Psychological Profiling of Sexual Murders: An Empirical Model

10.1177/030662402236739International Journal of Offender Therapy and Comparative CriminologyPsychological Profiling of Sexual Murders

Psychological Profiling of Sexual Murders:An Empirical Model

Richard N. KocsisRay W. CookseyHarvey J. Irwin

Abstract: Psychological profiling represents the investigative technique of analyzing crimebehaviors for the identification of probable offender characteristics. Profiling has progres-sively been incorporated into police procedures despite a surprising lack of empirical researchto support its validity. Indeed, in the study of sexual murder for the purpose of profiling, veryfew quantitative, academically reviewed studies exist. This article reports on the results of a 4-year study into Australian sexual murders for the development of psychological profiling. Thestudy involved 85 cases of sexual murder sampled from all Australian police jurisdictions. Thestatistical procedure of multidimensional scaling was employed. This analysis produced afive-cluster model of sexual murder behavior. First, a central cluster of behaviors was identi-fied that represents common behaviors to all patterns of sexual murder. Next, four distinct out-lying patterns—predator, fury, perversion, and rape—were identified that each demonstrateddistinct offence styles. Further analysis of these patterns also identified distinct offender char-acteristics that allow for the use of empirically robust offender profiles in future sexual murderinvestigations.

Criminal psychological profiling represents the investigative technique of analyz-ing crime behaviors for the identification of probable offender characteristics.However, profiling has progressively been incorporated into police proceduresdespite a surprising lack of empirical research to support its validity (Kocsis, inpress-a, in press-b; Kocsis, Hayes, & Irwin, 2002; Kocsis, Irwin, Hayes, & Nunn,2000; Pinizzotto & Finkel, 1990). Indeed, in the study of sexual murder for thepurpose of profiling, very few quantitative, academically reviewed studies havebeen undertaken. For example, a recent issue of the Journal of ContemporaryCriminal Justice especially dedicated to the topic of Psychological Profiling failsto report an original quantitative study on sexual murder profiling. Instead, aca-demic activity in this area has been predominated by articles focused on a theoreti-cal discussion of profiling issues (Jackson & Berkerian, 1997; Wilson, Lincoln, &Kocsis, 1997; Wilson & Soothill, 1996).

Possibly the most cited original empirical study in the area of sexual murderprofiling, and profiling in general, was undertaken in the early 1980s by the Amer-

NOTE: Correspondence concerning this article should be addressed to Richard N. Kocsis, 16 LyndenAve., Carlingford, Sydney NSW 2118, Australia; e-mail: [email protected]

International Journal of Offender Therapy and Comparative Criminology, 46(5), 2002 532-554DOI: 10.1177/030662402236739

2002 Sage Publications

532

at PENNSYLVANIA STATE UNIV on September 18, 2016ijo.sagepub.comDownloaded from

http://ijo.sagepub.com/

ican Federal Bureau of Investigation Behavioral Science Unit (Ressler, Burgess,& Douglas, 1988). This pioneering study reported on the analysis of 36 incarcer-ated sexual murderers within North America. The outcome of this research wasthe much renowned “organized/disorganized” behavior dichotomy. The underly-ing premise of this dichotomy was the interpretation of crimes by their level ofbehavioral sophistication and corresponding offender characteristics. Thus, the“organized” category represented a methodical, premeditated crime with corre-sponding offender characteristics such as maturity, resourcefulness, and typicallysexual perversion. Inversely, the “disorganized” category represented a haphaz-ard, almost random crime with the corresponding inverse offender characteristicsof immaturity, opportunism, and a likelihood to suffer from some mental disorder(Ressler et al., 1988).

Unfortunately, although this study has attracted enormous citation and dis-course, simple empirical replications of this pioneering research have been greatlylacking. Indeed, the empirical replication of this dichotomy in Kocsis, Irwin, andHayes (1998) appears to be the only quantitative, academically reviewed study toaddress this paucity. The findings of this research revealed that the concept ofinterpreting crimes by their level of behavioral sophistication does hold somemerit. However, a more realistic and utilitarian interpretation of crime behaviorswould require development beyond a simple dichotomy. Future research shouldaim to construct a model that would allow for the scientific and objective interpre-tation of crime behaviors and associated offender characteristics.

Aside from these few acknowledged empirical studies, the literature on sexualmurder profiling is dominated by various proposed offender taxonomies. Theforemost of these is the Crime Classification Manual of the FBI (Douglas, Bur-gess, Burgess, & Ressler, 1992), which uses the organized/disorganized dichot-omy as a basis for classifying sexual murder. However, this lexicon appears toacknowledge the limitations of this dichotomy by proposing the new intermediarycategory of the “mixed” offender. Furthermore, the proposal of another new cate-gory, labeled the “sadistic” offender, seems to be already encapsulated within theparameters of the organized offender.

In the sexual murder typologies of Hickey (1997) and Holmes and Holmes(1998), common characteristics are identifiable in the offender categories thatcorrespond with the work of the FBI. The Holmes categories rely primarily on theinference of offender motivations and associated psychogenic factors. Similarly,the Hickey categories are based on a combination of inferred motivations withdemographic features such as the gender of the offender, their sexual orientation,or the use of accomplices.

The typologies of both Holmes and Holmes (1998) and Hickey (1997) com-monly identify an offender category delineated by a pattern of selection ofunknown victims with physical and sexual assault being perpetrated for a psycho-logical gain. This gain Holmes described as being similar to sexual sadism(Holmes & Holmes, 1998, p. 62), whereas Hickey favored goals of power andcontrol with sexual sadism as a subsidiary drive (Hickey, 1997, p. 154). This com-

Psychological Profiling of Sexual Murders 533



mon offender category Holmes and Holmes (1998) encapsulated under the broadtitle of hedonism with subcategories of “lust” and “thrill” killers. Similarly,Hickey (1997) referred to this pattern as “men who kill women” (based on the pre-dominant gender pattern of these offences). Hickey, however, also assigned thelabel of “lust killers” to this category. Thus, common themes of anger, intercourse,and sadism seem to be inherently discernible from amongst these various sexualmurder typologies.

Both Hickey (1997) and Holmes and Holmes (1998) also elaborated on variousother sexual murderer categories that are identified primarily by the murder ofmultiple victims. Here a number of highly diverse categories are nominated. Themotivations for these categories include such factors as political beliefs, histrionicattention, or monetary gain. Such a vast diversity of categories poses problems forthe conceptualization and coherent understanding of sexual murder. Conse-quently, Kocsis and Irwin (1998) discussed the definition of serial crime and howthis title has become overly inclusive. They propose a more focused definition ofsexual murder where offenders are identified by an underlying profile of psycho-logical components.

In addition to the literature on sexual murder profiling, the material on sexualassault profiling is also relevant to the further development of a scientific andobjective model of sexual murder profiling. The first body of relevant work is byGroth, Burgess, and Holmstrom (1977). The central thesis of this study is that raperepresents the sexually violent expression of power and domination rather than anact of sexual desire. Two broad categories of rapist motivation are offered: “powerrape,” where the offender’s primary motivation is power and control over the vic-tim, and “anger rape,” where the offender’s primary motivation is an expression ofhatred and contempt for the victim.

Work has also been undertaken by the FBI Behavioral Science Unit into rapeprofiling (Hazelwood & Burgess, 1987). The starting point of this unit’s work wasthe adoption of the offender categories by Groth (Hazelwood & Burgess, 1995,pp. 160-161). However, for the purpose of developing information that would bedirectly relevant to the needs of law enforcement, a study of incarcerated NorthAmerican serial rapists was undertaken. This study resulted in the compilation ofan extensive list of serial rape demographic information.

Finally, Canter and Heritage (1989) offered five categories of sexual assaultbehavior: “intimacy,” where intimacy is attempted with the victim, “sexuality,”where intercourse is the crucial element in the assault, “violence,” characterizedby violent action, “impersonal,” where the victim is treated as an object, and“criminality,” where the assault is not overtly sexual in nature. This research ofCanter and Heritage is different from previous work on sexual assault in so far astheir categories are based exclusively on the empirical analysis of offenderactions. The basic hypothesis of profiling is that offenders differ in their actionsand that these differences in behavior relate to the offender’s characteristics. Con-sequently, the interpretation of crime actions requires the classification of offencebehaviors as distinct from any inferred motivation. On this point Canter and Heri-

534 International Journal of Offender Therapy and Comparative Criminology



tage argued that all previous profiling studies, such as those undertaken by Grothet al. (1977) that combine inferred motivations with actions, are empiricallyflawed. A further methodological issue to arise from the research of Canter andHeritage is in the distinct empirical discrimination between varying crime behav-iors. A flaw that has plagued previous studies is that they have made no allow-ances for the possibility of commonalties in behavior amongst the variousoffender typologies, whether in the crime of sexual assault or murder.

At the time of this study, no Australian research has been published that adoptsthis empirical approach of analyzing crime actions, as distinct from inferred moti-vations, in sexual murders for criminal profiling. Consequently, the primaryobjective of this study was to employ such an analysis. In addition, the selection ofcases for this study would employ the criteria for sexual murder as discussed byKocsis and Irwin (1998) and thus avoid any potentially confounding surplus ofcases.

It is envisaged that this analysis will produce an empirical model of sexualmurder behavior that will contain distinct behavior clusters. Furthermore, theseempirically distinct behavior clusters could be analyzed to identify associatedoffender characteristics. The direct statistical matching of behaviors to offendercharacteristics using this form of analysis has, surprisingly, never been previouslyundertaken. Instead, past literature has been characterized by qualitative associa-tions between offender behaviors and offender characteristics (Canter & Heri-tage, 1989). Finally, this study aims to identify potential behaviors that are nondis-criminatory, and thus commonly observable, in all behavior patterns of sexualmurder.

METHOD

DATABASE AND DATA SCREENING PROCESS

The study consisted of 85 murder cases whose characteristics fit the generalpattern of sexual murder as discussed by Kocsis and Irwin (1998). The cases datedas far back as 1960 and as recent as 1998 and were taken from all police jurisdic-tions across Australia. All offenders in the sample had been convicted of at leastone sexual murder. All cases had details recorded using the VICLAS recordingsystem as well as in case history information held by various Australian policejurisdictions. Although the full VICLAS form was used, many cases had missingdetails on numerous variables. Frequency distributions were computed for allvariables in the database. The goal was to achieve a stable and internally consis-tent refined database made up of variables possessing sufficient nonmissingentries as well as sufficient variability across categories within each variable.Variables presenting largely constant data values were removed prior to analysisas were variables that were missing across a large portion (i.e., greater than 50%)of the database.




VARIABLE CONDENSATION

To facilitate data analysis and interpretation, conceptually similar categoriesfor each variable were collapsed with a view to producing dichotomous (0, 1)measures having a reasonable number of category 1 responses. Most variableswere recoded on a presence/absence basis, whereas others were recoded into less/more-type categories. Some variables having multiple categories were dummycoded into several dichotomous variables (e.g., the variables for initial contactlocation, crime scene location, recovery site location, type of weapon used, etc.).Variables with very few or no category 1 responses were deleted with the excep-tion of the BURNING (V burned) variable (n = 2 category 1 responses).

As a result of this data screening and variable condensation process, some 260variables were reduced to the final set of 137 variables employed in this study.Variables were broadly grouped into conceptual sets for subsequent analyses: vic-tim characteristics (14 variables), offender characteristics (36 variables),offender-victim interaction characteristics (22 variables), and crime scene char-acteristics (65 variables). Due to publishing limitations, variable labels, codingdetails, and extended names used for multidimensional scaling diagrams couldnot be included in this article. Nonetheless, interested readers may obtain thismaterial by contacting the first author.

ANALYSIS PROCESS

The analysis consisted of several discrete steps. The first step involved anonmetric multidimensional scaling analysis of the 65 dichotomous crime scenecharacteristics to identify the appropriate number of dimensions, with the range oftwo- to five-dimensional solutions, to interpret (see Hair, Anderson, Tatham, &Black, 1998, chapter 10). For purposes of multidimensional scaling, 65 objectswould be considered as a large sample for scaling (see Hair et al., 1998, p. 533).This analysis was accomplished with the multidimensional scaling (MDS) pro-gram in SYSTAT 7.0 using the Guttman coefficient of alienation minimizationcriterion and the Jaccard measure of binary similarity.

The second step in the analysis involved a cluster analysis of the resulting MDScoordinates to facilitate a regional interpretation of the dimensional solution (aprocess recommended by Kruskal & Wish, 1978, and Coxon, 1982). The numberof entities clustered in this analysis was 85 pairs of coordinates that, given the waythe cases were selected, can be considered to be a representative sample of sexualassault crimes and would be considered adequate for the purposes of cluster anal-ysis (see Hair et al., 1998, p. 490). Ward’s minimum variance method of clusteringwas employed for this step using the squared Euclidean distance measure of dis-similarity. The dimensional coordinates of the MDS solution were standardized(converted to z scores) and then plotted on a scatterplot using cluster identifiers todifferentiate the plotting symbols.




The third step in the analysis involved the fitting of external property vectors,using variables from the victim, offender, and offender-victim interaction variablesets, to the MDS coordinates for each of the 65 crime scenes. It is important toacknowledge that the sample size for the canonical correlation analysis conductedhere would be considered too small for the purposes of cross validating of theresulting equation parameter estimates. However, as the purpose of this exercisewas primarily descriptive and relational, as opposed to establishing equations forpredictive purposes, this sample was considered sufficient for purposes of thisarticle. A new database was created that contained the standardized dimensionalcoordinates from the MDS analysis and conditional probabilities for eachVICLAS characteristic variable not used in the MDS analysis. Each conditionalprobability was found as the mean of the external VICLAS variable of interestwithin the category coded “1” for a specific crime scene characteristic. This meanwas, in effect, the conditional probability that the variable (e.g., VSEX) equaled 1(female) when a specific crime scene characteristic (e.g., UNPATTER) alsoequaled 1 (unpatterned wounds were present). These conditional probabilitiesbecame the external characteristic property vector variables to be fit to the MDScoordinates to aid in its interpretation. For statistical control purposes, two addi-tional new variables were created that indicated the number of nonmissing cases(out of 85 total cases) for each crime scene variable and the number of observa-tions making up the category coded as 1 for each crime scene variable.

Property fitting was accomplished using an extension of the multiple regres-sion procedure for fitting direction cosines described by Kruskal and Wish (1978,pp. 87-88). Each of the conditional probability variables within a specific concep-tual set (e.g., victim characteristic set, offender characteristic set, and offender-victim interaction characteristic set) were predicted from the standardized MDScoordinates using the set correlation analysis procedure in SYSTAT 7.0 (seeCohen & Cohen, 1983, for a discussion of set correlation as a multivariate rela-tional technique). Both the predictor set (MDS coordinates) and criterion set (con-ditional probability variables) were partialled for influence of number ofnonmissing observations and for number of observations making up the categorycoded as 1 to control for the possible influences of missing data and small cate-gory 1 membership. The appropriate multivariate canonical correlation, associ-ated with each variable set making up the cluster for each vector, was recorded andtested for overall significance.

Working from the univariate level, each significant (based on the omnibus Ftest for the regression analysis of that variable) external characteristic variablewithin a set was identified. Then, the standardized regression coefficients for eachof the MDS dimensions for each significant variable within the set of interest werecluster analyzed using Ward’s method with the Euclidean distance measure andthe most appropriate number of clusters of characteristics was then identified. Foreach such identified cluster, the standardized regression weights for each dimen-sion were averaged across the variables making up the cluster.1 On separate MDS




dimensional plots (one for each conceptual set of variables analyzed), the externalcharacteristic property vectors, for each cluster of variables, were fitted using theaveraged regression beta weights, which were transformed to yield a specific vec-tor angle that was then drawn onto the graph with an arrowhead. The angle com-puted ranged between 0 degrees and 360 degrees and thus directly indicated thedirection that the association arrowhead should point.2

RESULTS

MULTIDIMENSIONAL SCALING

After closely examining each of the two- through five-dimensional solutions,the two-dimensional solution was identified as most appropriate to interpret(coefficient of alienation = .264; R2 = .758). The higher order dimensional solu-tions, while producing marginally better data fit, were conceptually difficult tointerpret clearly. Interpretability was the ultimate criterion to be met by the solu-tion adopted. Figure 1 shows the plot of the standardized coordinates for the two-dimensional MDS solution for the 65 crime scene characteristics.3

The two-dimensional MDS coordinates were hierarchically clustered and fiveclusters of crime scene characteristics were identified. These five clusters dividedthe two-dimensional space of crime scene characteristics into five non-overlapping regions. The five clusters of coordinates are marked by distinct plot-ting symbols in Figure 1 and the cluster regions have also been sketched in. Thebest process for interpretation of Figure 1 was a regional analysis. Crime scenecharacteristics appearing in the same region of the plot were inspected for com-mon themes to reach an interpretation of what each region might be indicating.

The central cluster 1 (surrounded by the ellipse) represented crime scene char-acteristics that were not clearly discriminated by the two-dimensional MDS struc-ture; they were associated by virtue of having similar coordinate patterns centeredon or near zero for each dimension. Thus, these particular crime scene characteris-tics were not very useful for distinguishing between different crime scenes with aview to inferring anything uniquely meaningful about the crime scene. The fourregions surrounding the central regions represented distinct clusters of crimescene characteristics that tended to appear together sufficiently often to constitutea distinctive pattern.

Cluster 5, to the right of the central ellipse, seemed to capture a pattern of veryviolent crime scene characteristics, suggesting an element of deliberateness andcruelty in the behaviors. Following Kocsis (1999), the cluster 5 region suggested a“predator” pattern. Cluster 4, to the left of the central ellipse, seemed to capturecrime scene characteristics suggestive of less violent intent where the offenderand victim tended to be acquainted and brutality was not strongly evident—almost as if the death had not been intended. Cluster 4 suggested a “rape” pattern.Cluster 3 suggested a pattern of crime scene behaviors that had a very violent




-3.0 -2.0 -1.0 0.0 1.0 2.0 3.0

Crime Scene Behaviours Dimension 1

-3.0

-2.0

-1.0

0.0

1.0

2.0

Cri

me

Sce

ne

Beh

avio

urs

Dim

ensi

on

2

Other facial

O chgd att V

Cloth disturO disrobed V

V resisted

Care clothin

Used weapon

Redressed V

Bindng remov

Used binding

Force af sex

Fetish behav

Body moved

Hid body

Body other

O negotiated

Bindng brougPlan attack

Weapon brougRemoved weap

Strang/drown

Used knife

Took precaut

O manip beh

External sex

V stab/shot

O manip imag

Sex with V

No hide body

Postmort sexSemen elsewh

Dismember V

Binding neat

O reassured

O sex dysfun

Body prone

Used bludgeo

No care body

Multpl traum

Major trauma

Facial traum

Unpatt woundForce bf sex

Anger extrem

Took souveni

Took oth itm

Damagd cloth

Object inser

O acquaint V

O Comm Crime

Force resist

Minor trauma

Singl trauma

O threaten

V beaten

Body supine

Semen in bod

V burned

Force dg sex

Used props

Gagged VTortured V

Used firearm

Pattern woun

Used ligatur

12345

Clusters1

2

3

4

5

‘PREDATOR’

‘RAPE’

‘FURY’

‘PERVERSION’

‘UNDIFFERENTIATED’

Figure 1 Two-Dimensional Homicide Crime Scene Behavior Multidimensional Scaling Solution With Coordinate Cluster Structure Superimposed

539

at PE

NN

SY

LVA

NIA

ST

AT

E U

NIV

on Septem

ber 18, 2016ijo.sagepub.com

Dow

nloaded from


nature but with much less calculation and deliberation being evident, perhaps cou-pled with a motive of revenge—an anger theme seemed to underscore this region.Thus, cluster 3 seemed to clearly identify a “fury” pattern. Finally, cluster 2 cap-tured crime scene behaviors suggestive of an antisocial perversion theme butwithout the calculation evident in cluster 5. Cluster 2 therefore suggested a “per-version” pattern.

EXTERNAL PROPERTY VECTOR FITTING

Figures 2 through 4 summarize the property vector fitting analyses designed toexplicitly relate clusters of characteristics of the victim, the offender, and theirinteraction to facilitate interpretation and understanding of the dimensionality ofcrime scene behaviors. These variables are considered external propertiesbecause they were not used to generate the MDS solution itself. These analysesexplicitly link crime scene characteristics to conditional probability patterns that


VICTIM 1 (R=.80)VGLASSESVSCARSVOUTFEATVAGEVBUILDVSEXVHAIRLEN

VICTIM 3 (R=.81)VRACEVLIVEWTHVINCAPAC

VICTIM 2 (R =.61)VHEIGHTVLIFESTYVTRANSPT

-3.0 -2.0 -1.0 0.0 1.0 2.0 3.0


-3.0

-2.0

-1.0

0.0

1.0

2.0

Cri

me

Sce

ne

Beh

avio

urs

Dim

ensi

on

2

Other facial

O chgd att V


V resisted

Care clothin

Used weapon

Redressed V

Bindng remov

Used binding

Force af sex

Fetish behav

Body moved

Hid body

Body other

O negotiated



Strang/drown

Used knife

Took precaut

O manip beh

External sex

V stab/shot

O manip imag

Sex with V

No hide body


Dismember V

Binding neat

O reassured

O sex dysfun

Body prone

Used bludgeo

No care body

Multpl traum

Major trauma

Facial traum


Anger extrem

Took souveni

Took oth itm

Damagd cloth

Object inser

O acquaint V

O Comm Crime

Force resist

Minor trauma

Singl trauma

O threaten

V beaten

Body supine

Semen in bod

V burned

Force dg sex

Used props

Gagged VTortured V

Used firearm

Pattern woun

Used ligatur

Figure 2 Victim Characteristic Cluster Vectors Fitted to the Two-Dimensional HomicideCrime Scene Behavior Multidimensional Scaling Coordinates




would be most useful in profiling offenders and their victims. This sort of directstatistical linkage has not, to our knowledge, been made concrete within the con-text of a single database in previous research. The tendency has been to scale bothcrime scene behaviors and offender and victim characteristics in separate MDSanalyses, then making conceptual and qualitative arguments as to how the variousscaling patterns might relate (e.g., Canter & Heritage, 1989). Alternatively,Knight, Warren, Reboussin, and Soley (1998) employed a somewhat differentnonpattern-based bootstrapping methodology to predict rapist types from crimescene variables using two unrelated data sets.

Tables 1, 2, and 3 provide the numerical data (e.g., standardized regressionweights and multiple r values, averaged dimensional regression weights, andcanonical correlations) used to facilitate the property fitting exercise for the vic-tim, offender, and offender-victim interaction variable sets, respectively. Eachtable also shows the hierarchical clustering structure associated with each con-ceptual group of variables and provides the vector cluster labels to be employed inFigures 2 through 4.

Interpretation of Figures 2 through 4 is relatively straightforward, especiallywhen interpreted in conjunction with the regions identified in Figure 1. There is

-3.0 -2.0 -1.0 0.0 1.0 2.0 3.0


-3.0

-2.0

-1.0

0.0

1.0

2.0

Cri

me

Sce

ne

Beh

avio

urs

Dim

ensi

on

2

Other facial

O chgd att V


V resisted

Care clothin

Used weapon

Redressed V

Bindng remov

Used binding

Force af sex

Fetish behav

Body moved

Hid body

Body other

O negotiated



Strang/drown

Used knife

Took precaut

O manip beh

External sex

V stab/shot

O manip imag

Sex with V

No hide body


Dismember V

Binding neat

O reassured

O sex dysfun

Body prone

Used bludgeo

No care body

Multpl traum

Major trauma

Facial traum

Unpatt wound

Force bf sex

Anger extrem

Took souveni

Took oth itm

Damagd cloth

Object inser

O acquaint V

O Comm Crime

Force resist

Minor trauma

Singl trauma

O threaten

V beaten

Body supine

Semen in bod

V burned

Force dg sex

Used props

Gagged VTortured V

Used firearm

Pattern woun

Used ligatur

OFFENDER 3 (R=.93)OSEXHABOJOBTYPEODRUGALCOVEHAGEOPORNCOLOVEHCOND

OFFENDER 5 (R=.91)OCRIMSTOINTERSTOPRISEXOFOAGEOSEXPARA

OFFENDER 4 (R=.88)OGROOMOVEHTYPEODETCOLLOMARITALOMENPROBOLANGOVEHUSED

OFFENDER 1 (R=.89)OACCENTOINTERNAOLIVEWTHOHAIRLENOVEHSTATOSCAR

OFFENDER 2 (R=.69)OTRANSPTOHAIRSTYORACEOHEIGHTOMENTILL

ACCOMPLICES(r=.83)

MULTIPLE VICTIMS(r=.61)

Figure 3 Offender Characteristic Cluster Vectors Fitted to the Two-Dimensional HomicideCrime Scene Behavior Multidimensional Scaling Coordinates



one plot for each of the three conceptual sets of variables. Each fitted vector on aplot summarizes the average relationship that exists between a specific cluster ofexternal characteristics and the 2 dimensions (i.e., the spatial pattern) of the MDSsolution. The strength of the relationship is measured by the canonical correlationbetween the conditional probability scores for the variables in the cluster and thetwo dimension coordinates. The direction of the relationship directly reflects thecombination of signs of the averaged standardized regression weights for the twodimensions. Thus, movement toward the arrowhead for any vector is interpretedas reflecting an increasing tendency (i.e., the conditional probability that the vari-able takes on a value of 1 when a crime scene characteristic takes on a value of 1)for the variables making up the vector’s cluster to take on a coded value of 1 inconjunction with the crime scene characteristics near the arrowhead and viceversa when moving toward the tail of the vector.


-3.0 -2.0 -1.0 0.0 1.0 2.0 3.0


-3.0

-2.0

-1.0

0.0

1.0

2.0

Cri

me

Sce

ne

Beh

avio

urs

Dim

ensi

on

2

Other facial

O chgd att V


V resisted

Care clothin

Used weapon

Redressed V

Bindng remov

Used binding

Force af sex

Fetish behav

Body moved

Hid body

Body other

O negotiated



Strang/drown

Used knife

Took precaut

O manip beh

External sex

V stab/shot

O manip imag

Sex with V

No hide body


Dismember V

Binding neat

O reassured

O sex dysfun

Body prone

Used bludgeo

No care body

Multpl traum

Major trauma

Facial traum


Anger extrem

Took souveni

Took oth itm

Damagd cloth

Object inser

O acquaint V

O Comm Crime

Force resist

Minor trauma

Singl trauma

O threaten

V beaten

Body supine

Semen in bod

V burned

Force dg sex

Used props

Gagged VTortured V

Used firearm

Pattern woun

Used ligatur

INTERACT 5 (R=.76)ICOMMUNCVCLOTHCLIVQUARRCOMMUM

INTERACT 4 (R=.85)IOUTDOORRVCLOTHICONTACT

INTERACT 1 (R=.81)ILIVQUARCPUBPLACIPRIORACIPUBPLAC

INTERACT 2 (R=.81)CISAMERISAMEIVCLOTH

INTERACT 3 (R=.64)COUTDOORCINOUTRCSAMECFAMSITE

Figure 4 Offender-Victim Interaction Characteristic Cluster Vectors Fitted to the Two-DimensionalHomicideCrimeSceneBehaviorMultidimensionalScaling Coordinates



TABLE 1SUMMARIES OF THE COMPOSITION OF VICTIM CHARACTERISTIC CLUSTERS

USED IN THE PROPERTY VECTOR MDS FITTING ANALYSES

Victim Characteristics Variable Set

HierarchicalAverage Average Individual Individual Clusteringβ Weight β Weight Canonical β Weight β Weight Regression Cluster Using Ward’s

Dimension 1 Dimension 2 Correlation Dimension 1 Dimension 2 Multiple R Identifier Variable Method

–.536 .324 .61 VGLASSES–.467 .411 .60 VSCARS–.385 .395 .53 VOUTFEAT

–.334 .416 .80 –.236 .324 .39 VICTIM 1 VAGE–.266 .433 .49 VBUILD–.257 .478 .53 VSEX–.190 .550 .57 VHAIRLEN.339 .277 .44 VHEIGHT

.364 .118 .61 .546 .067 .53 VICTIM 2 VLIFESTY.234 –.052 .24 VTRANSPT.136 –.527 .54 VRACE

–.292 –.467 .81 –.160 –.355 .39 VICTIM 3 VLIVEWTH–.671 –.395 .78 VINCAPAC

NOTE: MDS = multidimensional scaling. Each variable included in this table achieved a significant (p< .05) multipleR value when predicted by the two-dimensional co-ordinates from the MDS solution.

543

at PE

NN

SY

LVA

NIA

ST

AT

E U

NIV

on Septem


Dow

nloaded from


TABLE 2SUMMARIES OF THE COMPOSITION OF OFFENDER CHARACTERISTIC CLUSTERS

USED IN THE PROPERTY VECTOR MDS FITTING ANALYSES

Offender Characteristics Variable Set

Average Average Individual Individual Hierarchicalβ Weight β Weight Canonical β Weight β Weight Regression Cluster Clustering Using

Dimension 1 Dimension 2 Correlation Dimension 1 Dimension 2 Multiple R Identifier Variable Ward’s Method

–.453 .448 .33 OACCENT–.612 .271 .34 OINTERNA

–.269 .362 .89 .052 .325 .42 OFFENDER 1 OLIVEWTH.019 .335 .63 OHAIRLEN

–.116 .405 .59 OVEHSTAT–.445 .402 .66 OSCAR–.460 –.101 .53 OHAIRSTY–.222 –.346 .42 OTRANSPT

–.334 –.160 .69 –.527 –.054 .47 OFFENDER 2 ORACE.124 –.470 .41 OHEIGHT

–.391 –.159 .49 OMENTILL.346 –.386 .75 OSEXHAB.707 –.276 .74 OJOBTYPE

.412 –.422 .93 .547 –.542 .50 OFFENDER 3 ODRUGALC.268 –.467 .55 OVEHAGE.222 –.446 .52 OPORNCOL.381 –.415 .49 OVEHCOND.374 .339 .37 OGROOM.351 –.109 .46 OVEHTYPE.220 .554 .25 ODETCOLL

.395 .198 .88 .219 .129 .59 OFFENDER 4 OMARITAL.403 .204 .45 OMENPROB.609 .477 .50 OLANG.462 –.011 .77 OVEHUSED.726 –.081 .72 OCRIMST.613 .011 .61 OINTERST

.678 –.029 .91 .711 .102 .63 OFFENDER 5 OPRSEXOF.625 –.029 .73 OAGE.822 –.040 .82 OSEXPARA

NOTE: MDS = multidimensional scaling. Each variable included in this table achieved a significant (p < .05) multipleR value when predicted by the two-dimensional coordinates from the MDSsolution.

544

at PE

NN

SY

LVA

NIA

ST

AT

E U

NIV

on Septem


Dow

nloaded from


TABLE 3SUMMARIES OF THE COMPOSITION OF OFFENDER-VICTIM INTERACTION CHARACTERISTIC CLUSTERS USED IN THE PROP-

ERTY VECTOR MDS FITTING ANALYSES

Offender-Victim Interaction Characteristics Variable Set

HierarchicalAverage Average Individual Individual Clusteringβ Weight β Weight Canonical β Weight β Weight Regression Cluster Using Ward’s

Dimension 1 Dimension 2 Correlation Dimension 1 Dimension 2 Multiple R Identifier Variable Method

–.445 –.487 .66 ILIVQUAR–.305 –.359 .81 –.181 –.419 .45 INTERACT 1 CPUBPLAC

–.274 –.272 .38 IPRIORAC–.202 –.239 .31 IPUBPLAC–.578 –.330 .66 CISAME

–.753 –.052 .81 –.619 –.096 .63 INTERACT 2 RISAME–.738 .066 .74 IVCLOTH–.323 .395 .51 COUTDOOR

–.331 .353 .64 –.305 .303 .43 INTERACT 3 CINOUT–.267 .126 .30 RCSAME–.175 .326 .37 CFAMSITE.517 .514 .75 IOUTDOOR

.524 .439 .85 .468 .423 .65 INTERACT 4 RVCLOTH.588 .380 .72 ICONTACT.695 .134 .72 ICOMMUN

.556 –.125 .76 .511 .019 .51 INTERACT 5 CVCLOTH.522 –.288 .58 CLIVQUAR.537 –.334 .61 RCOMMUN

NOTE: MDS = multidimensional scaling. Each variable included in this table achieved a significant (p< .05) multipleR value when predicted by the two-dimensional co-ordinates from the MDS solution.

545

at PE

NN

SY

LVA

NIA

ST

AT

E U

NIV

on Septem


Dow

nloaded from


Victim characteristics. Figure 2 shows the fitted property vectors for the threeidentified clusters of victim characteristics. The Victim 1 vector showed higherconditional probabilities with crime scene characteristics falling into the Furycluster. Thus, there were higher probabilities of victims, at crime scenes where theFury variables took on a coded value of 1, wearing glasses, having scars or othermarks or outstanding features, being older and of larger build, being female, andhaving longer hair length. The Victim 2 vector suggested higher probabilities ofvictims, at crime scenes where the Predator variables took on a coded value of 1,being taller, having a more criminal lifestyle, and relying on others for their trans-portation. The Victim 3 vector revealed higher probabilities of victims, at crimescenes where the Perversion variables took on a coded value of 1, being non-White, living with others, and being incapacitated at the time of initial contact.

Offender characteristics. Figure 3 shows the fitted property vectors for the fiveidentified clusters of offender characteristics. Also fitted were two offender-related single variable vectors, gleaned from the 65 case files, addressing whetheror not accomplices were involved, along with the offender, in the crime and thenumber of offences committed by the offender identified for each crime. TheOffender 1 vector suggested higher probabilities of offenders, at crime sceneswhere the Predator variables took on a coded value of 1, having an accent, havingtravelled internationally within the past 10 years, living with others, having longerhair length, not owning their vehicles, and having scars or other identifyingmarks. The Offender 2 vector revealed higher probabilities of offenders, at crimescenes where the Rape variables took on a coded value of 1, relying on others fortheir transportation, having an unkempt hair style, being non-White, being taller,and showing evidence of mental illness. The Offender 3 vector showed higherprobabilities of offenders, at crime scenes where the Perversion variables took ona coded value of 1 (spatially, this vector bordered on the Predator region as well),having homosexual/bisexual sex habits, being employed, showing evidence ofdrug/alcohol use, driving an older vehicle, having a collection of pornography,and having a vehicle in exceptionally good condition. The Offender 4 vector sug-gested higher probabilities of offenders, at crime scenes where the Predator vari-ables took on a coded value of 1, being well-groomed, driving a van/jeep/truck,owning a collection of detective magazines, being married, having a history ofmental problems, being bilingual, and using a vehicle in the crime. Similarly tothe Offender 4 vector, the Offender 5 vector revealed higher probabilities ofoffenders, at crime scenes where the Predator variables took on a coded value of 1,being on statutory release, having travelled interstate within the past 10 years,having a history of prior sex offences, being older, and having a collection of sex-ual paraphernalia. The Accomplices vector showed an increasing likelihood foraccomplices to be involved in crimes where the Predator variables took on codedvalues of 1. Similarly, the Multiple Victims vector showed an increasing likeli-hood for offenders with a longer series of offences to be involved in crimes wherethe Predator variables took on coded values of 1.




Offender-victim interactions. Figure 4 shows the fitted property vectors for thefive identified clusters of offender-victim interaction characteristics. The Inter-act 1 vector suggested higher probabilities of offender-victim interactions, atcrime scenes where the Perversion variables took on a coded value of 1 (spatially,this vector bordered on the Rape region as well), where the initial contact was inthe victim’s living quarters, where the crime scene was in a public place, where theoffender had a history of prior activity in the initial contact area, and where initialcontact was in a public place. The Interact 2 vector revealed higher probabilities ofoffender-victim interactions, at crime scenes where the Rape variables took on acoded value of 1, where the crime scene and initial contact scene were the same,where the recovery site and the initial contact site were the same, and where some-thing had been done to the victim’s clothing at the initial contact site. The Inter-act 3 vector showed higher probabilities of offender-victim interactions, at crimescenes where the Fury variables took on a coded value of 1, where the crime scene/site was outdoors, where the crime scene and the recovery site were the same, andwhere the offender was unfamiliar with the crime site. The Interact 4 vectorrevealed higher probabilities of offender-victim interactions, at crime sceneswhere the Predator variables took on a coded value of 1, where the initial contactsite was outdoors, and where something was done to the victim’s clothing at therecovery site. Similarly, the Interact 5 vector revealed higher probabilities ofoffender-victim interactions, at crime scenes where the Predator variables took ona coded value of 1, where the initial contact community was in a noncity location,where something was done to the victim’s clothing at the crime scene, where thecrime scene was in the victim’s living quarters, and where the recovery site com-munity was in a noncity location.

DISCUSSION

The results depicted in Figure 1 represent an attempt to present empirically acoherent model of sexual murder behaviors. The central cluster (undifferentiatedbehaviors) indicates behavior common to all offences of sexual murder. Sur-rounding this central cluster are four empirically different patterns of behaviorthat each correlate with distinctive offender characteristics (Figures 2-4). Conse-quently, this model allows for the interpretation of murder behavior patterns andthe identification of probable offender characteristics associated with each of thediscerned behavior patterns.

Three key themes appear to characterize the behaviors in the central cluster:intercourse with the victim, violence, and premeditation/precaution in the perpe-tration of the offence. The incidence of sex and violence is perhaps unsurprisinglya central theme of sexual murder and concurs with the commonly cited drives forsexual murder (Hickey, 1997; Holmes & Holmes, 1998). Indeed, this combina-tion of sexuality and violence also concurs with the basic premise by Groth et al.(1977) for sexual assault. However, the exact relationship between these two




themes—that is, whether the primary theme is the expression of violence and con-trol through sexuality or vice versa—is indeterminable. Given that the victim diesduring the encounter, it could be inferred that violence and force are indeed pri-mary factors.

The presence of preparatory and precautionary behaviors in the centralundifferentiated cluster serves as support for the expansion of the organized/disorganized behavior maxim beyond a simple dichotomy and into a more sophis-ticated continuum. The basic premise of the dichotomy is the categorical distinc-tion of crime behavior patterns by their level of sophistication. The key measurefor this level of sophistication is indications of planning for the offence such as theundertaking of precautions to elude apprehension. The presence of preparation/precautions as a central theme weakens this categorical distinction and indicatesthat all patterns commonly share some level of sophistication and then diverge outtoward the poles of a conceptual continuum.

The undifferentiated behaviors also assist in the construction of profiles via areductionist process. In line with previous literature (e.g., Ressler et al., 1988),behaviors located in this central pattern were interpreted as indicative of certainoffender characteristics. For example, the use of restraints or the removal of aweapon were identified as being key features of an organized offender, and fromthis conclusion various organized offender characteristics were espoused (Ressleret al., 1988, p. 123). With the current model, it can be seen that identification ofoffender characteristics cannot rely on the presence of these undifferentiatedcrime behaviors, as these actions are common to all patterns of sexual murderbehavior. Perhaps the only inference that can be made from the presence of theseundifferentiated behaviors is that the crime can be inferred to be a sexual murder.

In considering the outlying behavior patterns in Figure 1, there are a number ofconceptual similarities with previous research literature on sexual murder. Dem-onstrating the highest level of behavioral congruence is the predator pattern. Thispattern shares many similarities with the “hedonist” killer proposed by Holmesand Holmes (1998), the “lust” killer by Hickey (1997), and the archetypal “orga-nized” or “sadistic” offender of the FBI (Ressler et al., 1988; Douglas et al.,1992)—this being a sexually sadistic predator who tortures and rapes the victimfor pleasure.

The description of a hedonist killer given by Holmes and Holmes (1998)loosely correlates with the predator pattern. However, the subdivision made byHolmes and Holmes into the thrill and lust categories is based on the presence orabsence of postmortem sexual activity. Unfortunately, the current model identi-fies postmortem sexual activity as part of the perversion pattern. Consequently,the current model cannot support this subcategorization made by Holmes andHolmes (1998). Speculation that the perversion pattern may represent the lust cat-egory would be ill conceived, as this pattern holds many features that are highlyincompatible with the broad conception of hedonistic offenders. Clearly, a reeval-uation of the Holmes categories in light of these results is warranted.




Offender characteristics for the predator pattern concur well with previous lit-erature. Offenders are typically older, mobile, living with a partner, and from aWhite racial background. They are well groomed and are typically collectors ofcrime literature and sexual materials and are highly prone to reoffend. However,offenders in the predator pattern exhibit a high tendency to operate with an accom-plice. Here, a significant quandary arises as to how this result may relate toHickey’s (1997) distinction between “team” offenders, who are described asbeing driven by different psychological imperatives, influences, and consider-ations, and the subcategory of “male solo killers” (also known as lust killers).Both categories are equally well encompassed by the predator pattern and conse-quently serve to question this distinction.

In previous literature, the presence of souvenir or token collection behaviorswould indicate an offender in the predator pattern (Douglas et al., 1992, p. 126;Holmes & Holmes, 1998, p. 89; Hickey, 1997, p. 17). However, in the currentmodel, souvenir and token collection features in the adjacent fury pattern. Irre-spective of the statistical classification of these behaviors within the fury pattern,they appear in a close bordering proximity to the predator pattern. This impliesthat although these behaviors are statistically distinguished as being within thefury pattern, they can nonetheless appear to be associated with the predator pat-tern when adopting a broad directional interpretation of the model in contrast tothe present regional clusters.

The fury pattern represents an explosive, unfocussed obliteration of the victim.A number of similarities exist between the fury pattern and previous literature.The excessive uncoordinated violence and overall disorganization of the patterndoes demonstrate a similarity to the “visionary killer” espoused by Holmes andHolmes (1998) or the archetypal disorganized offender category from the FBIorganized/disorganized dichotomy. However, with both of these categories thereis the implicit assumption that the actions are the product of an individual behav-ing under some form of psychotic delusion. The present results, however, indicatethat there is an equal propensity for offenders to possess, or not to possess, a men-tal disorder. Consequently, although a proportion of offenders within the fury pat-tern may be identified as being of a violent, mentally ill orientation, an equal pro-portion are unlikely to suffer any such disorder. This obviously conflicts with thevisionary or disorganized offender hypothesis.

An alternative interpretation of the fury pattern could be developed from theliterature on sexual assault and specifically the “anger retaliation” rapist espousedby Groth et al. (1977). The anger retaliation rapist is described as a nonpsychoticoffender who possesses an irrational, deep hatred that is expressed as a violentsexual assault. If this category were extended into the domain of homicide wherethe offender’s assault actually kills the victim, it would plausibly explain thenonpsychotic element in the fury pattern. Indeed, this rational but hateful offenderhypothesis could explain the retention of souvenirs or trophies in the fury pattern.The taking of such items may serve as a reminder to the offender of their retribu-tion or as an added act of defilement and hatred.




In contrast to the general theme of violence exhibited within the predator andfury patterns, the rape pattern represents an offender primarily pursuing inter-course with only the necessary use of force to perpetrate the assault. Conse-quently, the victim is threatened for compliance and force is only applied toachieve control. This application of force is typically minimal with a few woundsbeing inflicted on the victim’s body. There is no indication of sexual dysfunctionwith penetration occurring and semen usually found in the victim. Thus, this pat-tern seems to resemble a sexual assault that has resulted in murder. Indeed, it isopen to speculation as to whether murder was the original intention of the offenderor an outcome of the offence. For example, in attempting to control the victim, theoffender may deliver a blow that actually kills the victim. The crime scene behav-iors and associated offender characteristics do not demonstrate any great degreeof perversion or sadism but rather a simple brutal pursuit of intercourse.

Indeed, a unique component of this pattern is that offenders often have someprior acquaintance with the victim. This does not necessarily imply a prior rela-tionship between the victim and offender but rather that the offender was aware ofthe victim’s existence prior to the offence. Interaction characteristics of this pat-tern indicate that these offences typically occur in a single indoor location.Through the combination of all of these features, a general scenario for this pat-tern emerges in the form of a single offender pursuing intercourse and invading avictim’s home. The offence is not planned but more characteristic of an impetuousyounger individual who discovers a potential victim and perhaps acts on animpulse of lust and/or desire.

The rape pattern represents another challenge to previous literature as it is dif-ficult to associate this pattern with any of the previous homicide categoriesdescribed by Holmes and Holmes (1998) or Hickey (1997). The pattern coincideswith the behavioral principles of a disorganized offender, however the disorga-nized category does not adequately explain the observable behavior themes. Theclosest link to the rape pattern stems from the literature on sexual assault. Follow-ing the categories of Groth et al. (1977), the rape pattern does exhibit many facetsof the “power assertive” rapist in pursuing sexual gratification for internal reas-surance. Concurrently, the primary behavioral theme of pursuing intercourse alsobears a similarity to the sexuality category of rapist identified by Canter and Heri-tage (1989).

Finally, a unique combination of factors makes up the perversion pattern. Vir-tually all crime behaviors in this pattern are related to extreme paraphilic/perverseactivity. In contradiction to these perversities, however, offenders are noted toengage the victim in conversation by offering reassurance. Various factors operat-ing in combination serve to explain this pattern. Victims in this pattern typicallytend to be younger and male, whereas offenders tend to be older with bisexual orhomosexual orientations. The combination of these offender and victim charac-teristics indicates that these offences typically involve some form of paedophilicassault that results in the murder of the victim. What may be the motive for the




final murder of the victim is contentious. Although this behavior pattern demon-strates a passivity in comparison to the violence observed in other patterns, thebehaviors present in this pattern are planned and consequently serve to indicatethat the murder of the victim is an intentional component of the assault. Given thatthis pattern is adjacent to the predator pattern indicates that ritualism is a compo-nent of the offence. Hickey (1997) described a specific category of offender whospecializes in the murder of children. The characteristics of the perversion patternmatch the behavior patterns, and offender characteristics indicate some form ofvariant lust killer of children.

In conclusion, although this study has demonstrated a coherent model for sex-ual murder behaviors and offender characteristics, it represents only a startingpoint from which further inquiry should be based. A start for future study wouldbe to undertake structured interviews with offenders and develop qualitativematerial to augment the present model. Indeed, such a study could probe issuessuch as family history, triggering activities to an offence, or the mental processinvolved in an offender’s selection of an ideal victim. Unfortunately, suchinsights, which would add both investigative and conceptual value to our under-standing of sexual murder, cannot be derived from the present analysis.

Clearly, the results demonstrate that sexual murder does involve a relationshipbetween sexual activity and violence. However, the exact dynamics behind thesetwo themes remain elusive. Whether sexual murder offences actually representserious rape offences that have escalated in violence and result in the victim’sdemise is contentious. The very occurrence of the rape pattern in the presentmodel is an indication of such an association. However, in sharp contradiction wehave the behaviors of the predator and fury patterns, where the victim’s death is anintegral element of the offence. Developing a better understanding of the relation-ship between serious rape and sexual murder will involve a closer study of seriousrape offenders (Kocsis, Cooksey, & Irwin, 2002).

Having established an empirical model of sexual murder, it would now beadvantageous to replicate such a study to assess its validity and utility in lawenforcement investigations. Previous attempts to construct offender typologieshave lacked empirical structure, and this has limited their suitability for quantita-tive comparisons. Furthermore, this model provides a direct empirical linkbetween crime behaviors and offender characteristics, another form of analysisthat has been lacking in previous research. This model represents the first everattempt to study sexual murder in Australia for the purpose of psychological pro-filing and its use in law enforcement investigations. Subsequently, a study usingsimilar sampling and methodological techniques from an overseas populationwould be beneficial in establishing any cross-cultural generalizations and indeedthe robustness of the present model. With this model the foundation has now beenclearly established for further research into this area and the development of sex-ual murder psychological profiling into a systematic and structured science.




NOTES

1. To illustrate the first few steps of the property fitting process, consider the steps taken to analyzethe variables in the victim characteristics set. The 14 variables making up this set were individuallyregressed onto the two standardized multidimensional scaling (MDS) dimension coordinates and theresulting standardized (beta) weights for the two dimensions were recorded (thus, 14 distinct regres-sion were run for this set—although the set correlation method in SYSTAT ran them all simulta-neously). Out of the 14 original variables in the set, only the victim’s marital status (VMARITAL) wasnot significantly predicted by the MDS coordinates and was therefore dropped from further consider-ation. The remaining set of weights (13 variables by two weights) was cluster analyzed and three dis-tinct clusters of variables were identified. One such cluster made up the VRACE, VLIVEWTH, andVINCAPAC variables and the standardized weights for each dimension were averaged across thethree, giving an average beta weight for Dimension 1 and an average beta weight for Dimension 2.These two averaged weights then formed the basis for fitting the property vector to the two-dimen-sional MDS solution plot.

2. The specific formulas used for this transformation were as follows, where a and b reflect the twodimensions that anchor the graph in question:

if (β a > 0 and β b > 0) angle = deg (cos ( / ))− +1 2 2β β βa a b

if (β a > 0 and β b < 0) angle = 360 1 2 2− +−deg (cos ( / ))β β βa a b

if (β a < 0 and β b > 0) angle = 180 1 2 2− +−deg (cos (| |/ ))β β βa a b

if (β a < 0 and β b < 0) angle = 180 1 2 2+ +−deg (cos (| |/ ))β β βa a b

3. The MDS figures are as clear as can be made without deleting actual plotted points in the center.As the major interpretative work for the article is linked to the regions around the center, the detail inthe center is not really essential in terms of understanding the overall interpretation of the patterns.

FINAL NOTE

During the submission and peer review process of this article, an MDS analysisof U.S. serial killers has appeared in book format. Although attaining earlier pub-lication, this study used different sampling procedures and thus analyzed differentdata from what would have been included in this study.

REFERENCES

Canter, D. V., & Heritage, R. (1989). A multivariate model of sexual offence behaviour: Developmentsin “offender profiling.” I. Journal of Forensic Psychiatry, 1(2), 185-212.

Cohen, J., & Cohen, P. (1983).Appliedmultiple regression/correlation analysis for the behavioral sci-ences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum.

Coxon, A.P.M. (1982). The user’s guide to multidimensional scaling. London: Heinemann Educa-tional Books.

Douglas, J. E., Burgess, A. W., Burgess, A. G., & Ressler, R. K. (1992).Crime classification manual.New York: Simon & Schuster.




Groth, N. A., Burgess, A. W., & Holmstrom, L. L. (1977). Rape: Power, anger and sexuality.AmericanJournal of Psychiatry, 134(11), 1239-1243.

Hair, J. F., Jr., Anderson, R. E., Tatham, R. L., & Black, W. G. (1998).Multivariate data analysis (5thed.). Upper Saddle River, NJ: Prentice Hall.

Hazelwood, R. R., & Burgess, A. W. (1987). An introduction to the serial rapist research by the FBI.FBI Law Enforcement Journal, 56(9), 16-24.

Hazelwood, R. R., & Burgess, A. W. (1995). Practical rape investigation: A multidisciplinaryapproach (2nd ed.). Boca Raton, FL: CRC Press.

Hickey, E. W. (1997). Serial murders and their victims (2nd ed.). Fresno, CA: Wadsworth.Holmes, R. M., & Holmes, S. T. (1998). Serial murder (2nd ed.). Thousand Oaks, CA: Sage.Jackson, J. L., & Berkerian, D. A. (1997). Does offender profiling have a role to play? In J. L. Jackson

& D. A. Berkerian (Eds.),Offender profiling: Theory, research and practice (pp. 1-7). Chichester,UK: Wiley.

Knight, R. A., Warren, J. I., Reboussin, R., & Soley, B. J. (1998). Predicting rapist type from crime-scene variables. Criminal Justice and Behavior, 25(1), 46-80.

Kocsis, R. N. (1999). Criminal profiling of crime scene behaviors in Australian sexual murders. Aus-tralian Police Journal, 53(2), 113-116.

Kocsis, R. N. (in press-a). Criminal psychological profiling: Validities and abilities? InternationalJournal of Offender Therapy and Comparative Criminology.

Kocsis, R. N. (in press-b). Psychological profiling in serial arson offenses: A comparative assessmentof skills and accuracy. Criminal Justice and Behavior.

Kocsis, R. N., Cooksey, R. W., & Irwin, H. J. (2002). Psychological profiling of offender characteris-tics from crime behaviors in serial rapists. International Journal of Offender Therapy and Com-parative Criminology, 46, 144-169.

Kocsis, R. N., Hayes, A. F., & Irwin, H. J. (2002). Investigative experience and accuracy in psycholog-ical profiling a violent crime. Journal of Interpersonal Violence, 17, 811-823.

Kocsis, R. N., & Irwin, H. J. (1998). The psychological profile of serial offenders and a redefinition ofthe misnomer of serial crime. Psychiatry, Psychology, and Law, 5(2), 197-213.

Kocsis, R. N., Irwin, H. J., & Hayes, A. F. (1998). Organised and disorganised behaviour syndromes inarsonists: A validation study of a psychological profiling concept. Psychiatry, Psychology, andLaw, 5, 117-130.

Kocsis, R. N., Irwin, H. J., Hayes, A. F., & Nunn, R. (2000). Expertise in psychological profiling: Acomparative assessment. Journal of Interpersonal Violence, 15(3), 310-330.

Kruskal, J. B., & Wish, M. (1978). Multidimensional scaling. Beverly Hills, CA: Sage.Pinizzotto, A. J., & Finkel, N. J. (1990). Criminal personality profiling: An outcome and process study.

Law and Human Behavior, 14, 215-233.Ressler, R. K., Burgess, A., & Douglas, J. E. (1988). Sexual homicide: Patterns and motives. New

York: Lexington Books.Wilson, P. R., Lincoln, R., & Kocsis, R. N. (1997). Validity, utility and ethics of profiling for serial vio-

lent and sexual offenders. Psychiatry, Psychology, and Law, 4, 1-12.Wilson, P. R., & Soothill, K. (1996, January-March). Psychological profiling: Red, green or amber?

The Police Journal, 69(1), 12-20.

Richard N. Kocsis, Ph.D.Forensic Psychologist in Private Practice16 Lynden AvenueCarlingfordSydney, NSW 2118Australia




Ray W. Cooksey, Ph.D.School of Marketing and ManagementUniversity of New EnglandArmidale, NSW 2351Australia

Harvey J. Irwin, Ph.D.School of PsychologyUniversity of New EnglandArmidale, NSW 2351Australia




Psychological Profiling of Sexual Murders: An Empirical Model

Documents

Transcript of Psychological Profiling of Sexual Murders: An Empirical Model