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

© 2000 by CRC Press LLC

tigation suffers from an incomplete pattern and the profile is missing impor-
tant data. Proper information collection — from the crime scene, during the
follow-up investigation, and as part of routine police practice — is particu-
larly important (Bekerian & Jackson, 1997); also, the value of a profile is
significantly enhanced when it is combined with other investigative tools,
both behavioural and traditional.

Profiling is not independent of other forms of investigative analysis. A

framework for crime analysis methods has been developed by Mario de Cocq
(1997), head of Interpol’s Analytical Criminal Intelligence Unit (ACIU). Ana-
lytic techniques are first classified as either strategic or operational, and then
grouped by focus on criminal incident, offender, or crime control method.
Within this framework, “specific profile analysis [unknown offender(s)]” —
psychological and geographic profiling — is categorized as opera-
tional/offender. Comparative case (linkage) analysis is categorized as opera-
tional/criminal incident. This approach allows for both similarities and
differences between analytic methods to be readily identified.

Part of the difficulty with profiling has been its probabilistic nature. Police

investigators are uncomfortable with such methods of inquiry, preferring
instead to rely upon “certainties.” This often leads to problems in the under-
standing, use, and evaluation of profiling services. The following example
clarifies the point. A man consults a statistician before gambling on a game
of dice. He is advised that the number most likely to turn up is 7, and he
bets accordingly. But when he rolls a 5, the bet is lost.

Was the advice inaccurate? Well, any student of probability knows that

7 is the most likely number to result from the roll of a pair of dice, with 2
and 12 being the least likely. Still, the probability of 7 occurring in any 

single

roll is only 1 in 6. The statistician’s advice was not a prediction — in fact, a
better forecast (with a 5/6 chance of being correct) would have been a number
other than 7. Rather, the advice was a statement regarding the most likely
single outcome. While the prediction was wrong, it was also accurate, and
any future prediction would still be 7. This information is thus of greater
value when used over the course of a series of games. In other words, while
no one can predict a given spin of the roulette wheel, there is little doubt the
house will make money at the end of the day. Similarly, profiling is optimally
employed by an investigation when it assists in the direction of repetitive
efforts and the prioritization of volume work.

5.7 Expert Testimony

Expert testimony from profilers has been introduced now in both American
and Canadian courts. Subject matter has included future dangerousness,

© 2000 by CRC Press LLC

threat level, similar fact evidence, case linkage, crime scene signature, staging,
and insanity (see Keppel, 1989, 1995; Keppel & Birnes, 1997). Profiling has
also played a role in premises liability litigation where the court must establish
if the proximate cause of damages was a breach of duty to provide proper
security (Kennedy & Homant, 1997). Opinion evidence regarding likely crim-
inal response to security measures influences establishment of a cause-in-
fact relationship between the alleged negligence and plaintiff injury. In these
circumstances, most of the profile is an attempt to predict the behaviour of
the typical offender — persistence, desistance, or displacement — under
certain security conditions. While such information may assist juries in
appropriate cases, prediction of specific individual behaviours with any
degree of confidence is difficult, even for experts (Homant, forthcoming).

This is particularly so when the offender is unknown (i.e., not identified

and apprehended). Homant (forthcoming) proposes a typology for offender
deterrence based on factors of criminal motivation and desire to avoid
capture: (1) calculating — motivated and cautious; (2) opportunistic —
unmotivated and cautious; (3) determined — motivated and incautious;
and (4) impulsive — unmotivated and incautious. Adjustments are made
based on offender intelligence, target significance, and evidence of irratio-
nality (e.g., use of drugs or alcohol). Profiling can play a role in premises
security litigation through interpreting signature, reconstructing crime
scenes, and assessing motivation.

To be so designated, an expert witness must meet three requirements:

(1) their testimony must be relevant; (2) their field must require scientific,
technical, or specialized knowledge; and (3) they must have the necessary
background to qualify as an expert in the field (e.g., skill, specialized training,
formal education, experience) (Garland & Stuckey, 2000). Relevance has been
defined as evidence that affects the probability of existence for facts of con-
sequence to the trial (Robertson & Vignaux, 1995). Until 1993, determination
of “scientific knowledge” used to follow the Frye test (

Frye v. United States

),

which asked if the knowledge was generally accepted as reliable within the
relevant scientific community. That year the U.S. Supreme Court decided
that U.S. Federal Rules of Evidence superseded Frye (

Daubert v. Merrell

,

1993). The Daubert test is now used to determine if a subject matter has
reached the stature of “scientific knowledge.” It is a more flexible and less
stringent test than Frye, and is based on the following factors: (1) falsifiability
(the testability of the technique); (2) peer review and publication; (3) the
actual or potential error rate and the maintenance of operational standards;
and (4) general acceptance of the methodology within the relevant scientific
community. These are currently the principles guiding admissibility of sci-
entific evidence in American courts.

© 2000 by CRC Press LLC

The Canadian test is not as strict as the U.S. one. In 

R. v. Mohan

 (1994),

the Supreme Court of Canada outlined the requirements for admissibility of
expert evidence: (1) relevance to a fact in issue; (2) necessity in assisting the
trier of fact; (3) absence of an exclusionary rule; and (4) a properly qualified
expert. Novel scientific theories or concepts are subject to special scrutiny
concerning their validity and reliability. This scrutiny is a flexible test that
considers acceptance within the scientific community; suggested criteria
include testability, peer review, and publication. Canadian courts have stated
that expert opinion evidence can be rendered to assist the trier’s understand-
ing of psychology, behaviour, and human conduct; but they have also cau-
tioned that the confirmation methods used in profiling “should be
considerably sharpened and disciplined” (

R. v. Clark

, 1998, p. 21). Profiles

are not generally introduced as evidence in British courts.

Profiling is based on inductive, probabilistic methods. What is the proper

role of such evidence in the courtroom? All science is inductive and scientific
“laws” are only predictions based upon repeated observations. By contrast,
deductive systems are derived from axioms or established rules. The only
true deductive system is mathematics, though many scientific theories are so
well accepted they are often treated as established facts in normal life for all
intents and purposes. It has been said that science is quantitative and law is
qualitative (Forst, 1996a). For example, “beyond a reasonable doubt,” the
standard for criminal conviction, is a subjective determination and has not
been translated by either statute or case law into a specific number.

28

 Gen-

erally, probabilistic information is valid, especially in cases involving circum-
stantial evidence, though experts are ill advised to go beyond the established
boundaries of their discipline. They must also be prepared to articulate the
logic and specify the data upon which their conclusions are based (Homant,
forthcoming).

The most appropriate manner of introducing and weighing profiling

expertise in court is through the use of Bayesian probability methods. Bayes’
rule is a logical theorem that provides the means of updating probabilities
given new information of relevance (Iversen, 1984). It can be expressed as
follows:

(prior odds)(likelihood ratio) = posterior odds.

(5.1)

The prior odds are those that existed before the new information, and

the posterior odds, those after. Probability always ranges between 0 and 1;
odds are the ratio of the probability that something is true to the probability
that it is not: 

p

 / (1 – 

p

). The likelihood ratio is the quotient of the probability

28 

In a survey of 1200 U.S. judges, two-thirds thought “beyond a reasonable doubt” repre-

sented a probability of guilt of at least 95%.

© 2000 by CRC Press LLC

of the evidence given that an assertion is true (i.e., the accused individual is
guilty), divided by the probability of the evidence given that the assertion is
false (i.e., an accused individual is innocent). This can be expressed as follows:

likelihood ratio = P(E|G)/P(E|I).

(5.2)

The probability of guilt given the evidence (P(G|E)) can be determined from
the probability of the evidence given guilt (P(E|G)). According to Bayes’ rule:

P(G|E) = P(G) P(E|G) / [P(G) P(E|G) + P(I) P(E|I)].

(5.3)

If the probabilities of guilt and innocence are equal (i.e., P(G) = P(I) =

0.5), then the above equation simplifies to: P(G|E) = P(E|G). The higher the
prior probability of guilt, the more attenuated the impact of the evidence
(likelihood ratio) on the posterior probability of guilt.

Expert testimony, however, should be limited to providing the likelihood

ratio resulting from the observation, profile, or test result. Behaviour science
evidence, whether it be similar fact, signature, psychological, or geographic
in nature, can only be justified on this basis. Estimating probability of guilt
or innocence assumes knowledge of the prior odds, and this determination
is the responsibility of the judge or jury — not the expert witness (Robertson
& Vignaux, 1995).

Errors in the use of probability within the court context unfortunately

are not uncommon.  The prosecutorís fallacy results from transposing the
conditional in either the numerator or denominator of the likelihood ratio
(see also Martin, 1992).  This occurs, for example, when the probability of
the evidence given guilt, P(E|G), is equated with the probability of guilt given
the evidence, P(G|E).  In other words, while all cows are four-legged animals,
not all four-legged animals are cows.  This type of error within a behavioural
science context results if:  (1) a crime scene indicates a certain personality
profile with a 90% level of confidence, and the assumption is made that a
person who matches the profile is 90% likely to be guilty (transposing the
conditional in the numerator); or (2) only 10 murders out of a database of
50,000 exhibit a specific crime scene behaviour, and the conclusion is reached
that the probability of such a similarity occurring by chance is one in 5,000
(transposing the conditional in the denominator, also referred to as the
coincidence fallacy). The former type of error occurred in the aftermath of
the 1996 Atlanta Olympics pipe bombing. Security guard Richard Jewell
became a major suspect in the investigation simply because he fit the FBI
profile for a certain type of bomber. Effectively branded guilty, it took several
weeks before he was eventually cleared (Reid, 1996).

© 2000 by CRC Press LLC

The defence attorney’s fallacy occurs when evidence is considered in

isolation, rather than as a totality. It is the combined impact of different, and
independent, evidential elements that is important.

29

 This type of error

within the behavioural science context results if a profile is ignored because
it lacks dramatic discriminatory power, instead of the findings considered in
conjunction with other existing evidence, and used to increase the posterior
odds appropriately.

5.8 Future of Profiling

Notwithstanding his claim to be a consulting detective, Sherlock Holmes
employed inductive methods almost as often as deductive ones (Teten, 1989).
Despite his renowned successes, such procedures by their very nature carry
a certain failure rate. Wrong predictions are part of any probabilistic-based
methodology, including profiling (Homant & Kennedy, 1998). It is important
that profiles be used by both providers and consumers in a careful and ethical
manner, with an awareness of their limitations and an understanding of the
proper application of stereotypes and prioritization methods. Grubin (1999)
also cautions that profiling is not about “getting into the mind” of the crim-
inal, and we should be concerned when the profiler becomes more newswor-
thy than the profile.

“In summary, there is enough research to suggest that crime scene pro-

filing may have sufficient reliability and validity to be useful for some pur-
poses. The literature suggests that the concept of behavioral traits and
consistency across situations is respectable, if measured in broad contexts”
(Homant & Kennedy, 1998, p. 338). The development of profiling and other
forensic behavioural science techniques is in its early days. While this may
sometimes result in frustration, it is also an exciting time with much potential
for future evolvement. For example, offender profiling may benefit signifi-
cantly from the application of fuzzy logic (Kosko & Isaka, 1993; Yager &
Zadeh, 1994). Traditional Boolean logic is dichotomous and follows the law
of the excluded middle — answers are either yes or no. Fuzzy logic allows
for subjectivity and “maybes.” These shades of meaning can be given inter-
mediate values between 0 and 1. Verma (1997) discusses how imprecise
features and characteristics may be fuzzy variables. A sufficient number of
such parameters can form a fuzzy prototype pattern class (e.g., young, tall,
heavy, violent). This is similar to the concept of frames, flexible and fuzzy
enclosures used to prioritize suspects during an investigation (Kind, 1987b,
1990). Estimating min-max values can help define the limits of the pattern

29 

These errors in logic can also occur within a profile. Use of spurious, intervening, and

non-independent variables is problematic and can distort an analysis.