Файл: [D._Kim_Rossmo]_Geographic_Profiling(BookFi.org).pdf

© 2000 by CRC Press LLC

(RISS) Programs now exist across the U.S. to provide multi-jurisdictional intel-
ligence and investigative analysis networks (U.S. Department of Justice, 1998).

Sex offences are much more common than murder, and when these

programs began to collect data on rapes and sexual assaults the power and
usefulness of computerized crime linkage systems became readily apparent.
Pattern recognition models for linking and predicting serial arsons, based on
cluster analysis of variables related to location, time of day, day of week, and
modus operandi, have also been developed (Icove, 1981; Icove & Crisman,
1975).

The Canadian Violent Crime Linkage Analysis System (ViCLAS), devel-

oped by the RCMP, went national in 1995 (Johnson, 1994). Since then,
ViCLAS has achieved international recognition and has now been adopted
for use by Austria, The Netherlands, Australia, Belgium, Germany, and Brit-
ain, along with certain American states including Indiana, New Jersey, and
Tennessee (Collins et al., 1998; Kocsis & Davies, 1997). Several other countries
are considering its introduction. Significant new versions of both VICAP and
ViCLAS were introduced in 1999.

ViCLAS uses a relational database within a Windows environment. Ter-

minals across the country communicate over a national wide-area network
(WAN) to a central database in Ottawa, Ontario. Data transmission is
encrypted on dedicated lines. There is also continual case tracking and audit-
ing. In Ontario, ViCLAS reporting is mandatory as the result of one of the
recommendations of the Campbell report, a judicial review into the investi-
gation of the Paul Bernardo murder case. As of mid-1999, there are over
70,000 cases on ViCLAS.

A major difficulty with computerized linkage systems is low reporting

rates. This causes a serious data shortage problem because case matches
are an exponential function of reporting level.

22

 For example, if only 50%

of the crimes are entered, then only 25% of the potential linkages can be
identified; with a 20% reporting rate, the linkage ability drops to 4%. It is
paradoxical that some police officers, who typically possess little faith in
offender rehabilitation, fail to see the value in a comprehensive criminal
tracking system.

One of the common reasons quoted for failure to report cases is the time

it takes an investigator to complete the input forms, which can involve over
150 questions. Therefore, the choice of behaviours used in the analysis is
critical. Both discrimination and utility are important. Certain offender

22 

Generally, if 

x

 represents the reporting rate, and 

y

 the linkage rate (expressed as propor-

tions of their potential totals), then:  

y

 = 

x

1/2

. This equation is an approximation because

the exact relationship depends on the mean number of crimes in a series, and on how a
series connection is defined (e.g., if 3 crimes out of a possible 10 are connected, does this
qualify as a successful series identification?).

© 2000 by CRC Press LLC

actions are so common (e.g., vaginal intercourse) as to be poor discrimina-
tors. Other behaviours are so rare (e.g., the offender writes on the victim’s
body) that they are unlikely to be encountered.

23

 Crime linkage is a holistic

process and the questions must also provide a full understanding of events.
System designers need to balance the requirement for parsimony with the
necessity for a comprehensive assessment of crime scene behaviour.

An underlying assumption of linkage systems is that the analyzed behav-

iours are more or less consistent across offences. An examination of child
abduction murders in the U.S. showed that victim gender and use of bindings
were the most important and consistent linkage variables (Hanfland, Keppel,
& Weis, 1997). Offender speech forms have been found consistent and useful
in connecting rape cases (Dale & Davies, 1994a), and the London Metropol-
itan Police Sexual Assault Index analyses verbal themes as part of the process
of linking crimes (Copson, 1993). However, FBI research on 108 serial rapists
determined that 58% of the 119 behavioural variables they examined
reflected zero or minimal consistency across crimes (Warren et al., 1995).
Alone, such questions make a poor basis for a case linkage system; together,
they may point to common underlying fantasies or behavioural themes.

The true strength of these systems therefore lies in their ability to make

concurrent comparisons between multiple variables. A study of British serial
rapists found that 28% of the offenders took steps to prevent their face from
being seen by the victim (Davies & Dale, 1995b). In the FBI research, it was
noted that 70% of the attacks took place indoors, and 60% within the victim’s
home (Warren et al., 1995). These are not powerful discriminators separately,
but assuming independence between questions, they can be combined in
powerful ways. A burglary rapist who covers his face might only represent
17% of the offender population. Additional variables can help increase the
focus. To facilitate this analytic process, the latest version of ViCLAS produces
multivariate statistics upon demand.

The process of recognition — whether it be of faces, scenes, automobiles,

voices, architecture — is based less on individual elements than on the
relationships between them. Humans are much better than serial-processing
computers at image and complex pattern recognition (Marshall & Zohar,
1997). As of today, most computerized case linkage systems are only designed
to manage and search through large volumes of information, leaving the
ultimate determination of case association to the analyst. As the volume of
information collected by these databases increases, the need for expert system
support becomes more crucial.

23 

A question can be reversed and phrased in the negative (e.g., “Was the victim trans-

ported?” becomes “Did the crime take place at a single location?”); an inversion makes a
common behaviour uncommon.

© 2000 by CRC Press LLC

Grubin, Kelly, and Ayis (1996) caution that because of “the softer and

more fluid substance of behavioural ‘evidence’ … [the process of linking
crimes] must be based on scientific principles rather than on a combination
of intuition, experience and theory” (pp. 12, 20). They summarize the
practical problems in using offender behaviour to link crimes as follows:
(1) consistent but common behaviours; (2) consistent but uncommon
behaviours; (3) preciseness of behaviour description; (4) behaviour influ-
enced by victim response; (5) weighting and relative importance of behav-
iours; (6) variation in behaviour consistency; (7) evolution of behaviour;
and (8) interpretation of behaviours.

Grubin et al. (1996, 1997) examined how offender behaviours at crime

scenes group and interact in an effort to link serious stranger sexual offences.
Four domains — control, sex, escape, and style — were used to identify 30
dichotomous variables for cluster analysis. The research assessed consistency
within a domain over a series of offences (vertical consistency), and the degree
to which similar domain patterns occur within an offence (horizontal con-
sistency). Behavioural consistency across offences was found, lending support
to the efficacy of this approach. This method of looking for the emergence
of constant patterns is similar to the process of DNA matching where genetic
alleles are compared.

The concept that linked crimes should be interconnected provides one

simple method for pattern determination. In other words, if crime A shows
similarities to crime B, and crime B shows similarities to crime C, then crime
C should also show similarities to crime A. If this is the case, then the
probability of these crimes being connected increases.

Consider an example involving the investigation of five prostitute mur-

ders in which police want to establish which crimes are connected. The
victims are Smith, Jones, Baker, Anderson, and Williams. Analysts have deter-
mined there are seven crime scene behaviours of interest. These are:

Variable 1 — victim strangled;
Variable 2 — victim found nude;
Variable 3 — bindings used;
Variable 4 — skid row prostitute;
Variable 5 — victim’s body concealed; 

Variable 6 — victim decapitated; and
Variable 7 — victim stabbed.

Table 4.2

 shows the case variable matrix for this example. Vn refers to

the variable number, and Cn to the victim (crime) number. An X indicates
the presence of that variable in a given murder. The results from the case
variable matrix are then plotted in a case linkage chart. This diagram depicts

© 2000 by CRC Press LLC

the strength of association, in terms of the behaviour variables of interest,
between the different crimes. The proper selection and phrasing of crime
behaviours are important parts of this analysis. 

Figure 4.1

 shows the case

linkage chart for this example.

In this example, crimes 1, 2, and 5 (Smith, Jones, and Williams) show

an interconnected pattern, suggesting that they were all committed by the
same offender. This approach can be refined by weighting the importance of
the variables under consideration. Variables that are less common (e.g., vic-
tim was decapitated) establish a stronger probability of connection than those
more common (e.g., victim was strangled). This method is particularly useful
when physical evidence exists establishing definitive links between some cases
but not others. Behavioural interconnections can help determine the prob-
able extent of the crime series. Such an approach was used in Lübeck,

Table 4.2 Case Variable Matrix

Victim/Variable

V1

V2

V3

V4

V5

V6

V7

C1 (Smith)

X

X

X

X

C2 (Jones)

X

X

X

X

X

C3 (Baker)

X

X

X

C4 (Anderson)

X

X

C5 (Williams)

X

X

X

X

Figure 4.1

    Case linkage chart.

© 2000 by CRC Press LLC

Germany, to help link offences in a series of arsons that lacked comparative
physical evidence between crime scenes.

In summary, the crime linkage methods discussed in this section can be

outlined as follows:

1. Physical evidence
2. Offender description
3. Crime scene behaviour

a) Proximity in time and place
b) Modus operandi

•

Find and attack the victim

•

Commit the crime

•

Escape from the scene

c) Signature

4.2.2 Other Investigative Tactics

Egger (1990, 1998) identifies additional responses that have been used or are
being developed and experimented with by police agencies to assist in solving
cases of serial murder (see also U.S. Department of Justice, 1991b). These
investigative approaches include:

•

Task forces set up at the local level, or in cases of multi-jurisdictional
crimes, at the regional level. It is also possible to implement centralized
coordination in the absence of a formalized task force;

•

Computerized analysis systems. These major case management sys-
tems store, collate, and analyze large volumes of investigative data. For
example, Scotland Yard detectives use the British Home Office Large
Major Enquiry System (HOLMES), and FBI agents use the Rapid Start
Information Management System (Doney, 1990; Federal Bureau of
Investigation, 1996);

•

Law enforcement conferences focused on a specific series of solved or
unsolved murders (see Green & Whitmore, 1993). The RCMP have
sponsored two such conferences in Canada to examine homicides of
prostitutes and young women. In 1991, Project Eclipse assessed 25
unsolved murders in Vancouver, and in 1993, Project Kayo examined
14 such crimes in Edmonton (MacKay, 1994; see also Lowman &
Fraser, 1995);

•

Geographical pattern analysis (geographic profiling or geoforensics);

•

Clearinghouses that provide information on an ongoing basis to police
agencies involved in investigating serial murder;

•

Investigative consultant teams, composed of investigators with expe-
rience in similar cases, that provide advice and assistance to the agency