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

© 2000 by CRC Press LLC

Figure 10.5

 CGT operational performance.

Figure 10.6

 CGT model learning curve.

© 2000 by CRC Press LLC

to help determine performance, but these estimates should be considered
optimal, rather than expected, performance levels of the model.

This process establishes that a minimum of 5 crime locations are neces-

sary to produce a median hit score percentages of 10%, and 6 crime locations
for a mean hit score percentage of 10%. The mean CGT hit score percentage
(6.0%) in the SFU serial murder study was higher than that predicted from
Monte Carlo testing (mean = 3.8%, median = 3.0%) for an average of 11.6
crime sites. This degradation reflects the impact of real world complexity,
and is most likely attributable to departures from uniformity in target back-
cloth (see note 68).

10.3.3.2 Reliability

The CGT model is reliable as the process is a computerized mathematical
procedure; however, but there is an element of subjectivity in determining
which crime sites in a given case are useful predictors of offender residence.
Theory, methodology, and experience help structure this interpretation pro-
cess (Benfer et al., 1991). The following guidelines should be considered when
choosing the input locations for a geographic profile:

• Generally, there should be a minimum of five distinct locations, of the

same type, available for analysis. It is usually assumed that the offender
has not moved or been displaced during the time period of these
crimes, but if this has occurred, then more locations are required. A
geographic assessment may be appropriate in cases involving fewer
locations (see Section 10.4.2.1 below).

• Only crime locations that are accurately known should be used. For

example, encounter sites may be imprecise if they have to be inferred
from last known victim sighting. In some investigations, the locations
of certain sites may be completely unknown.

• Analyses of the crime site type with the most locations results in lower

expected CGT hit score percentages. Multiple offences in the same
immediate area should not be double counted. The degree of spatial-
temporal clustering must be assessed as crime sites too close in time
and space are probably non-independent events (see previous discus-
sion on clusters). Contagion locations should be excluded from a CGT
analysis.

• Combining different site types to increase the total number of loca-

tions available for analysis can be advantageous when the number of
crimes is minimal. But two potential problems exist with this
approach. The first is that locations may be significantly correlated;
this is particularly likely when the offender travels directly to the dump
site from the encounter site. For example, the Yorkshire Ripper picked

© 2000 by CRC Press LLC

up prostitutes, parked a short distance away, and then killed and
dumped their bodies. While not identical, the encounter and body
dump sites in this series are close to being equivalent. The Hillside
Stranglers, by contrast, originated their victim disposal trips from
Buono’s home where the murders took place. The second problem
occurs when combined crime locations produce a hunting area larger
than that found with a single site type, resulting in the possibility of
a greater search area, even though the hit score percentage is smaller.
This problem is most likely to occur if the two crime site areas are
incongruent. Clifford Olson appeared to use different mental maps
for his victim searches and body dumps. Aileen Wuornos dumped
victim’s bodies and abandoned their vehicles at different sites, the
former suggestive of where she was coming from, and the latter of
where she was going to.

• Preference should be given to the crime site type that affords the

greatest degree of choice to the offender. Site types with constrained
target backcloths tell us little about the criminal. If victim specificity
leads to spatial bias then encounter locations may not be the best
profiling option. Similarly, body dump sites in isolated areas may
reveal only general detail about an urban killer.

The limitations of the CGT model are set by the hunting typology.

Because the process assumes the offender’s anchor point lies within the
hunting area, it cannot determine the residence of poaching offenders who
commute. Experience has shown, however, that often some form of second-
ary anchor point draws a poacher into the crime area — a workplace, former
residence, or the home of a friend or relative. The results of the geoprofile
then point to the secondary site, and to the degree the offender can be
associated with this location, investigators will find the geographic profile of
value.

10.3.3.3 Utility

No matter how valid or reliable a particular technique, it will have little value
if it cannot be used within the practical context of a police investigation
(Baeza, 1999). Even a highly accurate CGT prediction (i.e., a small hit score
percentage) will not lead police investigators to an offender’s door. Certain
psychics have allegedly been successful in finding criminal offenders,

66

 miss-

ing people, and buried bodies (Lyons & Truzzi, 1991; but see Kocsis et al.,
1999). Geographic profiling cannot do this. The process is an information
management system, and to that end it must be compatible with the overall
police investigation.

© 2000 by CRC Press LLC

The scope of the search area for a given CGT hit score percentage depends

upon the size of the hunting area. With a small hunting area the geoprofile
can outline the streets or blocks where the offender most probably resides;
with a large hunting area it will indicate the cities or towns. The resolution
of an analysis is bounded by its information input. Investigative strategies
based on a geographic profile must consider both expected hit score percent-
age and search area size. Some tactics possible at the local level are not feasible
on a larger scale.

Table 10.3

 lists average population and household estimates for five urban

area types in Vancouver, British Columbia (1991 Canada Census data; 

Van-

couver local areas 1981–1991

, 1994). Figures are provided for metropolitan

district, core city, and high, medium, and low density residential neighbour-
hoods (these estimates are somewhat high in that many areas contain unin-
habited regions such as industrial sectors, parks, water, farmland, or
wilderness). While numbers vary by city and region, it can be seen that even
a search area as small as one square kilometre contains hundreds of people.
This suggests there may be limited investigative value in a map that only
delineates the most probable area of offender residence. In other words,
geographic profiling does not result in an “x” that marks the spot; instead,
it provides an optimal search strategy. It is therefore important that the profile
be properly integrated with the overall investigation.

The Paperbag Rapist was responsible for at least 79 attacks in Greater

Vancouver from 1977 to 1985 (Alston, 1994; Eastham, 1989). These covered
an area of 1873 square kilometres (723 mi

2

). After John Horace Oughton was

identified as a suspect, an undercover policewoman gained access to his
apartment and observed a pin map on the wall showing the locations of his
crimes. He was being careful not to follow a pattern. Despite efforts to avoid
this, Oughton still left behind a “fingerprint” of his mental map; in a retro-

66 

Moscow police have allegedly used a pair of psychics with the ability to sense “death’s

geography” to help them locate bodies. But a Russian clairvoyant contacted during the
Chikatilo murder investigation incorrectly told police there were two offenders, who lived
high up a mountain (Lourie, 1993). A British medium claims to have accurately predicted
the Yorkshire Ripper’s name and street address 18 months prior to the arrest of Peter
Sutcliffe (Lyons & Truzzi, 1991; see also Nicholson, 1979); her assertion cannot be sub-
stantiated.

Table 10.3    Urban Population Density

Urban Residential Area

Total Population

Private Households

Metropolitan District

580/km

2

220/km

2

Core City

4190/km

2

1770/km

2

High Density Neighbourhood

19,800/km

2

13,530/km

2

Medium Density Neighbourhood

6290/km

2

3430/km

2

Low Density Neighbourhood

2380/km

2

790/km

2

© 2000 by CRC Press LLC

spective analysis the geoprofile located his home in only 0.08% of the total
hunting area. However, this still translated into a region of 1.4 square kilo-
metres (0.6 mi

2

), or approximately 1130 households and 3400 people. Sus-

pects can be prioritized in such a situation, but resource limitations make
door-to-door canvassing, for example, unfeasible. In other words, tactics and
analysis must be harmonious.

The utility of the CGT model is demonstrated by the various geograph-

ically-based investigative strategies that this process makes possible. Some
examples of such applications include suspect prioritization by address,
directed patrol saturation tactics, and the retrieval of computerized database
information on the basis of location, licence plate registration, telephone
number, and postal or zip code. These and other examples are discussed in
the following chapter.

10.4 Geographic Profiling

Clues derived from crime site location and place can be of significant assis-
tance to law enforcement agencies in the investigation of repetitive offences.
“Locating where an offender might be is the cornerstone of any detective
work ... and provides the foundations from which discovery of the criminal’s
identity may proceed” (Canter, 1994, p. 282). The probable spatial behaviour
of the offender may be determined from information contained in crime site
locations, their geographic connections, and the characteristics and demog-
raphy of the surrounding neighbourhoods.

Criminal apprehension efforts are assisted when some idea of offender

residence area exists. This information allows police departments to focus
their investigative activities, prioritize suspects, and concentrate saturation
patrol efforts in those zones where the criminal predator is most likely to
reside. Such investigative approaches are part of geographic profiling
(Rossmo, 1992, 1993a, 1995a).

Geographic profiling is a strategic information management system

designed to support serial violent crime investigations. It is part of the police
behavioural science response, and has relationships to both linkage analysis
and psychological profiling. It meets the definition of offender profiling: “an
approach to police investigations whereby an attempt is made to deduce a
description of an unknown offender based on evaluating minute details of the
crime scene, the victim, and other available evidence” (Copson, 1995, p. 1; see
also Dale, 1996). Not all types of offenders or crimes can be geographically
profiled, but in appropriate cases the process produces practical results.

Geographic profiling is a police-provided service for law enforcement

agencies or prosecuting offices, available upon official request (MacKay,