Citation
Exploring the persistence and uniqueness of modus operandi signatures in reported commercial burglary

Material Information

Title:
Exploring the persistence and uniqueness of modus operandi signatures in reported commercial burglary
Creator:
Rogers, D. J
Publication Date:
Language:
English
Physical Description:
ix, 64 leaves : ; 28 cm

Subjects

Subjects / Keywords:
Criminal methods ( lcsh )
Burglary ( lcsh )
Burglary ( fast )
Criminal methods ( fast )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Bibliography:
Includes bibliographical references (leaves 62-64).
General Note:
School of Public Affairs
Statement of Responsibility:
by D.J. Rogers.

Record Information

Source Institution:
|University of Colorado Denver
Holding Location:
|Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
71814608 ( OCLC )
ocm71814608
Classification:
LD1193.P74 2006m R63 ( lcc )

Full Text
EXPLORING THE PERSISTENCE AND UNIQUENESS OF MODUS OPERANDI SIGNATURES IN
j
REPORTED COMMERCIAL BURGLARY
By
D.J. Rogers
B.A., University of Colorado, 1988
A Thesis Submitted to the
University of Colorado at Denver
In Partial Fulfillment
Of the Requirements for the Degree of
Master of Criminal Justice
2006


This Thesis for the Master of Criminal Justice Degree
By
DJ. Rogers
Approved by
Dr. Mary Dodge Graduate Advisor
Dr. Eric Poole
Dr. Gerald Williams
04 |i4 | Qlp
Date


Rogers, D.J. (M.C.J Candidate)
Exploring the Persistence and Uniqueness of Modus Operandi Signatures in Reported
Commercial Burglary
Thesis Directed by Dr. Mary Dodge, University of Colorado at Denver GSPA MG Program
ABSTRACT
The ability to associate reported property crimes on the basis of a modus operandi (MO)
signature is the foundation of criminal profiling. It allows for the conceptualization of a
common offender and focuses tactical policing operations for the purpose of pattern
interdiction. This process operates on an assumption, recently called into question,
concerning the salience of the descriptive variables comprising an MO signature as they are
employed for criminal case-linking analysis.
The purpose of this research is twofold: (1) to produce an objective measure of salience, as
defined by meaningful degrees of differentiation and consistency, and (2) to conduct a
comparative analysis of the relative salience of each descriptive variable as compared against
a background of non-associated, temporally-coincident events. The method employed is an
adaptation of the Tversky-Kahneman Representativeness Heuristic. The derived measures of
salience, Uniqueness and Persistence, are graphically depicted as Cartesian coordinates
enhanced with an indicator for intensity.


The analysis is based on a review of 465 reported commercial burglaries, representing
activity in a single jurisdiction over a contiguous two-year period, presented as MO signatures
composed of 11 descriptive variables describing spatial, temporal and operant elements of
the offender's conduct.
This abstract accurately represents the content of the candidate's thesis. I recommend its
publication.
Signed


DEDICATION
To Christie Rogers, my beloved wife, who is, as Odysseus observed of Penelope," .. .
a wife in whom all virtues meet, .. who has proved herself so good and wise, so faithful in
her wedded love."
To Edward Clark, my long-suffering friend who, with boundless good cheer, endured
endless hours in my company only occasionally protesting. A man of courage, dignity,
integrity and above all honor.
To Mark & LeAnn Thieman; better parents I have never known, I'm humbled that
they chose to adopt me so late in life and I strive to be a fine son.
For Alta M. Rogers, my departed mother, requiescatin pace.


ACKNOWLEDGEMENT
To Dr. Mary Dodge, who served ably and cheerfully as my advisor; her guidance,
assistance and support were invaluable and her exertions, in presenting her students
opportunities to experience the fullness of their academic possibilities, were a true blessing.
To Dr. Eric Poole, who gave generously of his time, knowledge and experience, as
both professor and committee member. My time in his tutelage was an honor, and I am
much in his debt for a precision of thought, a scope of understanding and an appreciation of
virtue of which I was not previously endowed.
To Dr. Gerald Williams, who provided me with a voice bridging the academic and
practitioner experience as a member of the committee; my thanks for his willingness to give
so readily of his time, and for the efforts he has made throughout his career to better both
communities.
To Dr. Mark Pogrebin, who challenged me to expand my appreciation of matters
qualitative and, in his capacity as professor, improved me as a thinker through the sort of
vigorous, Socratic exchange that characterizes the academic ideal.
To Mr. Joseph Gerdom, the Research & Planning Manager of the Fort Collins Police
Service; my thanks for his support in developing the materials employed in this project and


my continuing gratitude for his friendship and his example of resilience in the face of
adversity.


TABLE OF CONTENTS
Figures.................................................................ix
CHAPTER
1. INTRODUCTION...................................................10
2. LITERATURE REVIEW............................................. 14
3. METHODS........................................................26
Purpose of the Study........................................26
Scope of the Data...........................................26
Considerations Regarding Spatio-Temporal Case Linking.......28
Methodology.................................................35
4. ANALYSIS.......................................................41
5. CONCLUSIONS....................................................54
APPENDIX
A. HUMAN SUBJECTS APPROVAL........................................59
GLOSSARY.................................................................60
BIBLOGRAPHY..............................................................62
viii


LIST OF FIGURES
FIGURE
1. Land Use / Zoning Pattern Map..............................................30
2. Locations of Reported Commercial Burglaries Map............................31
3. Commercial Burglary and Land Use Map......................................32
4. Temporal Clustering of Commercial Burglary Graph...........................33
5. Aoristic Signature of Commercial Burglary..................................34
6. Salience Graph Structure Type............................................39
7. Salience Graph Target....................................................42
8. Salience Graph Business Type.............................................43
9. Signature Graph Pawn Shop Crime Pattern..................................44
10. Salience Graph Structure Type............................................45
11. Salience Graph POE Type..................................................46
12. Salience Graph MOE.......................................................47
13. Salience Graph POE Side..................................................48
14. Salience Graph POE Level.................................................49
15. Signature Graph Medical Office Crime Series..............................50
16. Signature Graph Smash Entry Series.......................................51
17. Signature Graph Church Burglary Series...................................52
18. Signature Graph Pry Entry Series.........................................53
IX


CHAPTER 1
INTRODUCTION
Canter and Youngs (2003) . draw attention to the need to establish the
salient features of the crimes being examined. So many different aspects of
a crime can be considered when attempting to formulate views about that
crime, that there is the challenge, before any scientific arguments can be
developed, of determining which of those features are the behaviorally
important ones. Important in the sense of carrying information on which
reliable empirical findings can be built. . One aspect of these salient
features that also needs to be determined as part of scientific development is
that they are consistent enough from one context, or crime, to another to
form the basis for considering those crimes and comparing them with others.
. . [A]n offender's consistency is one of the starting points for empirically
based models of investigative inference; in order to use these models
operationally it is also necessary to have some indication of how offenders
can be distinguished from each other. . Research suggests that offenders
may share many aspects of their criminal styles with most other criminals but
there will be other aspects that are more characteristic, . rare,
discriminating features that may provide a productive basis for distinguishing
between offenses and offenders (Canter & Youngs, 2003 as cited in Canter,
2004, pp. 4-5).
Tactical Crime Analysis (TCA) is the policing function tasked with reviewing,
classifying and linking, on the basis of perceived similarity in exhibited modus operandi (MO)
signatures, the variety of criminal events reported to the police. Comparative Case Analysis
(CCA) is the TCA method employed to determine the degree of fitness between the prototype
and a candidate case for potential assignment to a linked case set (LCS) defining a criminal
offense series, pattern or spree. Excepting the work of Green and his colleagues (1976), TCA
has remained largely a practitioners' field receiving only limited academic scrutiny and, even
- 10-


in those rare occasions, focused more on the enhancement of technique than the
development of theory.
This inattention to fundamentals has produced methods with an undue reliance on
assumption. A compelling example of this exists in the predicate assumptions of CCA: (1)
that there is a significant, non-mundane degree of descriptive similarity between all cases
selected for inclusion in a LCS, (2) that there is a significant degree of descriptive consistency
in the MO signature of all cases selected for inclusion in a LCS and (3) that there is a
significant, non-mundane degree of descriptive difference between the stereotypical case in a
particular LCS and the population of non-selected spatio-temporally coincident cases.
With his observations regarding salience and case linking practices in "Offender
Profiling and Investigative Psychology," Canter (2004) challenged both academics and
practitioners to reexamine the fundamental assumptions of TCA / CCA in order to improve
the empirical efficacy of methods. His concerns, foundational in his more detailed
examination of investigatory inference, centered upon: (1) the descriptive salience of the MO
Signature and (2) the consistency of its appearance in exhibited offender conduct (Canter,
2004, pp.4-5). The centrality of Canter's critique obligates the academic and practitioner
communities to better establish, via deductively-reasoned theory or inductively-established
research, the reliability of their practices.
- 11 -


Canter's (2004) critique holds where the subject is property crime, but not where the
analysis turns to questions of violence. A considerable literature exists regarding MO-based
case linking in instances of serial violent crime, specifically crimes of a psycho-sexual nature
(e.g., Santtila et al., 2005; Sjostedt et al., 2004). Unfortunately, the applicability of those
findings to the behavior of a property criminal is tenuous. Given the radical separation
between the property criminal and the violent criminal, particularly as related to offense
motivation and operational conduct, it seems unlikely that the factors that drive MO
persistence in serial violence would be consistent with, or even analogous to, those
motivating the serial property offender. This perspective is similar, in form, to Chambliss'
(1967) observations concerning instrumental and expressive criminal acts. Regarding
prescriptive deterrence, as influenced by levels of subjective commitment and the nature of
the act, Chambliss noted that low-commitment instrumental offenses, like a typical
commercial burglary, were far more amenable to deterrence or more interdictable in the
context of tactical policing operations than were high-commitment expressive offenses
(e.g., sexual assault). Given that presumption, it is reasonable to assume that the same
factors that affect offender susceptibility to deterrence would inform signature MO behavior,
specifically in expressive conduct. Put concisely, the MO Signature of a serial violent offender
exists to manifest elements of his unique psychological disorder (Canter, 2004), whereas the
MO Signature of a serial property offender is more likely a product of technique, opportunity,
target selection preferences and iterative experience. Thus, while portions of the scholarship
concerning serial violence have relevance in this research as analogues, the emphasis here
will be on property offense-specific theory and research.
- 12 -


Canter's (2004) essay suggested a necessary corrective, an inducement to reflection
on the part of a community historically inclined to accept convenient assumptions in the
place of properly reasoned theory and rigorously constructed method. Unfortunately, as
Canter (2004) also noted, there is little substantive research upon which to build (p.5). Thus,
the initial efforts in this reexamination are, of necessity, exploratory. The research presented
herein is just such an effort, directed at establishing an objective basis for the salience of the
assumptions of differentiation (Uniqueness) and consistency (Persistence) in the MO
signatures of commercial burglars.
- 13 -


CHAPTER 2
LITERATURE REVIEW
The case-linking literature addressing MO profiling of property offenders begins with
the 1976 article by Green and his colleagues entitled "Cluster Analysis of Burglary M/Os" in
the Journal of Police Science and Administration. This effort focused principally on improving
case management methodology. Under the assumption that investigative effectiveness was
enhanced by systems promoting consistent assignment of similar cases, the authors sought a
statistically valid approach to identifying meaningfully similar burglary reports:
House burglaries are perpetrated by persons who have preferred targets and
specific M/Os ... The characteristics and conditions of the residence
burglaries reflected the criminals' behavior with a high degree of reliability. If
these characteristics can systematically be identified, it should be possible to
employ techniques of statistical analysis that objectify the basis of the
educated guess, hunch, intuition or whatever else the experienced officer
calls his discriminatory abilities. Such a statistical analysis will not provide
better information than good judgment, but it can systematically process
more information than any single police officer can possibly handle. Cases
are reported daily that may never come to the attention of the particular
officer who could readily recognize the M/O.... Lacking the big picture, each
investigator is handicapped by being unable to see the emerging outlines of
the characteristics of the criminal's styles (Green et al., 1976, p.383).
The authors employed an MO signature composed of seven primary elements (i.e.,
Location of Entry, Side of Entry, Location on Block, Method of Opening, Day of Week, Value
of Property, Type of Material Taken), with eight sub-elements in the "Type of Material Taken"
category, and then simulated 38 burglaries to provide a test data set. These simulated events
- 14-


were subject to pair-wise comparison, to provide a measure of similarity, which was then
employed in a non-metric, multi-dimensional scaling routine to generate event clusters that
suggested significant case groupings. While the results were considered encouraging and a
suggestion was made about the potential application of the technique to pattern/series
identification, little related research followed on the subject until 1999.
Hagen and Brown (2002) undertook the question of case matching, in the context of
developing enhanced data association applications, by refining the Green et al. approach to
measuring event similarity. The authors produced a Total Similarity Measure for a given
attribute (k), derived from Anderberg's (1973) work on cluster analysis, and developed a
series of sensitizing expressions (i.e., Binary Categorical Similarities, Transformed Categorical
Similarities, Dynamically Adjusted Weights) to enhance the quality and accuracy of weighting
variables. This approach was then tested, and compared to the findings of subject area
experts, on a sample data set of 39 cases 24 actual events from the Richmond Police
Department and 15 simulated cases. The findings of the assessment indicated excellent
fitness of association, as confirmed by the subject area experts, and an enormous increase in
volume efficiency in the case matching process. The expressions developed in this project
were incorporated in a software-application development project, the Regional Crime Analysis
Program (ReCAP). Later Lin and Brown (2002) re-examined the question of event linking,
with improved allowances for the imperfections of law enforcement case information, and
presented an outlier-based approach to data association. The basis of the outlier association
method was defined as the "Japanese Sword" convention:
- 15 -


Consider the weapon used in a series of robberies. When the weapon used is
a "gun" we can hardly assert that these incidents result from the same
person because "gun" is very common and everybody uses "guns." However,
if we have two incidents with an unusual weapon like a "Japanese sword,"
we are more confident to link these two incidents. We generalize the
Japanese sword claim as follows: when a group of records have some
common characteristics and these characteristics are outliers, these records
are more likely to result from the same criminal cause (criminal in crime
analysis) (Lin & Brown, 2002, p.4).
From this assumption a data association method, employing an Outlier Score
Function developed by the authors (Lin & Brown, 2002), was deployed against 1,198 robbery
cases defined by a six-descriptive element MO signature. In testing the outlier-based linking
methodology with the similarity-based approach proposed in Hagen and Brown (2002), Lin
and Brown (2002, p.12) found that the performance of the former was superior to that of the
latter and deemed the technique promising as an enhancement of police investigative
methods.
Yokota and Watanabe (2002) further refined the case-linking methodology by
introducing confounding concepts of differentiation and consistency in MO signatures to the
case-linking process. The authors employed a Choice Probability linking algorithm, sensitized
to possible variation in aspects of the MO signature, to a sample data set of 107,233
burglaries, with 12,468 identified offenders, reported between 1993 and 1998. This work
represented an interesting departure in the literature. While previous efforts focused on case-
linking for investigatory assignment (Green et al., 1976) or for pattern/series identification at
the case entry point (Hagen & Brown, 2002; Lin & Brown, 2002), Yokota and Watanabe
- 16-


(2002) phrase the problem from the perspective of linking a known offender's signature to a
collection of recently reported criminal events. By attempting to match the known signature
of an offender to a reported event the authors needed to demonstrate that such behavior is
sufficiently particular to be attributable to a specific offender and would display sufficient
stability over a meaningful portion of the offender's deviant career. The evidence for this
proposition was mixed (Yokota & Watanabe, 2002, pp.6-7), but adequate to serve as a basis
for further research. In running the linking algorithm against the data set, the authors
determined that the approach demonstrated a "high potential to link crimes to possible
offenders" (Yokota & Watanabe, 2002, p.13).
Coincident with the work of Yokota and Watanabe (2002), Bennell and Canter (2002)
employed logistic regression techniques to identify linked case pairs and receiver operating
characteristic (ROC) analysis to assess the predictive reliability of various signature elements
to the data association question. Their objective was to enhance the quality of the primary
linking process:
Comparative case analysis (CCA) ... has the goal of demonstrating that the
same offender has committed two or more crimes. . [Absent] a
confession, eyewitness testimony or other forensic evidence . behavioral
information must be relied upon to link crimes and the task usually involves
an examination of what happened at the crime scenes and where the crimes
took place. These aspects of the criminal event are popularly regarded as the
offender's modus operandi (MO) and they have been the subject of limited
empirical study.
MO is a rather vague term used in various ways by different police officers
and different crime fiction writers. The use of the concept assumes that
there will typically be a high degree of similarity between what an offender
- 17 -


does in one crime and what he or she does in another. CCA also assumes
that police officers are able to recognize these similarities and use them to
make effective investigative decisions. Yet research has shown that linking
decisions are often based on the limited, subjective impressions of
investigating officers, that these impressions often differ from officer to
officer, and that investigators often perform poorly on tasks like CCA.
There is, therefore, value in determining precisely which aspects of the
offenders' crime scene actions are most often repeated across crimes
(Bennell & Canter, 2002, p.154).
The approach developed by Bennell and Canter (2003) was tested against sample
data, created by selecting two benchmark cases for each of 43 serial commercial burglars
known to have operated in a major United Kingdom metropolis between January 1999 and
January 2000. As with Yokota and Watanabe (2002), this research was oriented from the
known offender signature and generalized to the body of reported events. The findings of the
sample analysis indicated that spatial proximity was the most "consistent and stable" case-
linking element, with "operant behaviors" (descriptive MO elements) being more "context
dependant" (Bennell & Canter, 2002, p.162), and thus a less reliable linkage factor. In 2005,
noting that MO case-linking methods were still applied only "sparingly and with extreme
caution" (Bennell & Jones, 2005, p.24), one of the authors undertook a second study to
replicate and extend the 2002 research. In this new project the data set was somewhat
expanded, to include both residential and commercial burglaries and involve alternative
spatial and temporal contexts. The findings were largely consistent with the original research.
Spatial proximity between events was the linking factor of choice and descriptive MO
elements were found to be applicable for linking only with "great difficulty" (Bennell & Jones,
2005, p.37).
- 18-


Oatley and Ewart (2003) evaluated a variety of statistical approaches, primarily as a
means of predicting potential or repeat burglary locations but with an eye towards
pattern identification, and concluded by recommending an approach based in Bayesian
probability analysis. This material was incorporated in a later work specifically concerned with
evaluating differentially constructed burglary MO signatures and the associated case
matching algorithms (Ewart et al., 2005). The MO signatures reviewed included one limited
to descriptive MO elements only (RCPA), the second signature included only temporal and
geographic elements (RPAL) and the final signature considered was a combination of the
RCPA and RPAL signatures (COMBIN) (Ewart et al., 2005). Interestingly, and in conflict with
the findings from the Bennell studies, the Ewart team concluded that the RCPA signature
algorithm, when evaluated against a data set of 966 residential burglaries involving 306
identified offenders, was more effective than the RPAL method:
For maximum operational performance, crime matching using a combination
of MO and temporal and geographic data may have to be a two-stage
process with initial matching being made on the basis of MO and refined
using the temporal and geographic features (Ewart et al., 2005, p.170).
Similar to the work following Yokota and Watanabe (2002), this research is offender-
centric, focused on matching cases to known offender MO signatures rather than suggesting
associations and indicating relative strength of linkage between recently reported criminal
events.
- 19-


Oatley, Zeleznikow and Ewart's (2005, p.4) unpublished research was, functionally, a
restatement of previous material, where they noted that the field is still characterized by how
little . has been achieved in the ability of 'soft' forensic evidence (e.g., the burglar's
modus operandi) to provide the basis of crime linking and matching." The distinction here
was in their conclusion that none of the reviewed methods possessed predictive reliability;
they have confidence in the potential of survival analysis, naive Bayes and Bayesian belief
networks, but acknowledge that further substantiating research would be necessary (Oatley
et al., 2005).
Adderly and Musgrove (2003), contemporaries of Oatley et al. and similarity
interested in the application of data association techniques to the problem of linking criminal
events to offender profiles, evaluated techniques specific to the practice of data mining (e.g.,
Multi-Layer Perception, Radial Basis Function, Kohonen Self-Organizing Maps). The authors
applied those techniques to a data set of 23,382 residential and commercial burglaries
reported to the West Midlands (United Kingdom) police services between January 1997 and
February 2001 (Adderly & Musgrove, 2003, p.269). Their findings were more consistent with
the Oatley et al./Ewart et al. research, thus conflicting with the Bennell results, in suggesting
the relative efficacy of the composite MO signature for case-linking.
Canter's (2004) observations on investigative psychology, which provided the
framing reference for this thesis, were intended as a "review of the field" assessment, not a
work forwarding a particular theoretical, analytical or evaluative objective. Canter (2004) did
-20-


supply a limited commentary on various linking theories and methodologies, but the material
was presented largely in passing.
The research of Meaney (2004) was taxonomical in nature, focusing on the
classification methods employed, and demographics observed, in assigning serial offenders
(e.g., burglars, rapists, arsonists) to a category of either "commuter" or "marauder." While
seemingly tangential to the subject at hand, Meaney was confronted with a similar research
problem concerning linking events in series on the basis of modus operandi factors. From the
MO perspective her findings addressed the question of persistence; the data indicated that
operational area, measured in terms of distance traveled to commit the offense, tended to
expand over time in a radial manner from the anchor location (Meaney, 2004). This
demonstrated tendency of serial burglars to expand operational area, to the effect that they
were relatively more likely to be classified as "commuters" (Meaney, 2004, p.128), raised
questions about the strength, reliability and consistency of the spatial clustering effect. The
research also described the basis for an additional dichotomy in serial burglars, between the
opportunistic and the professional offender. These burglars, depending on classification and
experience, exhibited a differential sensitivity to situational motivations and rational target
selection behaviors, although neither class was considered to be a product of "individual
psychopathology" (Meaney, 2004, pp. 132-133). This finding suggested that MO modification
over the series, due either to situational opportunities or learning, should be expected, thus
diminishing the reliability of the MO persistence assumption.
-21 -


The Santtila et al. (2004) research on offender characteristic prediction from
observed crime scene behavior was largely concordant with Meaney's (2004) research, as
applied from a different perceptive axis. Working from assumptions of MO persistence and
"bounded rationality" (Santtila et al., 2004, p.138), the authors sought to demonstrate that a
variety of demographic and descriptive elements would be significantly related to certain MO
signatures. This theory was tested against a data set of 633 burglaries, committed by 244
identified offenders, in a Finnish metropolitan area between 1990 and 2001. In matching
cases to offenders (with the understanding that multiple offenders may have been involved
in a single event, and vise-versa) a final data set of 913 events, with a known relationship
between a particular event and a particular offender, was developed. Logistic regression
analysis was employed to produce a component score for events and offenders and a
complex set of significant descriptive factors were identified for each of 14 types of reported
burglary (e.g., Basic, Opportunistic, Suburban, Balcony). While there was noteworthy
variation in areas, the authors generally concluded:
In spite of . potential threats to generalisability, the present study
highlights that it is feasible to predict certain offender characteristics on the
basis of crime scene behavior in urban burglary. Thus, theoretical
assumptions underlying psychological profiling may be as valid for property
crime as they are for violent crime (Santtila et al., 2004, p.147).
These results lent support to assumptions that were challenged in the findings of
Meaney (2004); thus, the matter of MO persistence remains an open question, but a
reasonable assumption upon which to predicate further analysis. The literature is mixed as
well on the question of MO uniqueness; from the perspective of what factors lent most
-22-


reliability to such differentiation rather than on the existence of uniqueness. Of greatest
interest, in light of the present work, is the fact that none of the literature addresses, in any
meaningful form, the question of salience as framed by Canter (2004).
As mentioned in the introduction, there is an extensive literature on the subject of
MO persistence and uniqueness in the domain of serial violent predation. This literature was
reviewed, in a limited respect, to evaluate opportunities to generalize from findings or to
reason analogously where that practice would not strain credulity too greatly.
In 2002 Santtila and his colleagues, building on Brantingham and Brantingham's
(1981) foundational work regarding the spatial analysis of crime, undertook an evaluation of
the predictive utility of geographic profiling. The test data set was composed of 50 sexual
assaults reported in northern Italy between 1973 and 1996, all the work of a single offender.
The purpose of the geographic profile, extrapolated from the spatial locations of the reported
assaults, was to identify areas with the highest probability of being the offender's anchor
point (i.e., residence, work place, primary recreation locale). In this variation on offender-
centric research, focused on the determinative value of spatial elements in the MO signature,
the offender's home-range would be identified through analysis of his conduct, just as the
offender himself, through a linkage of his known offense signature to the observed
signatures of reported crimes, was thought to be identifiable in the work of Yokota (2002)
and Adderly (2003). The findings suggested that the predictive value of this method was
limited and extremely sensitive to the taxonomical classification of the offender and the
-23 -


selection of certain analytical variables, in particular the distance-decay function (Santtila et
al., 2002, p.51). This observed weakness in spatial significance related to geographic profiling
may have implications regarding the relative importance of geographic attributes generally,
but the particularities of the reviewed methodology could explain the difference.
Santtila and colleagues (2005) revisited the subject, focused specifically on questions
of MO persistence and uniqueness, when employing techniques of multi-dimensional scaling
(MDS) and discriminate function analysis (DFA) to link cases based on behavioral signatures.
Their research findings suggested that the crime of rape was "hardly haphazard" and was
characterized by a consistent thematic structure which contributed to both persistence and
uniqueness in the behavioral signature (Santtila et al., 2005, p.102). This finding lent
credence to the line of investigation pursued in the present thesis, though the reliability of
such a generalization from a crime of sexual violence to a property offense is certainly
open to examination.
In a contemporaneous work, also addressed to the question of stability of MO
characteristics in sexual offenders, Sjostedt and his colleagues (2003) evaluated data for a
Swedish offender cohort of 1,303 former prison inmates who had committed sex crimes
between 1993 and 1997 and were currently on release. This analysis employed Cohen's
Kappa, to calculate inter-rater reliability, and odds-ratios to measure stability in a reduced
variable MO signature (Sjostedt et al., 2003). The authors concluded that there was a
significant degree of stability, as defined, in the behavioral MO of an offender over the span
-24-


of his known criminal career; this stability was particularly evident in the Victim Preference
element (Sjostedt et al., 2003).
A comparison of the literature examining MO Signature linkages for psycho-sexual
offenses the contrast, in terms of the specificity of analysis, with the work in the property
crime field is revealing. The psycho-sexual literature is more explicit in addressing the
foundational weakness commented upon by Canter; moreover, the specificity of the work to
the problems of persistence and uniqueness exceeds anything to be found in the property
crime literature. This is unfortunate, as the reliability of a generalization from one offense
domain to another is not well established and explanations for why such analogous
reasoning may be unreliable abound. Yet, the material at least provides some support for the
line of inquiry and suggests a variety of methods by which the issue may be framed and
examined, though consistent with Canter's (2004) critique it touches not at all on the
subject of salience.
-25 -


CHAPTER 3
METHODS
Purpose of the Study
This research proposes a response to Canter's (2004) critique, intending to produce
exploratory measures of the presence and extent of MO Persistence and MO Uniqueness -
factors deemed adequate proxies for salience (Canter, 2004) in a sample set of reported
commercial burglaries. This effort will compare descriptive MO Signature data from four
linked-cases sets (LCS) known to be the products of a common-offender(s) with coincident
data, some of which are convenience grouped into Crime Patterns or Crime Sprees.
Scope of the Data
The data set, composed of 465 commercial burglary event records, represents the
totality of commercial burglary reported to the Fort Collins Police Department in the period
between December 2002 and March 2005. This date parameter was defined so as to include
the entire known temporal activity range of the four identified common-offender LCS in the
data. The data were drawn from official police incident reports, as enhanced by a post-report
CCA process conducted by the agency's tactical crime analyst.
-26-


For each reported commercial burglary there is an event signature record composed
of 11 descriptive variables: Spatial X/Y Coordinates (State Plane), Temporal X/Y Coordinates
(168 Hour Week), Target, Business Type, Structure Type, Point of Entry (POE) Type, Method
of Entry (MOE), POE Side, and POE Level. This variable set, with the exception of the spatial
and temporal elements, has been recoded from a textually descriptive value to a numeric
value to simplify categorical and analytical processes.
Event records in this data set were also taxonomically classified, in the post-report
CCA process, for inclusion if the standards were satisfied in a three-tiered classification
schema; Crime Series, Crime Pattern or Crime Spree (See Glossary). This review identified
165 event records that were amenable to classification in one of 19 LCS; four Crime Series
(LCS 1, 2, 4 and 18), five Crime Patterns (LCS 3, 7, 10, 12 and 13) and ten Crime Sprees
(LCS 5, 6, 8, 11, 14, 15, 16, 17, 19 and 20). There were 300 event records in the
population data that were not classifiable as a member of any identified set of linked cases.
In terms of the relevant physical geography, the city of Fort Collins is a spatially
semi-isolated college community with a municipal jurisdiction of approximately 51 square
miles and a service population of 125,000. The property of spatial semi-isolation is relevant in
this analysis, as it tends to limit the amount of impact "cross-over" offending patterns has on
the CCA process (i.e. an offender in a multi-jurisdictional metropolitan area who commits
offenses in a number of the available municipalities, resulting in disjointed reporting and
interrupted series data) and on the completeness of an identified LCS. While there was
-27-


occasional cross-over offending, in particular in the nearby communities of Loveland and
Greeley, it occurred in a manner that had limited impact on the offense sequence or case set
completeness.
Considerations Regarding Spatio-Temporal Case Linking
Much of the work, both practitioner and academic, concerning the practical problem
of criminal case linking has focused on geo-spatial and temporal descriptive elements. It is
the contention of a majority of the existing case-linking literature that the MO Signature
variable of greatest associative reliability is that providing spatial, or spatio-temporal,
information (Bennell, 2002; Bennell, 2005). In this study, geographic and temporal elements
are considered as a part of the event signature, but their significance for case-linking is
minimized as compared with that of other descriptive elements. The reason for this reduction
in significance, in spite of research findings to the contrary, is the concern that spatial and
temporal clustering may prove to be artifacts, arising from the extant opportunity structure of
commercial burglary as influenced by land use patterns and operational considerations. In
furtherance of this "clustering-as-artifact" assumption, two analytical reviews of the data
were conducted; one, an application of the Nearest Neighbor Index (NNI) analysis described
by Eck and his colleagues (2005, pp. 17-19) as a test of spatial clustering, and the other, an
adaptation of Ratcliffe's Aoristic Temporal Analysis Method (2004, pp.8-10) for identification
of temporal clustering.
-28-


The results of the comparative analysis of NNI values indicated that, while known
common offender linked-case sets have the lowest NNI value (0.6807) of any of the
typologically classified series, this outcome is partially a result of the differences in sample
sizes, and remains 41% higher than the aggregate population NNI score. In reviewing the
NNI differential, comparing a linked-case set population with a comparably sized random
sample of unclassified cases as suggested by Eck (2005, pp.17-19), the Crime Series LCS
data had the smallest differential (0.1), indicating a higher likelihood of clustering, and the
Spree LCS data had the largest (0.29). From this result it is reasonable to infer that the
spatial clustering effect observed here for LCS known to have a common offender is actually
smaller than that observed for the convenience-linked Crime Pattern or Crime Spree data.
Considering that all three of the LCS typed classes had higher NNI values, thus lower
clustering effect, than the aggregate population the assumption of spatial proximity as an
indicator of offender commonality is untenable for this particular analysis.
A review of the cartographic presentations of the case data provides a visceral -
although anecdotal argument for considering the possibility that spatial clustering, in this
data at least, is an artifact of land-use. Figure 1 illustrates land-use patterns in Fort Collins
during the time frame in question. Blue and green areas are zoned for commercial or light
industrial uses, all other areas represented in grey are zoned for either residential or
some other non-commercial use (i.e. university, parks / open space, etc.).
-29-


Figure 1: Land Use / Zoning Patterns
Figure 2 depicts the location, by address (with no intensity indicator), of the reported
commercial burglaries in the population data set.
-30-


Figure 2: Locations of Reported Commercial Burglaries
In Figure 3 the burglary events overlay the land-use map; the coincidence between
commercial burglary locations and commercially zoned areas is almost perfect. This
seemingly obvious outcome is not always accounted for in clustering analysis in particular
in the practitioner's venue which frequently assumes the entire jurisdictional land surface to
be in equal jeopardy for victimization and thus discovers an unusual degree of spatial
clustering in the incidence of commercial burglary.
-31 -


Admittedly, this is not a perfect analysis of the employment of spatial data as a case
linking factor there are criticisms of the NNI method and issues unique to this particular
study population. Yet the findings present a reasonable basis, when taken in the context of
the larger study, to question the utility of the spatial data as a case linking element.
Figure 3: Commercial Burglary & Land Use
As for the temporal signature elements, an Aoristic analysis of both the aggregate
population data and the separate taxonomically-classified case sets supported a finding
-32-


minimizing the applicability of observed temporal clustering (Figure 4) as a case linking
factor.
Figure 4: Temporal Clustering of Commercial Burglary
The temporal signature which emerged (Figure 5) was suggestive of the expected
inversion of typical hours of business operation, thus illustrating the principal cause of
nighttime-centric event clustering in reported commercial burglary.
-33 -


Figure 5: Aoristic Signature of Commercial Burglary
A review of the Crime Series LCS data was undertaken, intended to identify specific
day/time preferences (e.g. Saturday: 02:00 to 03:00) that might have differed significantly
from the relatively even distribution in the aggregate population data. Only one of the LCS
groups (LCS 1) exhibited a notable day preference, with 46% of reported events taking place
on Wednesday (sharply contrasting the 14% exhibited in the aggregate population data),
though not with sufficient sequential persistence to apply as a linking factor (although not
without value as a targeting enhancer suggesting that deployments intended to interdict
this particular crime series should consider the offender's preference for Wednesdays). The
temporal window for the cases also lacked adequate specificity. In the aggregate population
data the average event window, describing the amount of time between when the victim
location was last known to be secure and the time the burglary was discovered, was 18
hours. This event-range was reduced only slightly, to approximately 15 hours, when just
-34-


those cases known to be the product of a common offender(s) were considered. In
evaluating the spatial and temporal elements as linking factors, in either their own right or
combined, there is a uncomfortable degree of uncertainty, in particular as regards the central
question of salience, which ultimately mitigates against employing these elements in
subsequent analysis.
Methodology
In their 1974 work "Judgment Under Uncertainty: Heuristics and Biases" Tversky and
Kahneman described heuristic reasoning as being "quite useful" (1974, p.1124) for
comparative case classification based on a subjective probability assessment of similarity.
They explained:
In the representativeness heuristic . probability ... is assessed by the
degree to which [the case] is representative of, or similar to, the stereotype
[case]. Indeed, research with problems of this type has shown that people
order [cases] by probability and by similarity in exactly the same way
(Tversky & Kahneman, 1974, p.1124).
The CCA process for criminal event linkage is typically heuristic. The analyst reads
cases for descriptive content with a stereotypical case, arising from either an amalgam or an
exemplary event, as the prototype. For each case a subjective determination of the degree of
relative similarity between prototype and candidate, based on a review of a limited collection
of assumed-to-be-salient descriptors, is generated; this estimation serves as the basis for the
case-linking decision.
-35 -


The concern with this approach, from both the academic and practitioner
perspective, is its inherent susceptibility to error. Tversky and Kahneman, while
acknowledging the functional value of heuristic reasoning, produced a catalogue of
potentially "severe and systemic errors" (1974, p.1124) arising from applications that were
not adequately sensitized to a variety of complications. These complications, which are the
primary source of bias in heuristic reasoning, include failures to consider base-rate frequency,
sample size, probability (in particular in sequence), predictability, validity, and the effects of
regression to the mean (Tversky and Kahneman, 1974, pp.1124-1127). Recognition of these
difficulties initially limited the consideration of a heuristic approach as the proper method for
this analysis.
Various alternatives for developing defensible measures of uniqueness and
persistence were considered, with fitness to the data, methodological reliability, explanatory
value and parsimony as the decisive factors. A number of approaches were evaluated, with
particular attention paid to scaling and clustering techniques, but ultimately the selection
criteria recommended a heuristic method designed to control for statistical complications and
to impart a degree of objective reliability to what is an innately subjective methodology. This
approach necessitated the calculation of the following metrics:
General Descriptive Values
-36-


The proposed adaptation requires a collection of base measures be in place upon
which the indices of differentiation and consistency will rely. These values, calculated
primarily for the aggregate population data, include:
Population Ratio (PR) = Count of the Variable Value in the Population / Total Number of Cases in the Population
Series Ratio (SR) = Count of the LCS Modal Variable Value / Total Number of Cases in the LCS
Maximum Sequential Run (MSR) = Count of Longest Continuous Sequence of the Modal Value in the LCS
Maximum Run Probability (MRP) = PR MSR
Run Persistence (RP) = MSR / Total Number of Cases in the LCS
Uniqueness Score
The first indicator metric determined was the Uniqueness Score which, in deference
to the critique of Tversky and Kahneman, was designed to respect the character of base-rate
frequency in the population data. The Uniqueness Score was calculated as follows:
Uniqueness Score (US) = SR / PR
This produces an index value, anchored on 1 (the point where the frequency in any
given LCS is equal to the frequency in the aggregate population), that indicates the degree of
atypicality possessed by the value of any given variable.
Persistence Score
-37-


The second indicator metric derived was the Persistence Score, addressing Tversky
and Kahneman's (1974, p.1125) concern that sequential probability was misestimated in case
matching heuristics. The method for calculating this value is:
Persistence Score (PS) = RP MRP
The Persistence Score produces an intensity value which ranges between 0 and 1. At
0 the metric indicates limited persistence, values approaching 1 indicate significant, or
absolute, relative persistence. This metric is intended to address the "Misperception of
Chance" problem, noted by Tversky and Kahneman (1974, p.1125), by discounting
persistence values that are the product of high-order of probability events.
The Uniqueness and Persistence Scores are then mapped as Cartesian coordinates,
producing a graphic depiction of the degree of distinction of either an individual variable or,
in the concluding portion of the analysis, a composite MO Signature. An example of such a
graphic, hereafter referred to as a Salience Graph, is provided in Figure 6. For the signature
variable Structure Type, Figure 6 presents the LCS average Uniqueness Score (X Axis) and
Persistence Score (Y Axis) for each classified typological set.
-38-



1

c J FWtVTt
Q i s


m a
UiQun
Figure 6: Salience Graph Structure Type
The purple marker in the lower left quarter of Figure 6 represents the Uniqueness
(1.000) and Persistence (-0.006) Scores for the modal value of the element Structure Type in
the population data. The relative size of the marker indicates the representativeness of the
class, in terms of the average number of matching cases contained by a particular LCS. In
the population data the value used is 1, as befits a grouping context in which all cases are
considered individually.
The average values of the four LCS known to be the product of a common
offender(s) are represented by the blue Crime Series marker. The relative position and size of
this marker indicates that this LCS is characterized by a greater degree of MO Uniqueness
(Uniqueness Score = 6.275), a stronger MO Persistence (Persistence Score = 0.488) and a
greater measure of representativeness (22.75 cases) than the population data. The red and
yellow markers represent the convenience-linked classifications of Crime Sprees and Patterns,
-39-


respectively. They are provided as points of reference, as is the average value for the three
classification typologies, which is represented by the green marker.
The optimal result of this analysis would be to isolate descriptive variables in the MO
Signature that have relatively high Uniqueness and Persistence Scores, either across all
classification categories or for a specific taxonomical class. A variable with relatively high
Persistence Scores and relatively low Uniqueness Scores would be, for purpose of the
analysis, mundane; as the persistence of the value is most likely a product of its prevalence.
Likewise, a variable presenting relatively high Uniqueness Scores and relatively low
Persistence Scores would be considered unreliable for sequential case matching, as the value
- while noticeably discordant with coincident activity possesses too little consistently to
support reliable event linking.
-40-


CHAPTER 4
ANALYSIS
Canter (2004) has suggested that a case-linking methodology, in advance of
developing scientific arguments, should positively determine, "which . features [of the
descriptive signature] are . behaviorally important," (Canter, 2004, p.4). The adaptation of
the Representativeness Heuristic, with its explicit measures of persistence and uniqueness, is
employed here in response. The measure of the descriptive salience of indicia imbued in a
typical MO Signature (e.g., Green, 1974; Gottlieb, 1994; Bruce, 2004) was developed in two
stages. An initial, variable-level assessment was conducted, as depicted in the Salience
Graphs, to discern the relative intra-variable descriptive validity. The variables were then
considered as a composite Crime Series signature, presented as Signature Graphs, which
should illustrate if the determinative salience discerned at the variable level persists when the
description is considered as a whole.
For purposes of illustrative clarity, the Salience Graphs presented in this section
reflect typological averages, as opposed to the unique scores of each of the 19 LCS. Where
such an average possesses a known weakness, in terms of its fitness as a measure of central
tendency, or produces other representative complications, such matters will be addressed in
the associated narrative. The majority of the analytical discussion will focus on the Crime
Series data, due to its unique ability to describe the behavior of a single offender entity over
-41 -


a series of offenses. Population, Crime Pattern and Crime Spree averages are provided as
reference points, lending context to the series data.
The first variable reviewed was the element Target (Figure 7), indicating the primary
type of property taken during the event, which was found to be descriptively mundane. The
modal value (i.e., Cash) was ubiquitous in offender conduct, as substantiated by a
Uniqueness Score of 1.220, a condition that likely accounts for the element's relatively high
Persistence Score (0.685). The Target was prosaic, as was the nature of its consistency.
VaWfeg*



Amo* fl |
(sj
r
a

Figure 7: Salience Graph Target
The Target metrics also provided a foundation for the argument that any analysis of
MO Signature data must consider the effect of taxonomical classification. In this case Crime
Spree data are observed to have an appreciable effect on the overall average, a result of an
extreme Uniqueness Score (19.862) paired with a strong Persistence Score (0.591). In an
analysis where the data selection method was based on matching a limited number of cases
-42-


to apprehended offenders it would be quite possible to over-sample Spree-type events, thus
producing atypical measures of salience.
The Business Type variable (Figure 8), which described the sort of commercial
activity characteristic of the victim location, exhibited the greatest degree of differentiation
(Uniqueness Score 9.819) in the analysis, but lacked the level of persistence necessary for
unquestionable salience. The consistency of the Crime Series data, as measured by the
Persistence metric, was limited because LCS 18 lacked a modal value and LCS 2 had
insufficient data for estimation.


g JHh*
mm o m
1

jpg
-5 th£ '3-1 lillr m m a x
L/iquerwe
Figure 8: Salience Graph Business Type
This element also presents a case-study of a known weakness in the existing
TCA/CCA case-linking methods. Namely, the over-representativeness generated by
tautological comparative methods. This is a defect, addressed in Tversky and Kahneman,
typical of stereotype-based heuristics. The example, in Figure 9 LCS 13 (a Crime Pattern)
-43 -


contained eight cases for which the modal Business Type value was a Pawn Shop. This value,
which occurred in only 2% of the population data, was reported in every LCS 13 member
event. Employing the proposed Representativeness Heuristic resulted in a Uniqueness Score
of 45.455 and a perfect Persistence Score (1.000). No other descriptive variable in LCS 13
exhibited a similar measure of salience, leading to speculation that linkages in this particular
case were the product of a convenient tautology. The stereotypical event involved a Pawn
Shop. Given the compelling salience of that indicator, prospective and retrospective CCA
would assign events involving Pawn Shops to LCS 13 (a decision that may also have been
influenced by spatial or temporal coincidences).
Figure 9: Signature Graph Pawn Shop Crime Pattern
This unitary-indicator method, while strengthening the salience of a single descriptor,
could also likely produce a diminished salience for the remaining signature variables. This is
the case in LCS 13; no other descriptive variable in the signature produces a Uniqueness
Score above 6 and the majority of the scores are in the sub-2 range. The possible source of
-44-


this disparity is an intermixing of offenders. If multiple offender entities targeted pawn shops
in a similar spatio-temporal frame, the unitary-indicator approach could produce results
similar to those observed in LCS 13. The visible spread arising from the consistency in target
selection combined with diversity in operational conduct characteristic of independent actors.
The variable Structure Type (Figure 10), describing the victimized building, is often
regarded as a complementary value to Business Type, given that specific types of commercial
activity are properly accommodated only by certain facility types. With a Uniqueness Score of
6.275 and a Persistence Score of 0.488 Structure Type exhibited the same sort of qualified
salience as Business Type, without the confounding distortion attributable to vagrities in the
data.
Of the elements considered in this analysis Business Type and Structure Type were
the two which exhibited the greatest salience in the relevant taxonomical classification, the
-45 -


Crime Series. Business Type, after considering the complications mentioned above, is likely
the variable of greatest descriptive reliability: a finding of greater inferential import if
eventually confirmed in an analysis of a more robust data set.
In POE Type (Figure 11), a variable describing the sort of entry point selected, the
analysis isolated another relatively mundane variable. The offender tended to select
commonplace entry points (Uniqueness Score 1.862), to which he was not highly committed
(Persistence Score 0.328). A score of this sort, while salient in comparison to the population
averages, was suggestive particularly when taken in combination (e.g., POE Type, POE
Level, POE Side, MOE) of an offender who is more dynamic and situationally malleable than
is assumed in some of the case linking literature (Green, 1974, p.382).

MM-,: ' i
<
T f: ! 1 %,
il .% :v.
1
m m m no

Figure 11: Salience Graph POE Type
Method of Entry (MOE), describing the forcible means employed to gain access,
possessed the strongest salience (Uniqueness Score 3.172, Persistence Score 0.584) of the
-46-


Operational Variables (Figure 12). MOE is also thought to be the signature element of
greatest indicative reliability concerning an offender's experience level. Under that
assumption a youthful or inexperienced offender would be characterized by MOE technique
that is unrefined, blunt and often unsuccessful. The experienced offender's technique would
exhibit a greater degree of subtlety employed to more frequent, and prosperous, effect.
Figure 12: Salience Graph Method of Entry
Of interest here was the degree of inconsistency between sequential MOE behavior
exhibited in the Crime Series LCS and the idealized offender of the TCA/CCA process. MOE
variation over the life-cycle of a crime series is assumed to be the product of technique
optimization through iterative experience. Thus, an offender initiates a crime series exhibiting
a consistent MOE eventually iterative experience induces experimental variation and
ultimately the offender either accepts the improved MOE, or rejects it and returns to the
original cycle. This is convenient for TCA/CCA, as it provides a level of sequential consistency
necessary for prospective pattern interdiction. Unfortunately, it is inconsistent with the
-47-


observed behavior of actual offenders. Two LCS, 1 and 4, described the activity of a common
offender entity over a significant period of time (685 days and 542 days, respectively). The
exhibited variation in MOE in those LCS was more haphazard, and potentially situational,
than an orderly and progressive career development arc might suppose.
As for the final two descriptive elements, POE Side (Figure 13) and POE Level (Figure
14), the analysis suggested they were quotidian. On the question of Side selection, the data
described a fairly even distribution of activity between a limited number of options.
Where the question concerned the entry Level, offenders overwhelmingly selected
the ground floor of a targeted structure. Given the community in question, this may not
indicate conscious selection as much as a dearth of options; the availability of multi-story
commercial structures is limited by comparison to alternative jurisdictions. This sort of
distribution offers some insight as to why the naive heuristic TCA/CCA approach occasionally
-48 -


succeeds. A "second-story man" operating in this environment would identify himself
relatively quickly, given that the primary strength of heuristic differentiation is exception
identification in variables with predominate-value response histories.
With varying degrees of salience in isolation identified the analysis then sought to
assess the quality of those distinctions in the MO Signature context. Business Type and
Structure Type, with MOE considered to a lesser degree, had been identified as relatively
salient descriptive elements under Canter's (2004) rubric. Would this quality adhere when the
analysis turned to a consideration of the actual conduct of known offenders?
The first series (Figure 15) considered contained 52 reported burglaries to medical
offices committed between December 2002 and November 2004. LCS 1 was an exemplar for
the approach taken in this thesis. Business Type, Structure Type and MOE presented
themselves as the variables of primary salience in the anticipated order and degree.
-49-


LCS1 Sgnabre


A-
w A
£ i
Ana POETyps
1 eas i V
Figure 15: Signature Graph Medical Office Crime Series
The second comparative analysis examined a series of 8 burglaries, linked largely on
their exhibited consistency in entry method, which took place between October 2003 and
January 2004 (Figure 16). Again, the descriptive variables exhibiting the greatest actual
salience were those distinguished by the variable-level review. With the theoretical ordering
slightly inverted in LCS 2, MOE exhibited the most salient reliability. Business Type remained
the most distinct variable (while lacking reliability), while Structure Type was distinct within a
cluster of variables otherwise difficult to differentiate.
-50-


Figure 16: Signature Graph Smash Entry Series
A series of 27 church burglaries, committed between March 2003 and September
2004, was the next LCS assessed, even though the outcome in this particular grouping was
eminently predictable (Figure 17). Churches, as target structures and entities, occurred with
a frequency of only 6% in the aggregate population data. They represented 100% of the
targeted Business and Structure Types in LCS 4, insuring that these variables would possess
considerable salience. MOE, to some degree lost in the cluster of values in the lower ranges
of the graph, still presents a reasonable degree of distinctiveness, though lacking a desirable
level of persistence.
-51 -


LCS4 Sfpwbre
ml

A
w
ET*
pc
pc EL** : > l 1 1 z>
Figure 17: Signature Graph Church Burglary Series
The final signature assessment considered a Crime Series (LCS 18) which exhibited
little salience and lacked even meaningful spatial or temporal patterns outside the realm of
the descriptive MO elements (Figure 18). This series represents a classic "failure scenario" for
the TCA/CCA approach: where the offenders exhibited consistency, it was of the mundane
sort (e.g., Target = Cash) and there was no consistency in the salient variables. An example,
regarding the element Business Type, this offender group had not yet victimized the same
sort of business twice prior to their apprehension. While it would be a mistake to over-
extrapolate from this LCS, given the relatively small number of cases involved (4) and the
limited temporal duration (26 Days), it does suggest the existence of actors who, by intention
or accident, adverse to identification via such heuristics.
-52-


LCS18 37&n

POE Type
A- POE 1da
StoJchreType


*
Figure 18: Signature Graph Pry Entry Series
Canter (2004) posed the question of salience. Is there a method of classifying a
criminal event, for purposes of comparative analysis, that can be shown to be descriptively
efficacious? To provide an answer, a method was proposed, an assessment was conducted,
and a comparison to actual offender data completed. On the basis of this exploratory
analysis, the Representativeness Heuristic seems to offer an objective means for the
evaluation of salience.
-53 -


CHAPTER 5
CONCLUSIONS
In the decade since Blau was writing it has become clear that developing a
scientifically derived investigative tool may be more difficult than the early
writings by the FBI Special Agents had implied (Mokros & Alison, 2002). But,
nonetheless, their promotion of the potential links between the action
associated with a crime and the characteristics of the person who committed
that offense have drawn an increasing number of serious scientists to
elaborate the central psychological questions that are implicit within the
concept of'profiling' and associated activities (Canter, 2004, p.4).
In laying the foundation for what Canter (2004) described as an empirically-based
model of investigative inference, the salience of the MO Signature needed to be convincingly
demonstrated. The difficulty or drudgery of this task, combined with the attractiveness of
accepting the assumption of salience and moving on to considerations of comparative
technique, has resulted in Canter's (2004) critique going largely unanswered.
This thesis presents a qualified response, based on the findings of an exploratory
analysis, to the effect that all of the signature elements possess salience, and some possess
it to a degree that analysts and academics may consider reliable. On average, the LCS modal
variable values tended to occur with a frequency 3.6 times greater than observed in the
population data, and they tended to appear in sequential runs accounting for 42% of their
overall presence. By Canter's (2004) standards such elements could be differentiated, as they
-54-


were distinct from coincident background activity, and they presented an opportunity for
sequential occurrence sufficient to meet the needs of pattern identification and interdiction.
The Representativeness Heuristic has provided a means, objective in descriptive
quality and sensitized to the various statistical problems typically associated with such
approaches, to examine the quality that is salience. As for future questions concerning
application and utility, this is an area where additional efforts should prove fruitful. Such is
the nature of exploratory analysis; it occasionally presents more complications than it
resolves, but it often provides promising premises upon which to proceed. A work returning
to first principles, addressing pedestrian concerns such as salience, should serve as an
inducement for a thoroughgoing reexamination of the practice and theory undergirding
TCA/CCA. A reconstructed methodology would prove to be a boon for the practice of crime
analysis and justify the pain that Canter (2004) notes will be the unavoidable consequence of
building that scientific tool.
In the practitioner realm these exploratory findings present an opportunity to
reexamine a range of practices and infrastructure in regards their ability to support
intelligence-led policing. The examination should commence with the event data upon which
the entire edifice rests. The holistic complication is the question of data completeness arising
from known underreporting of criminal activity. A range of factors, from issues of
convenience to concerns about law enforcement impartiality and efficacy, influence the
decision to report a crime. From the analytical perspective this creates a practical reality; the
-55-


data set employed for the TCA/CCA process will always be missing cases. Law enforcement,
if truly interested in the promise of TCA, should undertake an introspective review directed at
isolating the causal factors of underreporting and resolve them to the degree possible.
A secondary problem with data quality arises from the reporting process itself. The
capture of complete, timely, accurate and salient event signatures is predicated on a
reporting process that is onerous and is concerned primarily with documenting events for
statutory and procedural compliance. This is not to suggest that such intent is improper, for
one of the primary functions of law enforcement is documenting the nature of a criminal
event for prosecution. A second, important requirement of such a document is to insure that
form, policy and practice are adhered to in the process of policing. Developing data for
purposes of problem identification, pattern analysis and interdiction is a process responsibility
at a far remove from the core agency priorities. As the practice of policing changes, and
force-multiplying technologies and methods are better integrated, the reporting and
information gathering processes will need to be reengineered to improve analytical fitness.
This will, however, largely be a function of evolving management philosophy rather that one
driven by theory or technology.
Given a data collection and management environment optimized for TCA support,
and properly grounded and reasoned theory and method, the expected gains to the law
enforcement services should prove worth the investment. Of greatest import would be the
enhancement of the TCA function. An agency better capable of properly identifying and
-56-


interdicting emerging crime patterns in its jurisdiction is an agency delivering a manifest
social good. The quality of safety, of security in one's self and one's possessions is the true
product one expects of the police. An integrated TCA function, identifying the actual activity
of highly prolific offenders, is capable of delivering that good efficiently and effectively. This
refined TCA/CCA function should also improve internal services, in terms of both case
management and clearance, to at least the degree initially envisioned by Green and
associates (1976). TCA/CCA has already demonstrated some success in focusing
investigations and expanding the awareness-space of detectives regarding the totality of
activity related to known offenders. Enhancements that are directed at reducing the false-
positive and false-negative effects known to infect existing methods could only prove
beneficial to the quality of service delivered to the citizen.
The systems impact of ideas considered here would come in the area of adaptive
pattern-matching application logic. The TCA/CCA process has been the subject of an
impressive range of research and development in the last decade, ranging from traditional
statistical techniques to some exotic neural computing applications, but there has been little
reflection on basic, theoretical issues that have limited the impact of these efforts.
Application development tempered by theory and research, in particular on the foundational
issues of salience and variable interaction, would likely produce a toolset that could
simultaneously enhance the quality of the analyst and expand the variety and volume of
crimes that analyst could review. Currently an analyst is capable of managing a monthly case
load of between 200 and 500 cases, depending on the quality of the analyst and the
- 57 -


information infrastructure supporting the function. An application properly conditioned to the
inherent characteristics of criminal conduct could support an expansion of volume of an order
of magnitude, identifying the cases most likely to benefit from human review while
dramatically increasing the overall awareness-space of the analyst.
Canter (2004) and others doing yeoman work in this area have provided the
direction, and criticism, necessary to support a thoroughgoing reengineering of the TCA/CCA
process. This thesis is itself a response to one of those criticisms; an exercise that has
provided, at least, some indication of the nature of salience in a traditional MO signature.
Perhaps, given past interest in problem-solving and the emerging support for concepts such
as intelligence-led policing, these ideas will find a practitioner community better prepared to
address their long-held assumptions and explore the possibility of policing as a data-driven
endeavor.
-58-


APPENDIX A
HUMAN SUBJECTS APPROVAL
University of Colorado at Denver and Health Sciences Center
Human Subjects Research Committee Institutional Review Board
Downtown Denver
Campus Box 120, P O Box 173364
Denver. Colorado 80217-3364
Phone; 303-656-4060. Fax 303 5565855
DATE: December 23. 2005
TO: David Rogers
FROM: Dorothy Yates, HSRC Chair
SUBJECT: Human Subjects Research Protocor2006-054 Exploring the Persistence and
Uniqueness of Modus Operand! Signatures in Reported Commercial Burglary* *

Your protocol has been approved as exempt under CFR Title 45 Part 46.101,b. This
approval is good for up to one year from this date.
Your responsibilities as a researcher include:
If you make changes to your research protocol or design you should contact the
HSRC so that we can determine if your exempt status continues.
You are responsible for maintaining all documentation of consent. Unless
specified differently in your protocol, all data and consents should be
maintained for three years.
If you should encounter adverse human subjects issues, please contact us
immediately.
If your research continues beyond one year from the above dale, contact the
HSRC for an extension.
The HSRC may audit your documents at any time.
Thank you for submitting your protocol and good hick with your research.
Campuses Downtown Deovnr Htzsimons Aurora N -59-


GLOSSARY
A discussion of this subject is complicated by a specialist practitioner and academic
language. Employed to promote clarity, brevity and an economy of language, it may well
generate confusion for the uninitiated reader. Thus, in order that the reader may properly
discern the distinctions in some of the concepts employed throughout this thesis, the
following collection of definitions is presented:
Behavioral Signature: an exhibited pattern of offense-related behavior integral to the
offender's psychological motivation.
Comparative Case Analysis: a method of comparing and linking reported crime events, on a
qualitative/descriptive basis, to identify a pattern or series of activity by a presumed common
offender/offender group. The set of linked cases is employed to develop proactive
interdiction operations and clear cases in the event of an arrest/interdiction.
Crime Pattern: temporally non-contiguous, though often exhibiting temporal clustering, but
displaying a degree of spatial, target and MO similarity sufficient to suggest a common
offender/offender group.
Crime Series: temporally non-contiguous, though likely exhibiting a notable degree of
temporal clustering, and displaying a high degree of target, MO and signature similarity, with
spatial clustering frequently a product of the interplay of these factors, producing a high-
degree of certainty of a common offender/offender group. Cases linked, in addition to the
presence of MO signature similarity, by "hard" forensic evidence (i.e. fingerprints, hair/fiber,
confession).
Crime Spree: temporally contiguous, generally occurring as a continuous criminal episode,
and frequently (though not necessarily) exhibiting spatial, target and MO similarities.
Geographic Profiling: a methodology, largely based on concepts of environmental criminology
as formulated by Brantingham and Brantingham (2005, pp.27-54), that is used to identify
high-probability anchor locations (i.e., home, workplace, recreational sites) for a serial
offender based on observed geographic factors in his reported crimes.
-60-


Linked Case Set: a grouping of individual event records based upon either a known
commonality in responsible offender(s), which is the most reliable method of event linking, or
by convenience factors as described under the headings of Crime Pattern or Crime Spree.
Method of Entry (MOE1: the means by which entry is effected through the point of entry;
frequently characterized by methods like prying, breaking, defeating locking devices or
discovery of unsecured POEs.
MO Signature: a specific collection of elements describing target selection, methods and
techniques employed by the offender in furtherance of his strategic objective. Generally
collected via purpose-developed MO reports contained within a law enforcement agency's
standard field event documentation (e.g., a police incident report).
Operational Variables: variable elements of the MO Signature which describe the method and
point of entry to a victim structure. In the MO Signature employed in this analysis POE Type,
MOE, POE Side and POE Level comprise the set of Operational Variables.
Point of Entry (POEj: the type of portal selected, or created, to gain entry to a targeted
structure; typically selected points of entry include exterior doors, windows, and vents or
access panels.
Salience Graph: a Cartesian coordinate graph, with the Uniqueness Score indicated on the X-
axis and the Persistence Score indicated on the Y-axis. The relative size of the symbol
suggests the number of cases, in specific or on average, underlying the typological grouping.
Signature Graph: a Cartesian coordinate graph, similar in general construction to the Salience
Graph, but presenting the salience indicia for all of the descriptive variables of a particular
LCS, instead of scores typologically grouped for a specific descriptive variable. The relative
size of the symbol is indicative of the number of cases in the set that matched the modal
value represented by the salience metrics.
Target Selection Variables: variable elements of the MO Signature comprised of the Target,
Business Type and Structure Type which speak to the offender's conceptualization of what he
is seeking and where he feels it is likely located.
-61 -


BIBLOGRAPHY
Adderley, R. & Musgrove, P. (2003). MO Modeling of Group Offending: A Data Mining Case
Study. International Journal of Police Science and Management, 5(4), 267-276.
Anderberg, M. (1973). Cluster Analysis for Applications. New York, New York: Academic
Press.
Bennell, C. & Canter, D. (2002). Linking Commercial Burglaries by Modus Operandi: Tests
Using Regression and ROC Analysis. Science and Justice, 42(3), 153-164.
Bennell, C. & Jones, N. (2005). Between a ROC and a Hard Place: A Method for Linking Serial
Burglaries by Modus Operandi. Journal of Investigative Psychology and Offender Profiting, 2
(1), 23-41.
Brantingham, P. & Brantingham, P. (1981). Environmental Criminology. Prospect Heights,
Illinois: Waveland Press.
Bruce, C., Hick, S. & Cooper, J. (Eds.) (2004). Exploring Crime Analysis: Readings On
Essential Skills. Overland Park: IACA.
Canter, D. (2004). Offender Profiling and Investigative Psychology. Journal of Investigative
Psychology and Offender Profiling, 7(1), 1-15.
Canter, D. & Youngs, D. (2003). Beyond 'Offender Profiling': The Need for an Investigative
Psychology. In D. Carson & R. Bull (Eds.), Handbook of Psychology in Legal Contexts (2nd
Edition, pp. 171-206). Chichester: Wiley.
Chambliss, W. (1967). Types of Deviance and the Effectiveness of Legal Sanctions. Wisconsin
Law Review, Summer, 703-719.
Eck, J., Chainey, S., Cameron, J., Leitner, M. & Wilson, R. (2005). Mapping Crime:
Understanding Hot Spots. Washington, D.C.: National Institute of Justice.
Ewart, B. W., Oatley, G. & Burn, K. (2005). Matching Crimes Using Burglars' Modus Operandi:
A Test of Three Models. International Journal of Police Science and Management, 7(3), 160-
174.
Gottlieb, S., Arenberg, S. & Singh, R.. (1994). Crime Analysis: From First Report to Final
Arrest. Montclair: Alpha Press.
-62-


M ,1976) duster Analysis of Burglary M/O's. Journal of
toUw
Hagen, S. 8. Brown, D. (2002). DataMethdS ^
Enforcement. Decision Support Systems 34, 369-378.
Levine, N. (2004). CrimeStat III Version 3,0: A Spatial f' ^ ^ *
Crime Incident Locations. Washington, D.C.: National Institute
Lin, S. & Brown, D. (2002). An Outlier-Based Data Association Unking CnmM
Incidents. Unpublished Manuscript. University of Virginia p
Engineering, Charlottesville, VA.
Meaney, R. (2004). Commuters and Marauders: An Examination of the Spatial Behavior of
Serial Criminals. Journal of Investigative Psychology and Offender Profiling, 1(2), 121-137.
Bayesianttet Precfictions. ftpert Systems %%%££$ ***
vrx ^ ^ **
Manuscript. * techniques to Soft Forensic Evidence. Unpublished
Manuscript Ze'ezn,kow. 3- & Ewart, B. (2005). Matching and Predictin r
R^.(2004,Th u
Cnme Reduato"- " "**"* for the Soar
7. s Sand *** s(1>' ** M*Ta Targeting o
sanrt, Un,
&e"e ^ Mokros. A .. 'n9' '<& f S^"9er Rape
Sar>Wla
r-> . /
vior Tnf-' 01 "Oi
Internationa/ JC
)Kros, A. (2004). Predicting Burglar
Journal of Police Science *
-, Laukkanen, M. & Picozzi,
vjf-rt hc Profiling Approach in Three Rano r-
-ensic Science International. 10*'
fpW*- '
sJsteot r
,nse*uai£:: fan
Tv,
*
,dl Justict
versky, A. & Kahneman, D.
Science. (185), 1124-1"
, iv_ozz\, M. (2003). Testing the
... hi mree Rape Series of a Single Offender; A
ional, 131 (1), 42-52.
., Sturid'-'
-oi,y tsurglar Characteristics from Crim
a or Police Science and Management, 6(3), 136-154.
--Luia, P., Zappala, A., Laukkanen M TestinQ the Util
Ge^mphic Proflling An.-- tatter; ^ ^
4>- ^
N., Sturidsson, K.
ig. Criminal Justin

n. & Grann, M. (200*
~e and Behavior, 31 (5)(
(1974'

Of

%

Jr,


Yokota, K. & Watanbe, S. (2002). Computer Based Retrieval of Suspects Using Similarity of
Modus Operandi. Internationa! Journal of Police Science and Management, 4(1), 5-15.
-64-