Journal of Environmental Psychology (1999) 19, 117^129 0272-4944/99/020117+13 $30.00/0# 1999 Academic PressArticle No. jevp.1998.0108, available online at http://www.idealibrary.com on

DEFENSIBLE SPACE THEORYAND THE POLICE: ASSESSING THEVULNERABILITY OF RESIDENCES TO BURGLARY

KATHLEEN A. HAM-ROWBOTTOM, ROBERT GIFFORD AND KELLY T. SHAW

University of V|ctoria, Canada

Abstract

This study examines how defensible space theory and recent modi¢cations to it apply to assessments of single-family dwelling vulnerability to burglary by police o¤cers. Fifty photographs of detached houses were scoredon ¢fty-¢ve speci¢c physical cues in six defensible space cue categories. Forty-one police o¤cers reliably ratedthe vulnerability of the houses to burglary. A modi¢ed Brunswik lens model was used to determine which cuecategories police o¤cers use to assess vulnerability. Simple and partial correlation analyses show how roadand occupant surveillability, actual barriers, traces of occupancy, and house value are related to vulnerabilityassessments by police; symbolic barriers are unrelated to their assessments. Police assessments of vulnerabil-ity are, based on previous research, very similar to those of residents, but less similar to those of burglars.Suggestions for a more parsimonious and pluralistic defensible space theory are made.

# 1999 Academic Press

Introduction

The physical and social attributes of residences andsurrounding neighborhoods in£uence the percep-tions of both residents and visitors. In order to forman impression of a dwelling's security or vulnerabil-ity to crime, observers employ visible cues. Basedon ideas expressed by Jacobs (1961), Newman (1972)formulated defensible space theory, which holds thatcertain features of physical settings, such as indica-tors of territory and surveillance opportunities, canreduce crime. Newman proposed that environmen-tal designers can strategically use these features todefend an area against crime.

Initial research into defensible space theoryattempted to ¢nd direct links between physicaldesign features and crime rates (e.g. Newman, 1972;Pablant & Baxter, 1975) but met with mixed results.More recently, researchers have focused on the roleof mediating variables, such as perceptions orassessments of physical design features (Brower etal., 1983; McGahan, 1984; Macdonald & Gi¡ord,1989; Brown & Bentley, 1993; Shaw & Gi¡ord, 1994).The idea is that design features in£uence crime bycreating impressions in observers which rangefrom assessments that a given setting is defensible

against crime to assessments that it is vulnerableto crime.

An apparent assumption of defensible space theo-ry (Newman, 1972) is that all observers will employthe same physical cues to reach the same conclu-sions about a setting's defensibility. Following Mac-donald and Gi¡ord (1989), and Shaw and Gi¡ord(1994), we examine this proposition. The presentstudy has three main purposes. The ¢rst is to deter-mine whether or not the design features that defen-sible space theory (including some recent extensionsof it) proposes as indicators of vulnerability predictvulnerability assessments by police o¤cers. As issuehere is whether the evidence will support the theoryas it is, or whether it requires further modi¢cation.

The second purpose is to determine whether ornot the design features are weighted similarly byobservers who all play relevant parts in the crim-inal justice context, but have di¡erent experiencesand backgrounds (i.e. police, burglars, and resi-dents). This purpose concerns the universality ofthe theory's proposed indicators of vulnerability.

It is important to know how noncriminals assesssetting vulnerability. For example, when crime pre-vention programs are implemented the police per-spective must be understood so that o¤cers can

118 K. A. Ham-Rowbottom et al.

recognize the di¡erences, if any, between their ownway of assessing vulnerability and that of criminals.If the vulnerability assessment policies of police andcriminals di¡er, but police o¤cers assume that theirown assessment policies are the same as those ofcriminals, then o¤cers in crime prevention pro-grams may mis-advise residents by teaching themto attend to the wrong set of house and propertyfeatures.

If observer groups do di¡er Ð for example, if po-lice, burglars, and residents assess vulnerability dif-ferently Ð then architectural designs intended todeter crime must take these di¡erences into ac-count. If assessment di¡erences do exist, crime pre-vention will not be maximized by architecturaldesigns that re£ect residents' or police assessmentsof the setting's vulnerability, but rather by design-ing settings to re£ect the criminal's perspective.After all, the ultimate purpose of defensible spacetheory is to help deter criminal actions.

Therefore, this study examines the way policeo¤cers assess residential vulnerability to burglary.Understanding their perspective will permit com-parisons of the assessment policies of three majorgroups in the crime context: burglars, residents,and the police. Ideally, these comparisons shouldnot be confounded by di¡erences in the physicalsettings to be assessed. This study facilitates thisgoal by employing the same set of residencespreviously used to determine the assessment policesof burglars (Macdonald & Gi¡ord, 1989) and resi-dents (Shaw & Gi¡ord, 1994). Thus, our third pur-pose is to identify any di¡erences in assessmentpolicy between police, residents, and burglars inorder to inform and facilitate crime preventionprograms.

Defensible space theory and recently proposedmodi¢cations

Defensible spaces are said to possess several keyphysical elements. Two territorial factors, accordingto the theory, are actual and symbolic barriers. Ac-tual barriers are physical impediments to access,such as high fences and hedges. Symbolic barriersare physical characteristics of a house, such as or-namentation and nameplates, that indicate personalproperty and that the owners care about and areprepared to defend their residence.

The theory also predicts that the opportunity forsurveillance of the residence has an impact on theperception of vulnerability to burglary. Highly sur-veillable residences a¡ord a clear view of the prop-erty from both the house and street, which

increases security.When surveillance is di¤cult, in-creased concealment opportunities are presumed toincrease the dwelling's vulnerability to burglary.

Since Newman's formulation of defensible spacetheory, various modi¢cations have been proposed.For example, Brown and Altman (1981, 1983) andBrown (1985) suggested adding traces of occupancy,that is, environmental cues which indicate thepresence or absence of occupants. Dwellingsthat appear unoccupied were hypothesized to in-crease the perceived safety of access for burglars,which should lead to increased vulnerability toburglary.

The market value of the residence is another pos-sible vulnerability indicator. Macdonald and Gi¡ord(1989) obtained professional market appraisals of re-sidences which were assessed for vulnerability inde-pendently by burglars. The burglars assessed morevaluable houses as more vulnerable to burglary.1

Macdonald and Gi¡ord (1989) also proposed, onconceptual grounds, that surveillability should bedivided into two types: surveillability from theexternal perspective, such as that from the roadand neighbors' homes, and surveillability from aninternal perspective, such as the ease with which aresident can scan for approaching potential intru-ders from inside the house. Speci¢cally, roadsurveillability refers to the ability of passers-byand neighbors to view a house and property. Thenumber of windows visible from the road andwhether a neighbor's house is visible are examplesof cues within this category (Macdonald & Gi¡ord,1989). In contrast, occupants' surveillability has todo with residents' abilities to view their propertyand outside activity. Cues belonging to this categoryinclude the number of blocked/unblocked windows(e.g. by drapes or blinds), which can restrict orpermit residents to view their yard and the ap-proach to their house, and whether noise would becreated by someone nearing the house (Macdonald& Gi¡ord, 1989).

Of course, these two surveillability categoriesoverlap to some extent (r=0�61). Despite this, theusefulness of this distinction was demonstrated intwo previous studies (Macdonald & Gi¡ord, 1989;Shaw & Gi¡ord, 1994) in which both road surveill-ability and occupants' surveillability di¡erentiallypredicted vulnerability. Also, the two forms of sur-veillability are not strictly mirror images. For exam-ple, during the day it is easier to look out of awindow than it is to see inside a house from theoutside.

Newman's theory and these suggested modi¢ca-tions lead to the general defensible space hypothesis

FIGURE 1. A modi¢ed Brunswik lens model for comparing the de-fensible space cue utilization policies of two groups, in this casepolice and residents.

Defensible Space Theory 119

that houses with more actual barriers, moresymbolic barriers, more surveillability from thestreet and from inside the house, more traces of oc-cupancy, and lower market value will be assessed asless vulnerable to burglary.

The police perspective

Resident, burglar, and police perspectives are allimportant facets of Crime Prevention ThroughEnvironmental Design (CPTED) programs. Newman(1975; Newman & Franck, 1982) and others (e.g.National Crime Prevention Council [NCPC], 1989)emphasize that residents are `key agents inensuring their own security' (Newman, 1975, quotedin Taylor et al., 1980, p. 61). Others point outthe importance of the criminal's point of view(Macdonald & Gi¡ord, 1989; Cromwell et al., 1991;Brown & Bentley, 1993). In particular, someresearchers have begun to examine the di¡erentiale¡ect of defensible space features on criminalsand residents (Perkins et al., 1990; Shaw & Gi¡ord,1994).

Police should not be omitted from these compari-sons; they perform an integral role in teachingcrime prevention techniques, such as target harden-ing (NCPC, 1989), and they form a communicativelink between residents and criminals. Nevertheless,police assessments of the defensible space qualitiesof environments are not well understood. The ten-dency of police o¤cers to form images of larger en-vironments such as neighborhoods as `trouble areas'has been examined, and it appears police do usespeci¢c characteristics of these settings to struc-ture their assessments (McGahan, 1984, p. 125).

Logie et al. (1992) took perhaps the ¢rst step to-wards integrating and comparing information fromthree of the groups of interest in this studyÐpoliceo¤cers, residents, and (juvenile) burglars. Thatstudy's focus was on recognition memory, but its¢ndings have implications for CPTED. Their sub-jects ¢rst identi¢ed houses (from photographs) thatwould be appealing to burglars. Later, the subjectswere given a surprise recognition memory test inwhich some of the physical features of the houseshad been changed. The young burglars performedslightly better than police (and better than crim-inals who had committed other o¡enses), and boththe police and the young burglars performed betterthan the residents.

Logie et al. 's (1992) results also demonstrated thatburglars and police, but not residents, interpret thepresence of a car (a sign of occupancy) as a deter-rent to burglary. Both burglars and police indicated

that houses with hedges and fences (concealmentopportunities) are very attractive targets, whereasresidents seemed unaware of the potential of thisfactor. All three groups in the study agreed that aburglary attempt was less likely if there was analarm, but not if a lock alone was present.

The present study investigates the role of 55 spe-ci¢c physical cues and six cue categories representa-tive of related groups of cues as potential in£uenceson police assessments of residential vulnerability toburglary. It then compares police assessment poli-cies to those of burglars and residents, identi¢edin earlier studies (Macdonald & Gi¡ord, 1989; Shaw& Gi¡ord, 1994).

Brunswik's lens model as a framework

Brunswik's (1956) lens model provides a useful fra-mework for understanding the perceptual relationsbetween observers and environments. His modelwas adapted by Shaw and Gi¡ord (1994); theiradapted model will be used in this study (seeFigure 1). The adapted model includes cue utiliza-tion (the relative importance or weight that percei-vers place on each cue category) on both sides of themodel, so that the assessment policies of two obser-ver groups can be represented and compared in one¢gure.2

Cue utilization is represented by the simple (orpartialed) correlation coe¤cients between meanvulnerability assessments and the six cue cate-gories. Agreement between the observer groups isrepresented by the correlation between the groups'assessments of vulnerability across all targets(houses).

120 K. A. Ham-Rowbottom et al.

The di¡erential assessments of burglarsand residents

Macdonald and Gi¡ord (1989) assessed adult andyouth burglars' perceptions of a house's vulnerabil-ity to burglary. A partial correlation analysis oftheir data by Shaw and Gi¡ord (1994) to isolate thee¡ects of each cue category concluded that bothadult and youth burglars assess houses with highmarket value and low road surveillability as morevulnerable to burglary, which is consistent with de-fensible space theory and its extensions. However,contrary to the theory, actual barriers, symbolicbarriers, traces of occupancy, and occupant surveill-ability had no signi¢cant e¡ect on burglars' assess-ments. Occupants' surveillability (for adult burglarsonly) and symbolic barriers were signi¢cant as sim-ple correlations, but dropped out in the partial cor-relation analysis (i.e. when each cue category wascorrelated with vulnerability after all other cue ca-tegories were controlled).

Subsequently, Shaw and Gi¡ord (1994) used thesame experimental paradigm to examine residents'assessment policies. After partial correlation analy-sis (as in the earlier study), residents were found toassess greater vulnerability in houses with fewer ac-tual barriers, fewer traces of occupancy, and lessroad surveillability. Residents did not use symbolicbarriers, occupant surveillability, or house value assigni¢cant indicators of vulnerability. However, thesimple correlations revealed a di¡erent pattern foractual barriers, occupant surveillability, and tracesof occupancy; higher vulnerability was associatedwith actual barriers and reduced occupant surveill-ability, but occupancy traces were not signi¢cantlyrelated to burglary risk.

These studies, therefore, ¢nd di¡erent patternsof simple and partial correlations between cuecategories and perceived burglary risk; this dualpattern will be investigated further in this study ofpolice assessments of vulnerability. We believe thepattern of partial correlations is preferablewhen defensible space theory is tested and thatthe pattern indicated by the simple correlationsis preferable when everyday residences areconsidered, that is, when the various dimen-sions of defensible space are naturally intercorre-lated.

Based on partial correlations, residents and bur-glars used (only) one of the six cue categories (roadsurveillability) in the same signi¢cant way (Shaw &Gi¡ord, 1994). Neither the symbolic barrier nor oc-cupant surveillability cue categories were used bythese groups. The three remaining cue categories

were signi¢cantly used either by residents or bur-glars, but not by both.

Based on the simple correlations, the two surveill-ability cue categories were used in the same signi¢-cant way, but traces of occupancy were not utilizedby either group. The remaining three categorieswere signi¢cantly used by only one of the groups.Thus, the pattern of results di¡ers depending onwhether the goal is to test theory or to explicatethe dimensions of defensible space in a set of every-day buildings.

The two groups' across-houses assessments ofvulnerability were signi¢cantly correlated (r=0�68),and almost half their variance is shared. This agree-ment probably is based on their similar reading ofthe one cue category that signi¢cantly in£uencedboth groups, road surveillability, although some oftheir agreement could be based on similar readingsof cues that were present but not measured in thosestudies. Of course, over half the variance ofthe assessments by residents and burglars is notshared. This represents their di¡erential use of cuesto assess vulnerability (and some measurementerror).

Hypotheses

The present study investigates how police o¤cersutilize defensible space cues in their vulnerabilityassessments. No study has yet examined how policeo¤cers, as a group, assess vulnerability from defen-sible space cues.We know from the research just de-scribed that di¡erent observer groups (such asburglars and residents) use di¡erent sets of cues toassess vulnerability even when they are shown thesame houses. Thus, police may use the same or othercue categories to assess vulnerability. However, be-cause we have no basis for speculating which cuesor cue categories that might use, the most appropri-ate formal hypothesis is that police will make theirassessments in accordance with defensible spacetheory and its extensions.We do so because the the-ory is silent on the possibility of reliable di¡erencesin the ways that defensible space cues are inter-preted by di¡erent groups, and thus tacitly assertsthat all groups who assess the same physical settingwill do so in the same way.

Thus, our formal hypotheses are that, consistentwith defensible space theory and its extensions, thepresence of territorial markers or cues (i.e. actualand symbolic barriers) and opportunities for sur-veillance of the house and property (i.e. visibilityby the occupants and from the road) will be asso-ciated with lower assessed vulnerability to burglary.

Defensible Space Theory 121

We also expect that traces of occupancy and lowerhouse value will also lead to assessments of lowervulnerability to burglary by police o¤cers.

Method

Subjects

The subjects were 41 police o¤cers (33 male and 8female), between the ages of 24 and 52, from amunicipal police department of a medium-sizedCanadian city. The o¤cers had, on average, 11�6years of experience. After we received permissionfrom the Deputy Chief to recruit o¤cers for thestudy, each police o¤cer was contacted by phone.The o¤cers represented various ranks (includingtwo reserve o¤cers) from the patrol, street crime,and child and youth units. The response rate wasvirtually 100 per cent (one o¤cer had to go on pa-trol before the interview could begin).

Materials and measures

The same 50 photographs of single-family dwellingsfrom the Macdonald and Gi¡ord (1989) and Shawand Gi¡ord (1994) studies were used. As noted ear-lier, we used the same photographs to maximize thesimilarity of the stimuli for all groups, so that anydi¡erences could clearly be attributed to observerrather than environmental di¡erences. The 55 indi-vidual cues and six cue categories reliably rated inthe Macdonald and Gi¡ord (1989) study were alsoused in this study.3

Procedure

The police o¤cers were interviewed singly either ina private o¤ce or a conference room at police headquarters. They were seated across a table from theinterviewer. Cards with the numbers 1 to 7 printedon them were spread out in front of the o¤cer. Twocards, labeled `highly likely' and `not likely' wereplaced under the cards marked number 1 and num-ber 7, respectively.

The interviewer then verbally gave the followinginstructions:

I am interested in crime prevention through envir-onmental design. Speci¢cally, I would like to knowwhat police o¤cers think makes a house vulnerableto burglary. I would like you to look through thesephotographs and rate the houses in them on ascale of 1±7, as to how likely you believe they are tobe burglarized. Please the houses you believe most

likely to be burglarized under number 1, which ismarked `highly likely'; and place the houses you be-lieve least likely to be burglarized under number 7,which is marked `not likely'. Rate the houses basedon how they appear at the time the picture was ta-ken, and place at least three photographs undereach number.

The last instruction was intended to encouragethe o¤cers to use the full range of the vulnerabilityscale. After rating the photographs, each o¤cer wasasked to explain his or her reasoning for placingthe photographs in the two most extreme categories.The ratings for each photograph and the explana-tions were recorded by the experimenter. Each o¤-cer's sex, age, years of experience, main assignmentsand geographic locations of service were recorded,as was the approximate number of residential bur-glary (break-and-enter) investigations in which heor she had been involved. The presentation order ofthe photographs was counter-balanced across thepolice o¤cers.

Results

Reliability of assessments

In the original study (Macdonald & Gi¡ord, 1989),one researcher and three graduate students ratedsix of the 50 photographs on the 55 separate physi-cal cues. Because the reliability of these ratings wasextremely high (from 0�98 to 1�00), one researcherrated the other 44 photographs. The inter-rater re-liability (intraclass correlation) for the cue cate-gories was 0�95 for actual barriers, 0�72 for symbolicbarriers, 0�93 for traces of occupancy, 0�96 for roadsurveillability, and 0�89 for occupant surveillability.4

House value was reliably rated (0�91) by three profes-sional property appraisers.

In this study, all 50 photographs were rated onvulnerability to burglary by the 41 police o¤cers,whose assessments were highly consistent: the in-ter-rater reliability of their assessments was 0�97.As a group, the o¤cers strongly agreed amongthemselves which houses were more (and whichwere less) vulnerable to burglary.

Police o¤cers' perceptions of vulnerability

Individual cues. When the 55 individual cues werecorrelated with the police o¤cers' assessments, tenzero-order correlations were found to be signi¢cant(at least p50�01).5,6 Police believe that vulnerabilityto burglary is less if:

122 K. A. Ham-Rowbottom et al.

(1) the back yard is separated from the front yardby an actual barrier (r=70�50, p50�001);

(2) the front door is solid, with no glass (r=70�51,p50�001);

(3) a glass panel is not located next to the frontdoor (r=70�52, p50�001);7

(4) a carport or garage is present (r=70�52,p50�001);8

(5) at least three-quarters of the yard is visiblefrom the road (r=70�71, p50�001);

(6) at least three-quarters of the house is visiblefrom the road (r=70�75, p50�001);

(7) more windows are clearly visible from the road(r=70�45, p50�001);

(8) the front door is visible from the road (r=70�41,p= 0�003);

(9) a neighbor's house is visible (in the photo)(r=70�39, p= 0�006);

(10) the distance from the road is less than 20 feet(r=70�40, p= 0�004);

The ¢rst four of these cues are from the actualbarriers category; the last six are from the roadsurveillability category. Overall, these ¢ndings indi-cate that more actual barriers (e.g. solid front door)and greater road surveillabilitiy (e.g. windows visi-ble) lead police to believe that a house is less vul-nerable to burglary. However, individual cues o¡er,in general, a somewhat unstable picture; for thisreason we place more emphasis on groups or cate-gories of similar cues, in order to achieve more sta-bility and clarity in the constructs that predictvulnerability assessments. We turn next to thosecategories.

Cue categories. Multiple regression analysis showedthat the cue categories as a group strongly predictthe police o¤cers' vulnerability assessments [F(6,43) = 57�26, p50�001]. Together, they explained a verylarge proportion (adjusted R2 = 0�87, or 87%) of thevariance in the police o¤cers' vulnerability assess-ments.

How does each cue category contribute to this?Based on simple Pearson correlations, the presenceof actual barriers indicated to police o¤cers thatthe house was more likely to be burglarized(r=0�33, p50�02), which is the reverse of conven-tional defensible space wisdom.When the houses ex-hibited more opportunities of surveillance byoccupants (r=70�60, p50�001) or from the road(r=70�87, p50�001), they were judged to be less vul-nerable to burglary. Symbolic barriers, traces of oc-cupancy, and the market value of the house were

not signi¢cantly correlated with police o¤cers' vul-nerability assessments.

However, the simple Pearson correlations provideonly one view of the relation between the cue cate-gories and burglary risk. Although Macdonald andGi¡ord (1989) attempted to select a sample of resi-dences in such a way that the cue categories wouldbe uncorrelated, they did not completely succeed indoing so. Most cue categories were independent, butShaw and Gi¡ord (1994) noticed that some pairs ofcue categories were signi¢cantly correlated.9 Thus,the simple correlations represent the results as theymight occur in an everyday sample of actual resi-dences, in which the cue categories are somewhatinterrelated.

As noted earlier, a more theory-oriented analysiswould try to isolate the role of each cue category.Therefore, as in Shaw and Gi¡ord's (1994) study, anattempt was made to estimate the independent rela-tion of each cue category to the police o¤cers' vul-nerability assessments by computing partialcorrelations. In each case, the e¡ect of other cue ca-tegories was partialed from the simple correlationbetween a given cue category and the vulnerabilityassessment in order to gain an alternative perspec-tive of the relationship (see Table 1).

The partial correlations of three cue categories(symbolic barriers, road surveillability, and housevalue) with assessed vulnerability did not greatlydi¡er from their simple Pearson corrlations. How-ever, the partial correlations for the other threeshowed important shifts, as described in the follow-ing paragraphs.10

As shown in the last column of Table 1 and inFigure 2, when all other cue categories are con-trolled for, vulnerability is correlated 70�43(p50�05) with the presence of actual barriers, instrong contrast with the simple correlation of +0�33(p50�05). Thus, although the simple correlationsuggests that police o¤cers signi¢cantly associategreater vulnerability with more actual barriers,the partial correlation analysis indicates that whenthe e¡ect of other cue categories is controlled, po-lice o¤cers signi¢cantly associate greater vulner-ability with fewer actual barriers. The picturepainted by the partial correlation analysis is consis-tent with the results for the individual cues in theactual barrier category and with defensible spacetheory.

Second, the simple correlation between assessedvulnerability and traces of occupancy was 70�20(N.S), but the partial correlation was 70�68(p50�001). Thus, if other cue categories are con-trolled, it appears that police do assess houses with

TABLE1

Correlation

sbetw

eenthecuecategories

andpolice

o¤cers'v

ulne

rabilityassessments

Variab

lespa

rtialed

Allva

riab

les

Non

e1

23

45

6

1Actua

lba

rriers

733*

Ð732*

735

*729

*7

27733*

743

*2

Symbo

licba

rriers

709

700

Ð707

708

707

707

709

3Tracesof

ocup

ancy

720

722

719

Ð762***

722

719

768

***

4Roa

dsu

rveillab

ility

787***

787***

787***

792***

Ð7

80***

787***

789***

5Occup

ant's

surveillab

ility

760

***

758

***

760

***

761***

717

Ð761***

702

6Hou

seva

lue

705

705

700

703

710

711

Ð728

Note.

*p5

0�05**p5

0�01***p50�0

01.Decim

alpo

ints

omitted.

Defensible Space Theory 123

124 K. A. Ham-Rowbottom et al.

more traces of occupancy as less vulnerable to bur-glary. This also supports defensible space theory.

Third, the Pearson correlation between vulner-ability and occupant surveillability was 70�60(p50�001), but the partialed correlation was 0�02(N.S.). Thus, if the in£uence of all other cue cate-gories is controlled, police o¤cers' assessments arenot in£uenced by occupant surveillabilitiy. So, thisextension of defensible space theory is not sup-ported in the case of police, but it does in£uencevulnerability when considered in conjunction withother cue categories.

Regardless of whether simple or partial correla-tions are utilized, road surveillability appears to bethe most important in£uence on police o¤cers' vul-nerability assessments. Even when all other cue ca-tegories are controlled for, it accounts for 79 percent of the variation in their assessments. Table 1shows that road surveillability also had the mostdrastic e¡ects on the partial correlations for othercue categories. It probably accounts for the reversalof the actual barrier correlation, because actualbarriers and road surveillability are strongly corre-lated.

In summary, then, based on the partial correla-tions, police believe that houses with high road sur-

FIGURE 2. The cue utilization policies of police and residents. The dato a correlation signi¢es the simple (Pearson) correlation; a `p' appecategories controlled).

veillability, more actual barriers, and more traces ofoccupancy are less vulnerable to burglary. Thehouse's market value, occupant surveillabilitiy, andsymbolic barriers do not appear to in£uence policeassessments after the other cue categories are par-tialed.

The next section reveals that, consistent with the¢ndings just discussed, police frequently indicate(when explaining vulnerability in their own words)that opportunities for surveillance and concealmentand indicators of occupancy are important indica-tors of burglary risk.

Police o¤cers' verbal explanations

The frequency with which police o¤cers o¡ered var-ious explanations for their assessments is shown inTable 2. The most frequently o¡ered reasons for rat-ing a house as `highly likely' to be burglarized werethe presence of concealment opportunities (80%),that it was set back from the road (63%), that itwas a secluded property (59%), and that it appearedthat no-one was home (29%).

Their most frequently o¡ered reasons for ratinghouses as `not likely' to be burglarized were that itwas too open (68%), the neighbors were too close

ta for residents are from Shaw and Gi¡ord (1994). An s̀' appendednded to a correlation signi¢es the partial correlation (all other cue

TABLE 2Police o¤cers' explanations for vulnerability assessments

Reasons for sorting housesinto category

Per cent who mentioned

`Highly vulnerable'

Secluded property 59Concealment opportunity 80Appearance of value 10Sliding glass doors 2Set back from the road 63Number of exits 7Lots of windows 5Appears no-one home 29

`Not vulnerable'

`Beware of dog' sign 2Too open 68Neighbors too close 51Lack of value 24Appears occupied 39Too easy to see in windows 22Too much value 10Fences visible 32

Defensible Space Theory 125

(51%), that it appeared to be occupied (39%), fenceswere present (32%), the house had low market value(24%), and that it was too easy to see in the win-dows (22%).11

FIGURE 3. The cue utilization policies of police and burglars. The daare based on the averages of the adult and youth burglars' assessmcorrelated (r=0�84). The s̀' and `p' designations are explained in Fig

Police agreement with residents and burglars

Next, we examined the extent to which police as-sessments of vulnerability agree with those of bur-glars and residents. In the adapted lens model weuse, agreement is the Pearson correlation betweenthe (mean) assessments of each group across allthe target houses. Agreement between police and re-sidents was r=0�93 (see Figure 2). Because theyouth and adult burglars in the Macdonald and Gif-ford (1989) study strongly agreed with each other(r=0�84), their assessments were combined. Police-burglar agreements was r=0�61 (r50�001) (seeFigure 3), and burglar±resident agreement wasr=0�71 (p50�001).

Discussion

This study investigated whether police o¤cers' as-sessments of vulnerability to burglary are relatedto the presence of defensible space features in themanner predicted by Newman (1972) and otherswho later modi¢ed his theory. In addition, by com-paring the assessment policies of police o¤cers withthose of burglars and residents who viewed thesame stimuli in previous studies, we could examine

ta for burglars are from Macdonald and Gi¡ord (1989); the ¢guresents in that study because the two groups' assessments were highlyure 2.

126 K. A. Ham-Rowbottom et al.

the generalizability of the theory across di¡erentobserver groups.

Based on the partial correlations, which representthe independent in£uence of each cue category andare best used to test theory, three of the six defensi-ble space cue categories signi¢cantly a¡ect policeo¤cers' vulnerability assessments, and three donot. Only (more) actual barriers, (more) traces of oc-cupancy, and (greater) road surveillability seem tohave independent validity as predictors of (lower)vulnerability to burglary. The results for actual bar-riers vary with the analytic approach: independentof other cue categories, they lead to assessments oflower vulnerability by police, but in the context ofthe actual sample of residences, they appear tocause police to assess houses as more vulnerable.Results also vary for traces of occupancy. Indepen-dent of the other cue categories, traces of occu-pancy lead police to assess houses as lessvulnerable to burglary, but do not impact police as-sessments when these cues are viewed in context.Occupant surveillability is an important signalerof reduced vulnerability in the simple correlationapproach, but unimportant as an independent cuecategory. This probably occurred because occupantsurveillability is, after all, fairly redundant withroad surveillabilitiy. Symbolic barriers and housevalue have no in£uence on police assessments ofvulnerability in either the sample or partial correla-tion approaches.

The three signi¢cant cue categories together ex-plain a very signi¢cant proportion of the variancein police assessments. This is the basis of our claimthat pruning traditional defensible space theory tothree cue categories does not weaken its ability toexplain how assessments of defensibility are cre-ated; the three cue categories explain 87 per centof police assessments.

Furthermore, the same cue categories (regardlessor whether the pattern of simple or partial correla-tions is considered) explain the assessment policiesof residents (cf. Shaw & Gi¡ord, 1994) in almost ex-actly the same way. Police and resident assessmentsare correlated 0�93. The same pruned form of defen-sible space theory appears to work very well forboth police and residents.

However, the cue categories that so well explainthe police and resident assessment policies are notthe same as those that explain the burglar's perspec-tive (cf. Macdonald & Gi¡ord, 1989). Burglars do notpay signi¢cant attention to actual barriers or tracesof occupancy, as do residents and police.12 They dopay signi¢cant attention to the market value of thehouse, which police and residents do not.13 The sim-

ple correlations show that burglars attend to sym-bolic barriers (and occupant surveillability foradult burglars), but the partial correlation analysisindicates that the draw is house value; it just sohappens that more expensive houses have more sym-bolic barriers (Shaw & Gi¡ord, 1994, Tables 3 and 4).Only one cue category predicts assessed vulnerabil-ity in all three groups (regardless of the type of cor-relational analysis): road surveillability. This onecue category is the only portion of defensible spacetheory that holds across all three groups.

In terms of agreement, police vulnerability assess-ments shared 31 per cent of their variance withthose of youth burglars, and 36 per cent of their var-iance with those of adult burglars. Thus, police±resi-dent assessments are modestly related to those ofburglars. Perhaps this is just enough agreement tocreate the false impression that all observers assesscues the same way; if burglars' assessments weremore distinct it would be more obvious to the ca-sual observer that they assess residential vulner-ability di¡erently.

In sum, based on the partial correlations, defensi-ble space theory does not need all its hypothesizedcue categories and is not universal. Police and resi-dents perceive burglary risk quite similarly andboth groups see it di¡erently than burglars do. Itwould seem that more parsimonious versions of de-fensible space theory can be constructed, but thesewill be di¡erent for di¡erent observer groups.

Defensible space theory for police

The most interesting ¢nding is that, based on sim-ple correlations, police o¤cers judged houses withmore actual barriers as more vulnerable to burglary,but after partialing, the results conformed with thetraditional defensible space notion that actualbarriers reduce vulnerability. Why did this occur?Road surveillability clearly had the greatest impacton police perceptions of vulnerability, but it and ac-tual barriers are often naturally confounded in realhouses. Although actual barriers can impede acces-sibility, they may also obstruct visibility of thehouse and property from the viewpoint of occupantsand passers-by. Once unconfounded by partial cor-relation analysis, the `more expected' outcome (ac-cording to theory) was found.

Police assessments versus those of burglarsand residents

Perhaps it is surprising that the police and burglarperspectives are not more similar. Both groups are

Defensible Space Theory 127

trained observers who work on the same problem(albeit with opposite goals) and who communicateand presumably learn from one another. In contrast,residents are unaccustomed to critically observingtheir environment, tend to dislike thinking aboutburglary, and usually have no direct contact withburglars.

The di¡erences must be related to their di¡erentexperiences. Burglars, through their direct experi-ence with break-and-enters, have learned what tolook for when they c̀ase' a target house and whatto expect when they break into that house. In con-trast, police o¤cers are trained to observe crimescenes, interview criminals and, obviously, do notexperience burglaries as the perpetrators. Thus, po-lice o¤cers do not learn from breaking into a house,but from observing the after-e¡ects. Residents' ex-periences with burglary also are second hand, inthat they receive their education from the police(unless they have been victimized).

Agreement between police and resident vulner-ability assessments was very high: they shared 86per cent of their variance and did not signi¢cantlydi¡er in the importance they placed on any of thecue categories. Perhaps this agreement occurs be-cause police have successfully communicated theirperspectives on vulnerability to residents. Perhapsit is because police and residents share other ex-periences and values, which are not the same asthose of burglars. For example, burglars, but notthe others, must consider the probability of beingcaught, desperation, and the degree of challengethat a house represents, among other factors. Bur-glars may often be lower in socio-economic stand-ing. These and other factors may explain thediscrepancy between the police±resident and bur-glar assessments.

Practical implications

These ¢ndings are useful as the basis for recom-mending changes to defensible space theory, butthey also speak to practical issues in crime preven-tion. For example, when residents elect to screentheir houses with tall opaque fences or shrubs inthe pursuit of privacy, they make their houses ap-pear more inviting to burglars by decreasing roadsurveillability. Because burglars are the key to bur-glary, understanding their assessment criteria iscrucial to counteracting burglary. This study showsthat burglars have a somewhat di¡erent perspectivefrom the police and residents. This has at least twoimportant implications. First, police and residentsmust realize that there is a di¡erence, and second

they must base their prevention programs on theburglar's view of vulnerability rather than theirown.

Acknowledgements

We would like to thank Julie Macdonald for provid-ing the photographs used in this study and the po-lice o¤cers of Saanich, British Columbia, for theirparticipation and comments.

Notes

Reprint requests and correspondence should be addressedto Kathleen A. Ham-Rowbottom, Room 209, Social Ecol-ogy I, School of Social Ecology, University of California,Irvine, California 92697, U.S.A. E-mail: khamrow@soceco.uci.edu(1) Burglars typically report that they prefer more expen-sive targets, which they claim are more vulnerable toburglary (Maguire & Bennett, 1982; Bennett & Wright,1984; Cromwell et al., 1991). However, when they actuallychoose a site to burglarize, it is often located in a neigh-borhood close to their own homes. The target, however, islikely to be one of the more valuable sites in that area(Reppetto, 1974). In order to gain a clearer picture of bur-glars' decision-making about their target, further researchis neededÐspeci¢cally, an explicit comparison between`better' and `worse' houses within poor neighborhoodsand `better' and `worse' houses within richer neighbor-hoods.(2) The original Brunswik model includes cue utilizationon the right side of the lens and ecological validity on theleft. Ecological validity would be appropriate in a studydesigned to examine direct links between environmentalcues and crime occurrence. A full, traditional lens modelstudy of defensible space theory with ecological validity(`reality') on the left side, and cue utilization (assessmentsof observers) on the right side, in the tradition of Craikand Appleyard (1980), would make a useful addition to aliterature.(3) Originally, Macdonald and Gi¡ord (1989) rated eachphotograph on 64 individual cues and ¢ve cue categories.Nine of the 64 individual cues were sums of individualcues. House value, the sixth cue category, was created posthoc.(4) The cue categories of actual and symbolic barriers,traces of occupancy, and road and occupant surveillabil-ity do not represent summed individual cues; rather, eachcategory was comprised of ratings that were based onconceptual descriptions of each category along with sev-eral examples. For example, `Traces of Occupancy: thesecues indicate the probability that the occupants are homeor away. Cars in the driveway, smoke from the chimney,and interior lights are traces of presence [occupancy]. Un-collected mail and exterior lights on at inappropriatetimes are traces of absence' (Macdonald & Gi¡ord, 1989,p. 196).

128 K. A. Ham-Rowbottom et al.

(5) Although the hypotheses in this study are directional,all signi¢cance tests were two-tailed. This is because re-sults to date are inconclusive and we wished to avoidType II errors. When more than one comparison wasmade, a conservative alpha level was employed to lessenthe chances of making a Type I error. An alpha level ofp50�05 was applied to the results for the multiple regres-sion analyses and simple correlations between each cuecategory and the vulnerability assessments (because theoverall multiple regression was signi¢cant). An alpha le-vel of p50�01 was applied to the analyses involving the in-dividual cues and the explanations.(6) Vulnerability was coded 1=high vulnerability and7= low vulnerability; the cue categories were coded1=highly representative of the category and 7=not repre-sentative of the category. The individual cues were codedas follows: 1= absent and 2=present. For clarity of presen-tation, we have reversed the sign of the correlations be-tween individual cues and vulnerability (which is akin toreverse coding the individual cues prior to performingcalculations). Hence, it appears that the correlations forthe individual cues and vulnerability are in the oppositedirection to those for the categories, but this is actuallynot the case.(7) Upon closer inspection, it appears that this individualcue was incorrectly coded in the earlier studiesÐit shouldhave been reverse coded, but was not. As a result, correla-tions between this cue and vulnerability ratings were inthe incorrect direction in the previous studies. This cuewas recoded for the current study and as such, the resultsreported here are correct. In this case, it is the absence ofa glass panel next to the front door that is the actual bar-rier.(8) This individual cue was coded as an actual barrier;however, it could also be a cue for traces of occupancy.(9) Despite e¡orts to orthogonalize all predictors and be-cause the real world is di¤cult to orthogonalize, some cuecategories are correlated (e.g. actual barriers and roadsurveillability are correlated [70�52, p50�001]; symbolicbarriers and house value are correlated [0�58, p50�001]).(10) It is important to note that when the direction of theassociation between vulnerability and a cue category re-verses (e.g. from r=0�33 to r=70�43) or when there is alarge change in the size of the association, suppressorvariables may be present. Suppressors hamper interpreta-tion because they indicate that there is a pattern to theintercorrelations among the categories that constrainsthe error variance in other categories. As a result, caremust be taken when interpreting the result, care must betaken when interpreting the results when such a reversalor change is evident.(11) This is classi¢ed as a `not likely' explanationbecause it refers to the fact that if it is easy to see inthe windows, passers-by would by able to see an intruderin the house. It is also possible to interpret it as a `highlylikely' explanation, in that if it is easy to see in the win-dows, burglars would be able to view the residents' valu-ables.(12) Although burglars do not indicate that they pay at-tention to actual barriers or traces of occupancy duringa study, it is likely that they take note of concealment op-portunities and occupancy when they are assessing an ac-tual target. When in the ¢eld, burglars often reportlooking for actual barriers such as hedges in order to

locate places to hide (e.g. Cromwell et al., 1991). Also,many say that they discount actual barriers because theybelieve they can ¢nd a way into any house they choose(this is likely partly a boast and partly the truth [skill])(Macdonald & Gi¡ord, 1989). When assessing the occu-pancy of a real target, burglars are likely to use directmethods, such as calling the house or ringing the door-bell, over and above attending to traces (e.g. Crowmwellet al., 1991).(13) See Note 1.

References

Bennett, T. & Wright, R. (1984). Burglars on Burglary: pre-vention and the o¡ender. Aldershot, Hampshire:Gower.

Brower, S., Dockett, K. & Taylor, R. B. (1983). Residents'perceptions of territorial features and perceived localthreat. Environment and Behavior 15, 419±437.

Brown, B. B. (1985). Residential territories: cues toburglary vulnerability. Journal of Architectural andPlanning research 2, 231±243.

Brown, B. B. & Altman, I. (1981). Territoriality and resi-dential crime: a conceptual framework. In P. L. Bran-tingham & P. J. Brantingham (Eds) EnvironmentalCriminology. Beverly Hills, CA: Sage, pp. 55±76.

Brown, B. B. & Bentley, D. L. (1993). Residential burglarsjudge risk: the role of territoriality. Journal of Envir-onmental Psychology 13, 51±61.

Brunswik, E. (1956). Perception and the RepresentativeDesign of Psychological Experiments. Berkeley, CA:University of California Press.

Craik, K. H. & Appleyard, D. (1980). Streets of SanFrancisco: Brunswik's lens model applied to urban in-ference and assessment. Journal of Social Issues 36(3),72±83.

Cromwell, P. F., Olson, J. N. & Avary, D.W. (1991). Breakingand Entering: an ethnographic analysis of burglary.Newbury Park, CA: Sage.

Jacobs, J. (1961). The Death and Life of Great AmericanCities. New York: Random House.

Logie, R.,Wright, R. & Decker, S. (1992). Recognition mem-ory performance and residential burglary. AppliedCognitive Psychology 6, 109±123.

Macdonald, J. E. & Gi¡ord, R. (1989). Territorial cues anddefensible space theory: the burglar's point of view.Journal of Environmental Psychology 9, 193±205.

Maguire, M. & Bennett, T. (1982). Burglary in a Dwelling:the o¡ence, the o¡ender, and the victim. London: Heine-mann.

McGahan, P. (1984). Police imagery of the urban environ-ment. Canadian Police College Journal 8, 107±153.

National Crime Prevention Council. (1989). The success ofcommunity crime prevention. Canadian Journal ofCriminology 31, 487±506.

Newman, O. (1972). Defensible Space. New York:Macmillan.

Newman, O. (1975). Reactions to the `̀defensible space''study and some further ¢ndings. International Journalof Mental Health 4, 48±70.

Newman, O. & Franck, K. A. (1982). The e¡ects of buildingsize on personal crime and fear of crime. Populationand Environment 5, 203±220.

Defensible Space Theory 129

Pablant, P. & Baxter, J. C. (1975). Environmental correlatesof school vandalism. Journal of the American Instituteof Planners 41, 270±277.

Perkins, D. D., Florin, P., Rich, R. C., Wandersman, A. &Chavis, D. M. (1990). Participation and the socialand physical environment of residential blocks: crimeand community context. American Journal of Commu-nity Psychology 18, 83±115.

Reppetto, T. A. (1974). Residential Crime. Cambridge, MA:Ballinger.

Shaw, K. T. & Gi¡ord, R. (1994). Residents' and burglars'assessments of burglary risk from defensiblespace cues. Journal of Environmental Psychology 14,177±194.

Taylor, R. B., Gottfredson, S. D. & Brower, S. (1980).The defensibility of defensible space: a criticalreview and a synthetic framework for futureresearch. In T. Hirschi & M. Gottfredson (Eds)Understanding Crime. Beverly Hills, CA: Sage, pp.53±71.