .029383 .104076 Year_group .088247 .021371 2526.085 4.129 .000 .046340 .130154 Gender -.089823 .020549 2572.581 -4.371 .000 -.130116 -.049529 [Season_Of_Birth=1] .049340 .027755 2577.142 1.778 .076 -.005085 .103764 [Season_Of_Birth=2] .029193 .025831 2578.214 1.130 .259 -.021459 .079846 [Season_Of_Birth=3] .030989 .026293 2552.247 1.179 .239 -.020568 .082547 [Season_Of_Birth=4] 0a 0 . . . . . SQCENTREDCONTEXTUALRISK .035319 .016216 2585.121 2.178 .029 .003521 .067118 a. This parameter is set to zero because it is redundant. b. Dependent Variable: Behaviour mean T3 (items: 2, 3, 4, 6, 7, 8).




Lower Bound

Upper

Bound

Residual .230954 .006629 34.841 .000 .218321 .244319


School_No]

Variance .042300 .006107 6.927 .000 .031875 .056133





Criterion (AIC)

3920.400


(AICC)

3920.500


Schwarz's Bayesian

Criterion (BIC)

3985.143




T3 (items: 2, 3, 4, 6, 7, 8).

384

Appendix 6eii: Secondary Quadratic Model





Intercept .502526 .062043 391.701 8.100 .000 .380547 .624505

XBeh_mean_T1 .476305 .021316 1617.552 22.344 .000 .434495 .518116

Year_group -.062812 .029707 1627.997 -2.114 .035 -.121079 -.004545

Gender -.098584 .031772 1616.497 -3.103 .002 -.160902 -.036266

centerRisk .104873 .020333 1625.134 5.158 .000 .064992 .144754

SQcenterRisk .033714 .017124 1600.419 1.969 .049 .000127 .067302






Residual .324135 .011569 28.018 .000 .302236 .347621


School_No]

Variance .041756 .011982 3.485 .000 .023794 .073276





Criterion (AIC)

2876.320


(AICC)

2876.409


Schwarz's Bayesian

Criterion (BIC)

2919.481




T3 (items: 2, 3, 4, 6, 7, 8).

385

Appendix 6fi: Primary Independent Additive Model





Intercept .224251 .039120 2149.915 5.732 .000 .147534 .300968

[Season_Of_Birth=1] .049212 .027786 2578.228 1.771 .077 -.005273 .103697

[Season_Of_Birth=2] .027421 .025847 2578.994 1.061 .289 -.023262 .078104

[Season_Of_Birth=3] .031691 .026321 2553.309 1.204 .229 -.019922 .083304

[Season_Of_Birth=4] 0a 0 . . . . .

XBeh_mean_T1 .533372 .016319 2640.739 32.684 .000 .501372 .565372

Year_group .090479 .021359 2522.390 4.236 .000 .048596 .132363

Gender -.090972 .020565 2574.875 -4.424 .000 -.131297 -.050647

CONTEXTUALRISK .089410 .015969 2646.262 5.599 .000 .058096 .120724

a. This parameter is set to zero because it is redundant.






Residual .231543 .006645 34.847 .000 .218880 .244940


School_No]

Variance .041831 .006059 6.904 .000 .031493 .055563





Criterion (AIC)

3923.131


(AICC)

3923.214


Schwarz's Bayesian

Criterion (BIC)

3981.988




T3 (items: 2, 3, 4, 6, 7, 8).

386

Appendix 6fii: Secondary Independent Additive Model




Lower Bound

Upper

Bound

Intercept 1.522877 .223744 1309.923 6.806 .000 1.083941 1.961813

[Bully_role_T=1] -.080734 .049362 1285.120 -1.636 .102 -.177573 .016105

[Bully_role_T=2] .214530 .070693 1282.979 3.035 .002 .075844 .353215

[Bully_role_T=3] .076180 .064224 1296.231 1.186 .236 -.049813 .202174

[Bully_role_T=4] .210760 .085067 1308.127 2.478 .013 .043878 .377642

[Bully_role_T=6] 0a 0 . . . . .

XBeh_mean_T1 .431154 .032716 1313.231 13.179 .000 .366972 .495337

Year_group -.096129 .033171 1314.000 -2.898 .004 -.161202 -.031056

Gender -.100187 .034857 1312.058 -2.874 .004 -.168569 -.031805

EngbaselineZscore1 -.040429 .017330 1292.896 -2.333 .020 -.074428 -.006431

FSM_eligible .076624 .036011 1310.470 2.128 .034 .005977 .147270

Attendance_Year1 -.010862 .002207 1285.708 -4.922 .000 -.015191 -.006533







Residual .302653 .012105 25.002 .000 .279833 .327333


School_No]

Variance .051420 .016003 3.213 .001 .027940 .094635





Criterion (AIC)

2258.186


(AICC)

2258.466


Schwarz's Bayesian

Criterion (BIC)

2325.537




T3 (items: 2, 3, 4, 6, 7, 8).

387

Appendix 6gi: Primary Protective Model





Intercept .285549 .188754 1965.625 1.513 .130 -.084629 .655727

[Season_Of_Birth=1] .084246 .030429 1961.313 2.769 .006 .024569 .143922

[Season_Of_Birth=2] .036923 .027989 1951.627 1.319 .187 -.017967 .091814

[Season_Of_Birth=3] .043333 .028435 1931.797 1.524 .128 -.012433 .099099


[SENDsupport=3] .037653 .055344 1982.022 .680 .496 -.070885 .146191

[SENDsupport=4] .048090 .024924 2013.290 1.929 .054 -.000790 .096971

[SENDsupport=5] 0a 0 . . . . .

[Bully_role_T=1] -.071868 .047307 2001.400 -1.519 .129 -.164644 .020907

[Bully_role_T=2] -.037824 .183102 2016.722 -.207 .836 -.396913 .321265

[Bully_role_T=3] .049958 .044109 2016.976 1.133 .258 -.036547 .136462

[Bully_role_T=4] .047498 .055620 2010.563 .854 .393 -.061581 .156576

[Bully_role_T=6] 0a 0 . . . . .

XBeh_mean_T1 .456211 .025253 1986.117 18.066 .000 .406687 .505736

Year_group .086012 .023802 1981.212 3.614 .000 .039331 .132692

Gender -.087786 .022655 1958.386 -3.875 .000 -.132216 -.043355

Ethnicity -.026981 .033220 1976.528 -.812 .417 -.092132 .038169

EngbaselineZscore -.004056 .011723 2013.936 -.346 .729 -.027047 .018935

Attendance_Year1 -.000171 .001842 1964.992 -.093 .926 -.003783 .003441

Bully_mean_T .020985 .029872 2014.717 .702 .482 -.037599 .079569

FSM_eligible .008773 .091506 2016.876 .096 .924 -.170683 .188229

BullyRISK 0a 0 . . . . .

Sch_urbanicity -.037853 .056664 276.675 -.668 .505 -.149400 .073694

CSize -.006088 .018786 235.700 -.324 .746 -.043097 .030922

CFSM .001589 .001999 365.977 .795 .427 -.002343 .005520

CEAL -.000218 .000972 358.938 -.224 .823 -.002129 .001694

CSEND -.004641 .003033 244.347 -1.530 .127 -.010615 .001334

CSEND2 -.002446 .003883 288.396 -.630 .529 -.010088 .005196

CAcad -.005535 .001662 303.405 -3.329 .001 -.008806 -.002263

CATT -.037541 .020384 269.955 -1.842 .067 -.077673 .002591

CEX .016263 .022846 246.746 .712 .477 -.028735 .061262

CPOSrel -.194639 .076219 2013.983 -2.554 .011 -.344115 -.045163

FSM_eligible *

Sch_urbanicity

.080842 .077211 2016.971 1.047 .295 -.070580 .232265

FSM_eligible * CSize .043812 .024343 2015.004 1.800 .072 -.003928 .091552

388

FSM_eligible * CFSM -.001587 .002314 2014.510 -.686 .493 -.006125 .002951

FSM_eligible * CEAL -.001064 .000984 2007.988 -1.082 .280 -.002995 .000866

FSM_eligible *

CSEND

.000914 .003459 2004.389 .264 .792 -.005870 .007698

FSM_eligible *

CSEND2

.007187 .004608 2015.666 1.560 .119 -.001849 .016223

FSM_eligible * CAcad -.000515 .001945 2016.999 -.265 .791 -.004329 .003299

FSM_eligible * CATT .014793 .023305 2014.323 .635 .526 -.030911 .060496

FSM_eligible * CEX -.023733 .025363 2016.969 -.936 .350 -.073474 .026008

Sch_urbanicity *

CPOSrel

.060428 .062541 2014.822 .966 .334 -.062223 .183079

CSize * CPOSrel -.025691 .020292 1974.621 -1.266 .206 -.065488 .014106

CFSM * CPOSrel -.000976 .002037 1986.770 -.479 .632 -.004970 .003018

CEAL * CPOSrel -.001219 .000924 2016.992 -1.320 .187 -.003031 .000592

CSEND * CPOSrel .008778 .003050 2012.291 2.879 .004 .002798 .014759

CSEND2 * CPOSrel -.002222 .004013 2016.879 -.554 .580 -.010092 .005648

CAcad * CPOSrel .004427 .001770 2010.434 2.501 .012 .000956 .007898

CATT * CPOSrel .029828 .020419 2015.152 1.461 .144 -.010217 .069873

CEX * CPOSrel -.040606 .022005 1972.477 -1.845 .065 -.083762 .002550

BullyRISK *

Sch_urbanicity

.236210 .147408 2002.181 1.602 .109 -.052879 .525299

BullyRISK * CSize -.001668 .051231 1995.454 -.033 .974 -.102139 .098803

BullyRISK * CFSM .000545 .004411 1999.492 .124 .902 -.008105 .009195

BullyRISK * CEAL .001233 .001927 1971.129 .640 .522 -.002546 .005011

BullyRISK * CSEND .003180 .006662 1939.625 .477 .633 -.009885 .016245

BullyRISK * CSEND2 .000338 .008908 1995.770 .038 .970 -.017132 .017809

BullyRISK * CAcad -.000610 .003866 1987.199 -.158 .875 -.008192 .006972

BullyRISK * CATT .039287 .046932 1946.111 .837 .403 -.052755 .131328

BullyRISK * CEX -.084514 .045662 1921.610 -1.851 .064 -.174066 .005038



389




Lower Bound

Upper

Bound

Residual .202101 .006701 30.160 .000 .189385 .215671


School_No]

Variance .031831 .005677 5.607 .000 .022442 .045150








The information criteria are displayed in smaller-is-

better forms.

a. Dependent Variable: Behaviour mean T3 (items: 2,

3, 4, 6, 7, 8).

390

Appendix 6gii: Secondary Protective Model




Lower Bound

Upper

Bound

Intercept .716654 .119024 943.622 6.021 .000 .483072 .950235

[Season_Of_Birth=1] .008537 .044821 1188.098 .190 .849 -.079400 .096475

[Season_Of_Birth=2] -.024705 .045162 1187.915 -.547 .584 -.113312 .063902

[Season_Of_Birth=3] -.007168 .048039 1188.581 -.149 .881 -.101420 .087083


[SENDsupport=3] -.051924 .059777 1229.004 -.869 .385 -.169199 .065352

[SENDsupport=4] -.012711 .038204 1221.134 -.333 .739 -.087664 .062243

[SENDsupport=5] 0a 0 . . . . .

[Bully_role_T=1] -.074142 .062465 1221.556 -1.187 .235 -.196693 .048410

[Bully_role_T=2] .212676 .077612 1209.699 2.740 .006 .060407 .364946

[Bully_role_T=3] .119350 .073943 1222.004 1.614 .107 -.025719 .264418

[Bully_role_T=4] .236203 .099274 1229.909 2.379 .017 .041438 .430968

[Bully_role_T=6] 0a 0 . . . . .

XBeh_mean_T1 .423077 .039189 1227.273 10.796 .000 .346192 .499962

Year_group -.117075 .035509 1214.518 -3.297 .001 -.186741 -.047408

Gender -.113202 .035637 1219.542 -3.176 .002 -.183120 -.043284

Ethnicity .022269 .053202 1216.424 .419 .676 -.082110 .126647

FSM_eligible .084368 .043199 1217.934 1.953 .051 -.000384 .169120

EngbaselineZscore1 -.026982 .021422 1227.501 -1.260 .208 -.069009 .015045

Rel_mean_T -.065647 .038212 1224.982 -1.718 .086 -.140615 .009322

Bully_mean_T -.042949 .041721 1229.797 -1.029 .303 -.124801 .038903

RISK_BULLY 0a 0 . . . . .

RISK_BYSTANDER 0a 0 . . . . .

Ccsize .020158 .013329 55.466 1.512 .136 -.006549 .046864

Sch_urbanicity .123556 .130158 32.755 .949 .349 -.141329 .388441

CFSM .009763 .009104 71.035 1.072 .287 -.008389 .027915

CEAL -.003575 .003940 78.625 -.907 .367 -.011417 .004268

Csend -.003989 .006856 66.637 -.582 .563 -.017675 .009697

Csend2 -.008882 .009340 53.998 -.951 .346 -.027607 .009842

Cac .007189 .004324 50.337 1.663 .103 -.001495 .015873

Cab .045848 .042245 48.177 1.085 .283 -.039083 .130779

Cex -.000477 .011319 49.359 -.042 .967 -.023220 .022266

CAttendance -.009390 .002599 1207.880 -3.613 .000 -.014488 -.004291

Cex * CAttendance -8.257263E-5 .000915 1215.081 -.090 .928 -.001877 .001712

Cab * CAttendance .002610 .003154 1226.265 .828 .408 -.003577 .008797

Cac * CAttendance .000195 .000369 1213.659 .529 .597 -.000529 .000919

Csend2 * -1.102817E-5 .000739 1229.783 -.015 .988 -.001460 .001438

391

CAttendance

Csend * CAttendance -.000420 .000525 1227.672 -.799 .424 -.001450 .000611

CEAL * CAttendance -.000156 .000313 1224.887 -.498 .619 -.000771 .000459

CFSM * CAttendance .000328 .000779 1229.950 .421 .674 -.001200 .001856

Sch_urbanicity *

CAttendance

.004062 .008340 1188.241 .487 .626 -.012300 .020425

Ccsize * CAttendance .000932 .001098 1221.287 .849 .396 -.001222 .003085

EngbaselineZscore1 *

Sch_urbanicity

-.137960 .066580 1219.761 -2.072 .038 -.268583 -.007337


Ccsize

-.002445 .008508 1219.126 -.287 .774 -.019137 .014247


CFSM

.002192 .005885 1158.384 .372 .710 -.009355 .013739


CEAL

-.000728 .002365 1167.403 -.308 .758 -.005367 .003911


Csend

-.006078 .003893 1082.035 -1.561 .119 -.013717 .001561


Csend2

-.002234 .005274 1215.298 -.424 .672 -.012581 .008113


Cac

-.000786 .002700 1140.942 -.291 .771 -.006083 .004512


Cab

-.017310 .021613 1186.814 -.801 .423 -.059714 .025094


Cex

.005888 .007892 1184.646 .746 .456 -.009597 .021372

FSM_eligible * Ccsize -.006012 .017850 1229.606 -.337 .736 -.041032 .029008

FSM_eligible *

Sch_urbanicity

-.090528 .134267 1204.195 -.674 .500 -.353952 .172896

FSM_eligible * CFSM -.000528 .011405 1159.804 -.046 .963 -.022905 .021848

FSM_eligible * CEAL -.001998 .004565 1197.816 -.438 .662 -.010953 .006958

FSM_eligible * Csend -.008825 .007845 1160.034 -1.125 .261 -.024217 .006567

FSM_eligible *

Csend2

.001489 .011075 1224.183 .134 .893 -.020239 .023216

FSM_eligible * Cac .000376 .005245 1226.970 .072 .943 -.009915 .010666

FSM_eligible * Cab .062287 .048280 1215.339 1.290 .197 -.032435 .157009

FSM_eligible * Cex -.000398 .014327 1229.551 -.028 .978 -.028507 .027711

RISK_BULLY * Ccsize .011666 .024764 1222.389 .471 .638 -.036919 .060250

RISK_BULLY *

Sch_urbanicity

-.075637 .188836 1225.289 -.401 .689 -.446114 .294841

RISK_BULLY * CFSM .014457 .015303 1227.727 .945 .345 -.015567 .044480

RISK_BULLY * CEAL -.003304 .006874 1229.293 -.481 .631 -.016791 .010182

RISK_BULLY * Csend .006028 .012036 1221.568 .501 .617 -.017586 .029642

RISK_BULLY * -.011769 .015528 1229.874 -.758 .449 -.042233 .018696

392

Csend2

RISK_BULLY * Cac .005669 .007611 1211.635 .745 .457 -.009264 .020602

RISK_BULLY * Cab -.020030 .070095 1211.863 -.286 .775 -.157551 .117491

RISK_BULLY * Cex .002786 .026274 1229.614 .106 .916 -.048760 .054332

RISK_BYSTANDER *

Ccsize

.050363 .030358 1229.551 1.659 .097 -.009196 .109923

RISK_BYSTANDER *

Sch_urbanicity

.093103 .286863 1194.732 .325 .746 -.469708 .655914

RISK_BYSTANDER *

CFSM

-.010596 .028641 1197.604 -.370 .711 -.066787 .045596

RISK_BYSTANDER *

CEAL

.010573 .014107 1210.156 .750 .454 -.017103 .038249

RISK_BYSTANDER *

Csend

.020762 .020298 1093.631 1.023 .307 -.019065 .060588

RISK_BYSTANDER *

Csend2

.052299 .029225 1225.056 1.790 .074 -.005038 .109636

RISK_BYSTANDER *

Cac

.006036 .012359 1224.487 .488 .625 -.018212 .030284

RISK_BYSTANDER *

Cab

.081053 .115436 1226.805 .702 .483 -.145420 .307526

RISK_BYSTANDER *

Cex

.046555 .027322 1227.037 1.704 .089 -.007048 .100158



393



95% Confidence

Interval

Lower

Bound

Upper

Bound

Residual .287076 .011928 24.068 .000 .264624 .311433


School_No]

Variance .025725 .010937 2.352 .019 .011181 .059188








The information criteria are displayed in smaller-is-

better forms.

a. Dependent Variable: Behaviour mean T3 (items:

2, 3, 4, 6, 7, 8).

394

Appendix 7a: Data assumptions and requirements (Primary Model)

395

Appendix 7b: Data assumptions and requirements (Secondary Model)

396

Appendix 8: MISSING DATA ANALYSIS Missing data was analysed through SPSS missing value analysis. Missing data was acknowledged so assurances can be made they are not having an undue effect or biasing the results in any way. There may well be specific reasons why data has not been completed across certain variables e.g. in the present study a pattern could emerge with missing data on both the behaviour and bullying scales, as teachers completing them could have disliked the negative wording within these constructs. If such patterns exist and are not acknowledged this could lead to inaccurate generalisations. Patterns in missing data across the continuous variables can be assessed using Little’s (1988) missing completely at random (MCAR) test. If data were MCAR then it can be assumed no predictable patterns in missing data have emerged and there is no undue influence on the findings. If data are not MCAR then some explanation is required.

A.8.1 Phase 1: The sample taken for phase 1 of the analysis included the optimum sample of 13223 participants (6377 in primary schools, 6846 in secondary schools). The purpose of the analysis was to assess whether those who had a valid survey at baseline (8375 participants, 4609 in primary and 3766 in secondary) differed from those who did not. Analyses were conducted to assess whether the mean of the continuous variables differed when the survey at baseline was present or absent, and also whether those with a survey at baseline present differed from those without one across the different groups within the categorical variables52

.

Table A.8.1: Means and standard deviations (in parentheses) for the continuous predictor variables when the survey was present and absent at baseline, in primary and secondary schools. (Cohen’s d showing effect sizes differences). PRIMARY SCHOOLS SECONDARY SCHOOLS

Variable Survey

Present at Baseline

Survey Absent at Baseline

Cohen’s d Survey Present at Baseline


Cohen’s d

School Size

334.03 (188.90)

336.01 (165.55)

0.01 1187.71 (384.97)

1235.93 (378.14)

0.13

School % FSM

25.93 (15.66)

25.73 (16.07)

0.01 19.41 (10.52)

20.93 (11.62)

0.14

School % EAL

21.72 (28.15)

25.78 (30.18)

0.14 14.29 (20.02)

18.60 (24.37)

0.19

School % SA

14.06 (7.05)

14.30 (7.95)

0.03 16.82 (7.79)

17.60 (7.92)

0.10

School % SAP & ST

10.41 (5.62)

9.59 (4.72)

0.16 11.34 (5.79)

9.67 (4.77)

0.31

School Academic Attainment

69.46 (15.17)

68.59 (15.64)

0.06 45.24 (11.69)

44.14 (13.39)

0.09

School Overall Absence

6.13 (2.72)

6.31 (4.11)

0.05 7.84 (1.09)

8.18 (1.20)

0.30

School Exclusions

- - - - - -

52 This analysis included all variables for which data were collected independently from the survey

397

Individual Academic Achievement

12.29 (5.23)

12.80 (5.46)

0.10 26.23 (7.92)

26.78 (7.56)

0.07

Individual Attendance

93.12 (6.38)

92.61 (8.23)

0.07 91.22 (10.03)

91.15 (10.57)

0.01

Table A.8.2. Differences in observed and expected counts across different levels of the categorical predictor variables when the survey was present and absent at baseline in primary and secondary schools. (Phi or Cramer’s V showing effect size differences). PRIMARY SCHOOLS SECONDARY SCHOOLS

Variable Survey

Present at Baseline


53

Cramer’s V

Phi/ Survey Present at Baseline


Phi/ Cramer’s V

Urbanicity Urban 4090 (4125.9)

1575 (1539.1)

0.04 3914 (3846.0)

2346 (2414.0)

0.07

Rural 553 (517.1)

157 (192.9)

292 (360.0)

294 (226.0)

Year group

Year 1/7 2000 (2050.3)

816 (765.7)

0.04 2133 (2110.4)

1302 (1324.6)

0.01

Year 5/10 2643 (2592.7)

918 (968.3)

2073 (2095.6)

1338 (1315.4)

Gender

Male 3073 (3049.7)

1115 (1138.3)

0.02 2605 (2565.0)

1570 (1610.0)

0.03

Female 1570 (1593.3)

618 (594.7)

1601 (1641.0)

1070 (1030.0)

Season of Birth

Autumn 937 (924.7)

333 (345.3)

0.02 938 (906.8)

538 (569.2)

0.03

Winter 1062 (1074.7)

414 (401.3)

950 (978.7)

643 (614.3)

Spring 1209 (1202.8)

443 (449.2)

1104 (1124.3)

726 (705.7)

Summer 1435 (1440.9)

544 (538.1)

1214 (1196.2)

733 (750.8)

FSM yes 1637 (1625.1)

572 (583.9)

0.01 1191 (1194.0)

749 (746.0)

0.01

no 2994 (3005.9)

1092 (1080.1)

3010 (3007.0)

1876 (1879.0)

Ethnicity White British 3529 (3467.9)

1185 (1246.1)

0.05 3475 (3405.9)

2059 (2128.1)

0.05

Other 1102 (1163.1)

479 (417.9)

726 (795.1)

566 (496.9)

53 Phi was used for categorical variables with 2 groups, Cramer’s V was used for variables with more than 2 groups

398

As can be seen from table A.8.1 the means between variables when the survey was present and absence are comparable, as Cohen’s d for all variables equates to small or less than small effect sizes54. From table A.8.2 the difference between the observed and expected values across the different levels of categorical variables when the survey was present and absent across are considered equal as values of Phi or Cramer’s V are lower than what are considered to be small effects 55

. These results show that the difference between the optimum sample and the sample with a baseline survey completed can therefore be considered comparable.

A.8.2 Phase 2: The sample taken for phase 2 of the analysis included only those with a valid survey at baseline, (8375 cases, 4609 in primary and 3766 in secondary). The purpose of the analysis was to assess whether those who had a valid survey at follow-up (4288 participants, 2788 in primary and 1683 in secondary) differed from those who did not. Analyses were calculated to assess whether the mean of the continuous variables differed when the survey at follow-up was present or absent, and also whether those with a survey at follow-up differed from those without one across the different groups within the categorical variables56

.

Table A.8.3: Means and standard deviations (in parentheses) for the continuous predictor variables when the survey was present and absent at follow-up in primary and secondary schools. (Cohen’s d showing effect sizes differences). PRIMARY SCHOOLS SECONDARY SCHOOLS

Variable Survey

Present at Baseline


Cohen’s d Survey Present at Baseline


Cohen’s d

School Size

330.89 (210.93)

338.96 (147.96)

0.04 1073.4 (365.04)

1292.14 (372.92)

0.59

School % FSM

26.02 (16.08)

25.76 (14.97)

0.02 20.06 (10.60)

18.79 (10.41)

0.12

School % EAL

21.12 (28.29)

22.65 (27.92)

0.05 14.00 (19.31)

14.55 (20.67)

0.03

School % SA

14.53 (7.09)

13.31 (6.92)

0.17 15.72 (6.92)

17.82 (8.38)

0.27

School % SAP & ST

10.51 (5.83)

10.26 (5.26)

0.04 11.02 (5.60)

11.62 (5.94)

0.10

School Academic Attainment

68.44 (15.37)

71.04 (14.71)

0.17 46.80 (12.42)

43.82 (10.79)

0.26

School Overall Absence

6.23 (3.30)

5.96 (1.30)

0.11 7.90 (1.07)

7.79 (1.09)

0.10

School

-

-

-

-

-

-

54 According to Cohen (1992) small effect sizes for these analyses are considered to be values of 0.20, medium effects, values of 0.50, and large effects, values of 0.80. 55 According to Cohen (1992) small effect sizes for these analyses are considered to be values of 0.10, medium effects, values of 0.30, and large effects, values of 0.50. 56 Within this analysis the same variables were included as of phase 1 with the addition of variables measured within the survey at baseline

399

Exclusions Individual Academic Achievement

12.33 (5.23)

12.21 (5.23)

0.02 25.02 (7.65)

27.26 (8.00)

0.29

Individual Attendance

93.22 (6.37)

92.96 (6.39)

0.04 91.98 (8.97)

90.48 (10.91)

0.15


2.06 (0.55)

2.01 (0.55)

0.09 2.06 (0.58)

2.12 (0.60)

0.10

Bullying 0.55 (0.59)

0.60 (0.62)

0.08

0.50 (0.66)

0.59 (0.66)

0.26

Behaviour Baseline

0.63 (0.71)

0.69 (0.72)

0.08

0.58 (0.77)

0.67 (0.77)

0.26

Table A.8.4. Differences in observed and expected counts across difference levels of the categorical predictor variables when the survey was present and absent at follow-up in primary and secondary schools. (Phi or Cramer’s V showing effect size differences). PRIMARY SCHOOLS SECONDARY SCHOOLS

Variable Survey

Present at Baseline


57

Cramer’s V

Phi/ Survey Present at Baseline


Phi/ Cramer’s V

Urbanicity Urban 244 (2492.1)

1650 (1597.9)

0.07 1825 (1868.6)

2089 (2045.4)

0.08

Rural 389 (336.9)

164 (216.1)

183 (139.4)

109 (152.6)

Year group

Year 1/7 1199 (1218.6)

801 (781.4)

0.02 1076 (1018.3)

1057 (1114.7)

0.06

Year 5/10 1630 (1610.4

1013 (1032.6)

932 (989.7)

1141 (1083.3)

Gender

Male 1838 (1872.4)

1235 (1200.6)

0.03 1142 (1243.7)

1463 (1361.3)

0.10

Female 991 (956.6)

579 (613.4)

866 (764.3)

735 (836.7)

Season of Birth

Autumn 574 (570.9)

363 (366.1)

0.03 468 (447.8)

470 (490.2)

0.03

Winter 667 (647.1)

395 (414.9)

428 (453.5)

522 (496.5)

Spring 742 (736.6)

467 (472.4)

534 (527.1)

570 (576.9)

Summer 846 (874.4)

589 (560.6)

578 (579.6)

636 (634.4)

FSM yes 987 (999.7)

650 (637.3)

0.01 580 (568.4)

611 (622.6)

0.01

no 1841 (1828.3)

1153 (1165.7)

1425 (1436.6)

1585 (1573.4)

Ethnicity White British 2165 1364 0.01 1678 1797 0.03 57 Phi was used for categorical variables with 2 groups, Cramer’s V was used for variables with more than 2 groups

400

(2155.0) (1374.0) (1658.5) (1816.5) Other 663

(673.0) 439 (429.0)

327 (346.5)

399 (379.5)

SEND Category

Cognitive 1583 (1566.1)

1003 (1019.9)

0.05 1118 (1121.2)

1219 (1215.8)

0.06

Behaviour 399 (419.1)

293 (272.9)

447 (448.6)

488 (486.4)

Communication

541 (557.2)

379 (362.8)

194 (187.6)

197 (203.4)

Physical 112 (107.2)

65 (69.8)

124 (105.1)

95 (113.9)

Other 106 (91.4)

45 (59.6)

96 (114.7)

143 (124.3)

SEN Support

School Action 1693 (1651.3)

1014 (1055.7)

0.04 1024 (1077.7)

1231 (1177.3)

0.10

School Action+

914 (945.5)

636 (604.5)

676 (631.8)

646 (690.2)

Statement 163 (173.2)

121 (110.8)

222 (183.5)

134 (104.9)

Bully Role Victim 208 (218.4)

147 (136.6)

0.09 209 (226.9)

256 (238.1)

0.15

Bully 160 (179.0)

131 (112.0)

160 (174.2)

197 (182.8)

Bully-victim 318 (340.8)

236 (213.2)

219 (238.1)

269 (249.9)

Bystander 104 (131.6)

110 (82.4)

68 (94.6)

126 (99.4)

Not Involved 1861 (1781.3)

1035 (114.7)

1118 (980.1)

891 (1028.9)

As can be seen from table A.8.3 the means between variables when the survey was present and absence are comparable, as Cohen’s d for all variables equates to small or less than small effect sizes58. The only notable exception is a medium effect was established for school size in secondary schools, with pupils attending larger schools less likely to have a survey completed at follow-up. From table A.8.4 the difference between the observed and expected values across the different levels of categorical variables when the survey at follow-up was present and absent across are considered equal as values of Phi or Cramer’s V are considered lower than that is considered to be small effects59

. These results show that the difference between the sample with solely a valid survey at baseline and those with a valid survey at baseline and follow-up are comparable.

58 According to Cohen (1992) small effect sizes for these analyses are considered to be values of 0.20, medium effects, values of 0.50, and large effects, values of 0.80. 59 According to Cohen (1992) small effect sizes for these analyses are considered to be values of 0.10, medium effects, values of 0.30, and large effects, values of 0.50.

401

A.8.3 Phase 3: The sample taken for this analysis included all participants who had a valid survey at baseline and follow-up, resulting in a total sample of 4471 (2788 in primary schools and 1683 in secondary schools). The aim of this phase was to assess whether there were any meaningful patterns in missing data across all predictor variables. Table A.8.5. The percentages of missing data for each of the predictor variables across school type VARIABLES

PRIMARY SCHOOLS % of missing data

SECONDARY SCHOOLS %of missing data

SCHOOL LEVEL Urbanicity 0.0 0.0 School size 0.4 0.0 School %FSM 1.8 0.0 School & EAL 1.8 0.0 School % SA 6.7 1.1 School % SAP & ST 6.7 1.1 School Achievement 10.3 1.1 School Absence 7.0 1.1 School Exclusions 0.0 0.0 INDIVIDUAL LEVEL Year group 0.0 0.0 Season of Birth 0.0 0.0 Gender 0.0 0.0 Ethnicity 0.0 0.1 Free school meals 0.0 0.1 SEND Category 3.2 0.2 SEND support 2.0 0.8 Attendance 2.3 0.7 Academic Achievement 5.2 9.7 Positive Relationships 0.5 1.2 Bullying 1.1 5.1 Bully role 5.9 2.7 As can be seen from the above table in the primary school data set 6 variables had 5% or more missing data these included School % SA and School % SAP & ST which were both missing 6.7%, school level academic achievement missing 10.3%, School absence missing 7.0%, individual academic achievement missing 5.2% and role in bullying incidences missing 5.9%. In the secondary data set only 2 variables were missing over 5% these were individual academic achievement and bullying mean score. It has been noted however, that what is potentially more important is not the amount of missing data per se but the patterns within that data that have the most influence on findings (Tabachnick & Fidell 2007). A pattern analysis was therefore conducted in order to assess whether there were any meaningful patterns in missing data, i.e. whether missing data were related across variables.

402

The pattern analysis revealed that in primary schools about 5% of cases were missing on the school level variables of % of SA, % of SAP & ST, School absence and School academic achievement. It is likely that it was the same schools where this data were unavailable, i.e. a failure by the school to return this information, the pattern is therefore unlikely to be related to individual pupils, but only to the school pupils are attending. No pattern over 5% was found within the secondary data set. The final analysis therefore involved conducting Little’s (1988) test whether these patterns of missing data are missing completely at random or not. This test revealed data for primary schools were not missing completely at random (χ₂ =2195.609, (df 131), p < .001). This is not a surprising finding from the information provided above about patterns of missing data across schools, rather than for specific characteristics of pupils. For secondary schools the test also revealed data were also not missing completely at random (χ₂ =446.678, (df 95), p < .001). This is perhaps a more surprising finding as no pattern in missing data in secondary schools was observed for over 5% of cases. Nonetheless patterns of missing data could have emerged more frequently at values of less than 5% of cases, and in combination these could have resulted in data not being missing completely at random. The results from these analyses reveal that although the missing data was not MCAR, it is likely to be a product of schools not completing and returning the school level data rather than being related to a specific individual. Furthermore with the majority of missing data less than 5% and only a few notable patterns in missing data it is unlikely that missing data has had an excessive influence on the results.

Behaviour difficulties in children with special education needs ...

Documents

Transcript of Behaviour difficulties in children with special education needs ...