Professional Documents
Culture Documents
http://jiv.sagepub.com/
Violence
Published by:
http://www.sagepublications.com
On behalf of:
American Professional Society on the Abuse of Children
Additional services and information for Journal of Interpersonal Violence can be found at:
Subscriptions: http://jiv.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://jiv.sagepub.com/content/18/5/490.refs.html
What is This?
C. GABRIELLE SALFATI
Centre for Investigative Psychology, University of Liverpool, United Kingdom
INSTRUMENTAL
foreign object
covered
naked anal
prop.not.id arson
prop.v weapon fr.
bound manual neck
clothdam
sexual
forensic
torso face partundr vaginal
blunt head MWD
prop.id
limbs
weapon to
hidden
transported
shot
suffocated
blindfolded
water
drugged/poisoned
EXPRESSIVE
Figure 1: Smallest Space Analysis Illustration of 36 Behaviors Across 247
Homicides
SOURCE: Adapted from Salfati (2000).
NOTE: For explanation of variable names, please see Tables 1 through 4.
between the behaviors. In this case, the closer two behaviors are on the plot
the more often they occur in the same case across the sample (for a more
detailed explanation of SSA, see the “Method” section of this article).
With expressive homicides, there was often an extreme physical attack.
Moreover, these behaviors, when looked at together, are suggestive of actions
centered on needing to separate from the victim and the place of the crime, as
these elements might have aided the identification of the killer. All of these
behaviors, Salfati (2000) suggested, show a prior relationship between the
two parties, or at least suggest that the offender knew the victim to some
extent. Actions in the instrumental theme, on the other hand, centered on
behaviors that were not singularly directed at the victim as a person. Rather,
the actions were part of a larger theme of the offender using the victim to fur-
ther attain an ulterior aim such as sex or money. When taken together, these
actions suggested a behavioral theme where the offender regarded the victim
not as a person with whom they were personally interacting but as an object
ultimately to be used for personal gain.
Through the analysis of the co-occurrences of the actual behaviors used by
offenders at homicide crime scenes, Salfati’s (2000) study brings attention to
the actual behavioral components that make up different themes of homicide,
such as expressive and instrumental crime scenes. The study also takes into
consideration that a number of behaviors occurred in the majority (50% and
above, see dark points in the middle of the plot) of all the cases, and these
behaviors were thus not used to discriminate between cases. However,
although this study did take a more empirical approach to the examination of
behaviors as they occur at crime scenes, it was not in the scope of that study to
provide a more in-depth discussion of how the actual frequencies of the
behaviors in each of the two themes related to the structural makeup of each
of the themes.
The aim of this study will therefore be to more fully investigate the pat-
terns of co-occurrences of crime scene actions as evidenced by the different
types of behaviors that were executed at homicide crime scenes. Moreover, it
is suggested that the patterns of co-occurrences will show a consistency in the
thematic content of different subgroups of behaviors that tended to co-occur
at homicide crime scenes. Based on the literature, particular emphasis will be
put on understanding these differences in terms of the meaning of the victim
to the offender and the way the offender manifested impulsivity.
THE DATA
So that the results from the current study could be directly used to build on
the previous literature, the data used were the same as those used for the study
by Salfati (2000). These data consisted of 247 British single offender-single
victim solved homicide cases spanning from the 1970s to the early 1990s.
The data were collected from various British police forces around the
country, as well as the Crown Prosecution Service. A thorough randomiza-
tion of cases was not possible due to the nature of the data and issues of avail-
ability and access. However, the data did include cases from many areas of
England and Wales, throughout the time period investigated.
to the hypotheses being tested in this study, a number of cases were excluded
from this study. These are outlined below.
The excluded cases included cases of death by reckless driving, as this
type of homicide does not explicitly deal with the interpersonal relationships
between the offender and the victim. Nor were cases included that very
clearly involved professional hit men, as it was felt that the professional
aspect of these crimes would interfere with the issues at hand. Cases that were
clearly defined as euthanasia cases, where the offender(s) was deemed
legally sane when the life of the victim was taken and where the victim was
killed as an act of mercy rather than as an explicit act of aggression, were also
excluded. Cases where offenders killed family members for the reason that
they did not want them to suffer from any type of pain, whether caused by dis-
ease or not, were included in the sample only if the offenders were diagnosed
as mentally ill, or not legally sane, at the time of the offense.
THE VARIABLES
Data from police reports of homicide crime scenes are rich in information
about the actions that occur at crime scenes and were therefore used as the
basic data source. From the case files, 36 crime scene behaviors were selected
for the analysis of co-occurrences across 247 cases of homicide. This selec-
tion was based on several criteria:
1. Behaviors that were clearly observable at the crime scene and were not easily
misinterpreted.
2. Behaviors that reflected the crime of homicide, such as the type of wounding
the victim had sustained and where the victims were found.
3. Behaviors that indicated more specialized behaviors regarding how the
offender acted toward the victim, in particular with reference to the intensity
of this involvement, such as an “overkill” attack (e.g., multiple wounds dis-
tributed all over the victim’s body) as well as more removed or controlling
interaction such as binding or blindfolding the victim.
These variables were recorded in dichotomous form with values based on the
presence or absence of each behavior in any one file. Dichotomies were used
because the information was drawn from police records not initially collected
for research purposes. Previous research has also indicated that for content
analysis, any distinctions more refined than presence-absence dichotomies
are likely to be unreliable (Canter & Heritage, 1990).
SAMPLE DEMOGRAPHICS
For the purposes of generalizability of the results from the analysis of the
data sample, comparisons were made between trends of the current data set
and national figures relating to homicide (as provided by Home Office Crimi-
nal Statistics England and Wales, 1977-1995) as well as the figures provided
by the general literature on homicide.
Looking at those variables that could be compared to Home office crimi-
nal statistics (see Salfati, 1998, for more details), it could be seen that the
present sample had a larger percentage of victims known to the offender.
Similarly, the present sample had a smaller sample of stranger cases. In part,
this may be due to the issue of definition in that these variables may have been
defined slightly differently for the present purpose. Regardless of problems
of definition, it can be said however that the present sample contained more
cases where there was a prior relationship between the offender and the
victim.
Victim Demographics
Of the sample of 247 victims, the gender of the victim was known for 246
cases. In all, 44% were male, and 56% were female.1
For the 232 cases (94%) where information about the victim’s age was
available, the range of ages spanned from newborns to victims who were 93
years old. The mean age for the total sample was 39 years. Separating the
sample into males and females revealed a slightly younger mean age of 36
years for male victims (n = 99, range 0 to 82) and a slightly older mean age for
female victims of 41 years (n = 133, range 0 to 93).
Offender Demographics
Information regarding the gender of the offender was known for 246 of the
247 cases. Of these, 89% were male, and 11% were female.
For the 239 cases where information was available as regards the age of
the offender, the range was from 16 years to 79 years, with the mean age being
33 years. Splitting the sample into male and female, the mean age was 32 years
for male offenders (n = 212, range 16 to 79) and 33 years for female offenders
(n = 27, range 17 to 70).
Of the 247 offenders, 201 (81%) had a previous conviction. Of these 201
offenders, 116 (58%) had previous convictions for theft, 77 (38%) for bur-
glary, 74 (37%) for fraud, 68 (34%) for violence, 54 (27%) for disorder, 51
(25%) for public disorder, 50 (25%) for traffic offences, 46 (23%) for vehicle
theft, 22 (11%) for sexual offenses, 17 (9%) for drugs, 15 (7%) for firearms
possession, and 7 (4%) for arson. And 73 (36%) had previously served a
prison sentence.
METHOD
The data were analyzed using SSA (Lingoes, 1973). SSA allows a test of
hypotheses concerning the co-occurrence of every variable with every other
variable. In essence, the null hypothesis is that the variables have no clear
relationship to each other. SSA is a nonmetric multidimensional scaling pro-
cedure based on the assumption that the underlying structure, or system of
behavior, will most readily be appreciated if the relationship between every
variable and every other variable is examined.
Initially, association coefficients between all variables are computed. It is
these coefficients that are used to form a spatial representation of items with
points representing variables. The more often variables co-occur during
homicide the closer the points will be representing those variables in the SSA
space. The pattern of points can hence be examined and thematic structures
delineated.
The hypotheses of this study are built on the assumption that actions with
similar underlying themes will be more likely to co-occur than those that
imply different themes. These similarly themed actions will co-occur in the
same region of the plot. This regional hypothesis has previously been seen as
an appropriate way of interpreting co-occurrences of behaviors and has suc-
cessfully been used to interpret both studies of emotion and personality (see
Plutchik & Conte, 1997).
The coefficient of alienation2 (Borg & Lingoes, 1987) is an indication of
how well the spatial representation fits the co-occurrences as represented in
the matrix. The smaller the coefficient of alienation is the better the fit (i.e.,
the fit of the plot to the original matrix). However, as Borg and Lingoes
(1987) emphasized, there is no simple answer to the question of how “good”
or “bad” the representations are. This will depend on a combination of the
number of variables, the amount of error in the data, and the logical strength
of the interpretation framework. In the present study, the data are mainly
derived from police crime records, which are not created for research pur-
poses and thus do not adhere to strict collection protocol and procedures. It
would therefore be expected that the data are not error free and could contain
considerable “noise” that could reduce the possibility of interpreting the
results. A relatively high coefficient of alienation would therefore be accept-
able, provided the plot produced a meaningful interpretation.
Although SSA and similar techniques have been widely used by social
scientists over the past 30 years (e.g., Canter, 1985), using multidimensional
scaling techniques in the study of crime is a relatively new concept (e.g., Can-
ter & Heritage, 1990). Using multidimensional scaling for classifying homi-
cides has recently successfully been used by Salfati and Canter (1999),
Salfati (2000), Salfati and Haratsis (2001), and Santilla, Canter, Elfgren, and
Häkkänen (2001).
Using this methodology to analyze the co-occurrences of actions at a
crime scene enables a test of the hypothesis that sets of actions will co-occur
to form different types of interaction between the offender and the victim.
RESULTS
Figure 2 shows the same distribution of the 36 crime scene behaviors for
the 247 cases of homicide on the 1 × 2 projection of the three-dimensional
SSA analysis, as used by Salfati (2000). The coefficient of alienation of this
analysis was also 0.17224, showing a good fit of the spatial representation of
the co-occurrences of the behaviors.
The regional hypothesis of SSA states that items that have a common
underlying psychological theme will be found in the same region of the SSA
space. The manifestations of the behaviors on the plot therefore reflected
how associated each of the behaviors were to each other across the sample.
To examine the pattern of frequency of the behaviors and how these
related to each other, the frequencies for each of the behaviors were examined
(see Tables 1 through 4). Furthermore, by superimposing frequency bands
onto the SSA plot (see Figure 3), a discernible pattern of the frequencies of
occurrence of the behaviors can be seen in terms of a circular pattern of
descending frequencies from the middle of the plot out to the periphery of the
plot. Each one of these frequency bands is further expounded below.
foreignobj
covered anal
naked
arson
prop.notid
prop.v weap.fr
bound faceup neck
scene manual
clothdam
faceunh stab sexual
forensic mwoa
face
torso partundr vaginal
blunt head mwd
prop.id
limbs
weap.to
hidden
transport
shot
suffocated
blindfold
water
drug/poison
foreignobj
covered anal
naked
arson
prop.notid
prop.v weap.fr
faceup neck
bound scene manual
clothdam
faceunh stab sexual
forensic mwoa
torso face
partundr vaginal
blunt head mwd
prop.id
limbs
weap.to
hidden
transport
shot
suffocated
blindfold
water
drug/poison
60.0
50.0
43.3
40.0
30.0
25.1
20.0
19.8
Percent (%)
10.0
10.5
0.0
0 1 2 3 4
As can be seen in Figure 4, 68.4% of the 247 crime scenes had three or four
of the high-frequency variables present. As such, this further validates the
idea that these high-frequency behaviors should not be used to attempt to dif-
ferentiate between cases. The percentages also suggest that there were few
cases that had none (1.2%) or only one (10.5%) behavior present. This shows
that most of the cases included the majority of these behaviors, and thus these
behaviors can be said to exemplify the norm of behavior within the sample.
Further analysis showed that 57% of the sample had a higher percentage3
of variables present in this high-frequency band than the percentage of vari-
ables present in the other three frequency bands (described below) added
together. This also suggests that the majority of the cases in the sample could
be classified as highly impulsive in nature.
The represented case study below, taken from the sample investigated,
was typical of many impulsive homicides where the offender reacts in an
impulsive way toward the emotions engendered through the conflict-ridden
interpersonal relationship with the victim. This crime scene exemplifies the
direct, impulsive, and messy crime scene of the impulsive type of murder that
defines the majority of the cases in this sample.
When the behaviors in which the offender engaged in at the crime scene
are plotted onto the SSA plot (see variables in italics on Figure 3), the distri-
bution of the presence of the actions committed by the offender during the
homicide can be ascertained. As can be seen, the offender engaged in all of
the high-frequency actions at the crime scene (variables more than 50%),
such as leaving the victim face up, with the face unhidden, and at the same
scene as where the crime took place. The offender also engaged in most of the
other high-frequency actions in the next frequency band (30% to 50%, see
description below), such as killing the victim manually, that is through hit-
ting, kicking, and strangling the victim, as well as using a blunt instrument
and stabbing the victim. The offender in this case left wounds all over the
body of the victim and wounded certain areas of the victim’s body intensely.
In this case, the victim was wounded on the neck, the face, the torso, the head,
and the limbs. The offender was also careful not to leave forensic evidence at
the scene.
Case Study
The offender (36 years old) and the deceased (27 years old) shared a very
stormy and violent relationship. The offender was a heavy drinker and drug
user and had a long list of previous convictions for assault, theft, burglary,
disorder, possession of drugs, and the possession of a weapon. Consequently,
this lead the deceased to seek an injunction on the offender, which was
granted.
However, despite the injunction, there was a violent argument between the
two, during which the offender kicked and punched the deceased and ulti-
mately killed her through wounds caused by a blunt instrument. The
deceased sustained extensive bruising to her head, face, neck, and upper and
lower limbs and to the back of her hands. There was also bruising consistent
with the application of manual pressure on the neck.
60.0
50.0
40.0
30.0
25.1
20.0 23.1
16.2
Percent (%)
10.0
10.1 10.5
5.7 6.1
0.0
0 1 2 3 4 5 6 7 8
gation shows that 39.3% of the 247 crime scenes had four or more of the vari-
ables in this frequency category present. Only very few cases had none
(2.4%) or only one (10%) of the variables present.
60.0
50.0
40.0
30.0
27.9
26.3
20.0 22.7
Percent (%)
14.6
10.0
6.5
0.0
0 1 2 3 4 5 6
centages also show that more than half the sample had none (26.3%) or only
one (27.9%) behavior present at the crime scene. This patterns shows a clear
shift away from offenders using a range of the behaviors in this frequency
band, to offenders using only a few, very select, number of behaviors. As
such, there is evidence that these behaviors form a much more select, special-
ized group of actions.
60.0
50.0 51.8
40.0
30.0
20.0 23.1
Percent (%)
14.6
10.0
7.7
0.0
0 1 2 3 4 5
(thus not entering into direct intimate contact with the victim), and blindfolds
the victim (again, the offender removes the impact of the victim’s face).
These are all cognitive behaviors where the offenders “removes” themselves
from the victim physically and psychologically.
As can be seen in Figure 7, no offender engaged in all 16 variables within
this frequency band. Indeed, the maximum number of behaviors was 5, a
third of the possible variables. Of these 5 variables, only a very small number
of cases (10.5%) had more than half (i.e., 3 to 5 variables) present. Looking at
the distribution of these behaviors, compared with those in Figures 4, 5, and
6, a marked decrease in the number of offenders who engage in these actions
can be seen. In this group, as many as two thirds of the sample had none
(51.8%) or only one (23.1%) behavior present in this frequency band. This
again shows the specialized nature of the actions within this category.
frequency behaviors, on the other hand, have less to do with killing the victim
and more to do with offenders using actions that are controlling and that sug-
gest they are psychologically distancing themselves from the victim. In these
cases, the transaction with the victim becomes more one where the offender
removes himself or herself from the victim, both physically (e.g., shot, trans-
port, foreign object, etc.) and emotionally (e.g., blindfold, drug/poison, etc.).
The frequency structure of behavior can thus be seen to follow a contin-
uum from where the offender reacts in an impulsive way toward the emotions
engendered through the conflicted interpersonal relationship with the victim,
to where the offender interacts with the victim much more at a removed level,
both physically and emotionally.
This distribution of the frequencies of behaviors at these homicide crime
scenes pertinently reflects that homicide is an impulsive act that is acted out
at times of stress and the consequences of which may therefore not be at the
front of the mind of the offender. The majority of the behaviors reflect an
impulsive or reactive dimension to the offenders’ actions, whereas the less
frequent behaviors are more organized or purposeful and directed in nature.
Like Cornell et al. (1996) suggested, these can be seen as more psychopathic,
in the sense that the offender’s actions reflect an ulterior motive rather than
just a reactive reaction to an interpersonal conflict.
INSTRUMENTAL
Foreign
Low % variables
Medium % variables
High % variables
Drug/poison EXPRESSIVE
how the offender is distancing himself or herself from the victim. But the sug-
gestion here is that the offender does this differently, depending on whether
they are generally committing an expressive or instrumental homicide.
In conclusion, this article has attempted to illustrate that by understanding
behaviors within a multidimensional context in terms of their occurrence fre-
quency, as well as in terms of their thematic associations with other behaviors
during the crime, a more comprehensive modeling of homicide behavior may
be established.
NOTES
1. Please note that all percentages in the demographics section are rounded up or down to the
nearest whole number.
2. All smallest space analyses were analyzed using Jaccard’s correlation coefficient.
3. Please note that the percentage of variables was used rather than the actual number due to
each frequency band containing a different number of variables.
4. Please note that the behavior “wound to limbs” in the next frequency category fits in with
this categorization but because of its frequency is left in the next frequency category table.
REFERENCES
Borg, I., & Lingoes, J. C. (1987). Facet theory: Form and content. New York: Springer.
Canter, D. (Ed.). (1985). Facet theory. New York: Springer.
Canter, D., & Heritage, R. (1990). A multivariate model of sexual offence behavior: Develop-
ments in “offender profiling” I. Journal of Forensic Psychiatry, 2, 185-212.
Cornell, D. G., Warren, J., Hawk, G., Stafford, E., Oram, G., & Pine, D. (1996). Psychopathy in
instrumental and reactive violent offenders. Journal of Consulting and Clinical Psychology,
64(4), 783-790.
Curtis, L. A. (1974). Criminal violence: National patterns and behavior. Lexington, MA:
Lexington Books.
Davies, A., Wittebrod, K., & Jackson, J. L. (1997). Predicting the criminal antecedents of a
stranger rapist from his offence behavior. Science and Justice, 37(3), 161-170.
Gillies, H. (1976). Homicide in the west of Scotland. British Journal of Psychiatry, 128, 105-
127.
Home Office Criminal Statistics England and Wales. (1977-1995). The government statistical
service. London: Her Majesty’s Stationary Office.
Huesmann, L. R., & Eron, L. D. (1989). Individual differences and the trait of aggression. Euro-
pean Journal of Personality, 3, 95-106.
Kratcoski, P. C. (1987). Families who kill. Marriage and Family Review, 12(1/2), 47-70.
Lingoes, J. C. (1973). The Guttman Lingoes nonmetric program series. Ann Arbor, Michigan:
Mathesis.
Plutchik, R., & Conte, H. R. (1997). Circumplex models of personality and emotions. Washing-
ton, DC: American Psychological Association.
Pokorny, A. (1965). A comparison of homicides in two cities. Journal of Criminal Law, Crimi-
nology and Police Science, 59, 449-508.
Ressler, R. K., Burgess, A. W., & Douglas, J. E. (1988). Sexual homicide: Patterns and motives.
Lexington, MA: Lexington Books.
Salfati, C. G. (1998). Homicide: A behavioural analysis of crime scene actions and associated
offender characteristics. Unpublished doctoral dissertation, University of Liverpool, United
Kingdom.
Salfati, C. G. (2000). Profiling homicide: A multidimensional approach. Homicide Studies, 4(3),
265-293.
Salfati, C. G., & Canter, D. V. (1999). Differentiating stranger murders: Profiling offender char-
acteristics from behavioral styles. Behavioral Sciences and the Law, 17, 391-406.
Salfati, C. G., & Haratsis, E. (2001). Greek homicide: A behavioral examination of offender
crime-scene actions. Homicide Studies, 5(4), 335-362.
Santilla, P., Canter, D., Elfgren, T., & Häkkänen, H. (2001). The structure of crime scene actions
in Finnish homicides. Homicide Studies, 5(4), 363-387.
Silverman, R. A., & Mukherjee, S. K. (1987). Intimate homicide: An analysis of violent social
relationships. Behavioral Sciences and The Law, 5(1), 37-47.
Voss, H. L., & Hepburn, J. (1968). Patterns in criminal homicide in Chicago. Journal of Criminal
Law, Criminology, and Police Science, 59, 499-506.
C. Gabrielle Salfati is a lecturer in the Centre for Investigative Psychology at the Univer-
sity of Liverpool in the United Kingdom, and the course director for the M.Sc. in forensic
behavioral science. Her main areas of interest are homicide and violent sexual crimes, in
particular with reference to offender profiling, classifications of crime, and cross-
cultural comparisons. She has presented widely both nationally and internationally on
homicide crime scene pattern analysis and offender profiling. For more information, see
www.i-psy.com.