Behavioral Consistency Among Serial Burglars

Introduction

In the absence of the three most effective means of identifying offenders—forensic evidence, eyewitnesses, and offender confessions—suspect prioritization through the use of behavioral information present at a crime scene becomes a valuable tool to help police identify or narrow down the range of potential suspects in unsolved cases (Deslauriers-Varin & Beauregard, 2013). While the practice of offender profiling involves the more general use of behavioral and psychological crime scene information for suspect prioritization in law enforcement investigations (see, for example, Douglas, Ressler, Burgess, & Hartman, 1986), the sub-field of crime linkage analysis (CLA) refers to the process of linking a known offender to one or more unsolved crimes based on high levels of consistency and distinctiveness in crime scene behaviors across a series of offenses (Deslauriers-Varin & Beauregard, 2013). Unlike other practices in offender profiling, which tend to take a more clinical approach to evaluating offenders and their crime scene behavior (Copson, Badcock, Boon, & Britton, 1997), CLA relies on the statistical analysis of police data to uncover patterns in offending, and links an unsolved crime to another behaviorally similar cases in which the responsible offender is usually already known (Woodhams, Hollin, & Bull, 2007).

Although some variation has been found in the type and level of behavioral consistency shown by different types of offenders, and when using different analytical methods, research on CLA has been fairly successful in identifying solved and unsolved cases committed by the same offender. Specifically, behavioral consistency of serial offenders has been successfully conducted for burglary (e.g., Bennell & Canter, 2002; Bennell & Jones, 2005; Goodwill & Alison, 2006; Green, Booth, & Biderman, 1976; Yokota & Canter, 2004), arson (e.g., Santtila, Fritzon, & Tamelander, 2005), sexual assault (e.g., Canter et al., 1991; Deslauriers-Varin & Beauregard, 2013; Grubin, Kelly, & Ayis, 1997; Grubin, Kelly, & Brunsdon, 2001; Knight, Warren, Reboussin, & Soley, 1998), robbery (e.g., Woodhams & Toye, 2007), and homicide (e.g., Melnyk, Bennell, Gauthier, & Gauthier, 2011; Salfati & Bateman, 2005). Within these offense types, the most consistent offending behaviors have been the distance an offender traveled to the crime, the time interval between offenses, and a combination of related behaviors committed at the crime scene, known as the modus operandi or offense style ¹ (e.g., Markson, Woodhams, & Bond, 2010; Tonkin, Woodhams, Bull, Bond, & Palmer, 2011).

Although results on the behavioral consistency of serial offenders are encouraging for the field of CLA, there are several issues in behavioral consistency and crime linkage research that still need to be addressed. For instance, while past research has shown that certain offenders (within a crime type ² ) are highly consistent in their crime scene behaviors, other studies suggest that some offenders are highly inconsistent (Deslauriers-Varin & Beauregard, 2013; Lussier, Leclerc, Healey, & Proulx, 2008). It therefore remains unclear if offenders that commit a specific style of offense are more consistent than offenders who commit different styles of the same offense, and how characteristics of consistent and inconsistent offenders may differ?

Disparities are also evident among studies using the same data, but different analytical methods, to assess offender behaviors and consistency (see, for example, Bennell & Jones, 2005; Deslauriers-Varin & Beauregard, 2013; Melnyk et al., 2011; Tonkin, Santtila, & Bull, 2011; Woodhams, 2008; Woodhams, Grant, & Price, 2007). This has led to some ambiguity in the interpretation of findings from studies using different methods, which are still debated in terms of their level of accuracy and efficiency, and ultimately what may be the “best” method of case linkage and behavioral consistency analysis.

This study aims to address these issues by conducting a comparative analysis of the consistency in offending behaviors and the results of analyses used in CLA research. To do this, three behavioral consistency analyses will be employed to evaluate the consistency of “offending styles” in a sample of serial burglars from the United States. The results of each analysis will then be compared to evaluate the similarities, differences, and relative utility of the methods. Next, any variation found in the burglary offending styles will be evaluated to determine the level of consistency or versatility among serial offenders, depending on the style of crime that they commit. Through this research, we aim to shed new light on the consistency of crime styles among serial burglars, and how case linkage practices may be improved by taking behavioral and methodological issues into consideration before employing CLA in real-world police investigations.

Offending Consistency and Behavior

Criminal behavior, much like non-criminal behavior such as one’s taste in music or choice of hobbies, is believed to be a reflection of many stable traits including the individual’s personality, physiology, developmental features, and demographic factors (Caspi & Bem, 1990; Horning, Salfati, & Crawford, 2010). While psychological research has shown that situational influences may play a role in altering the behavior of an individual in some circumstances (see, for example, J. R. Meyer, 1990; Mischel, 1999), personality traits have also been significantly correlated ³ with behavioral outcomes such as marital success, educational and occupational attainment, health, psychiatric disorders, criminal behavior, and life expectancy, above and beyond the effect of influential factors such as socio-economic status and cognitive abilities (Caspi, Roberts, & Shiner, 2005; Ozer & Benet-Martinez, 2006; Roberts, Kuncel, Shiner, Caspi, & Goldberg, 2007). For instance, longitudinal studies have shown that individuals who are more conscientious and agreeable tend to stay married, while those who score high on neuroticism are more likely to experience divorce, even when controlling for factors such as socio-economic status (Kinnunen & Pulkkinen, 2003; Roberts & Bogg, 2004). Similarly, children exhibiting poor self-control and irritability are at heightened risk of unemployment decades later in adulthood (Caspi, Wright, Moffitt, & Silva, 1998), while childhood levels of neuroticism, extraversion, agreeableness, and conscientiousness predicts occupational success more than 45 years later, even after controlling for childhood IQ (Judge, Higgins, Thoresen, & Barrick, 1999). Children’s personality traits at age three have also been shown to significantly predict behavioral problems in teenage years, psychiatric disorders, quality of interpersonal relationships, and criminal behavior at age 21, and unemployment in adulthood (Caspi, 2000). In other words, while certain situational factors could influence behavior in the short term, these findings suggest that personality traits can predict behavior, beyond other influential factors, up to several decades in the future.

For the purposes of CLA and behavioral consistency research, it is important to note that as many longitudinal predictors of behavior, such as personality, race, gender, socio-economic status, genetics, and developmental factors, change slowly, if at all, behavior should similarly remain relatively predictable and stable over time. In short, there is reason to believe that the way serial offenders commit their crimes may be relatively consistent across offenses.

While behavioral consistency research has mainly been conducted by psychologists, and more recently by those studying offender profiling, there is striking similarity in the themes and methodologies used in each of these fields and those of mainstream criminology (Deslauriers-Varin & Beauregard, 2013). For instance, one of the main goals of developmental and life-course criminology (DLC) is to assess the longitudinal sequence of crimes committed by individuals, and assess the extent to which offenders are specialized or versatile in their criminal career (Blumstein, Cohen, Roth, & Visher, 1986; Nieuwbeerta, Blokland, Piquero, & Sweeten, 2011).

Whereas CLA evaluates the distinctiveness and consistency (or lack thereof) in the behaviors that criminals exhibit in their offenses, DLC assesses the consistency in the general types of crimes that offenders commit (and ultimately, why they persist and desist). In other words, DLC focuses on specialization in offense types, while CLA focuses on specialization in offense styles and behaviors. ⁴ This similarity in research means that CLA could benefit from drawing on the scientific methods and findings on offending specialization in DLC, as they are highly relevant to the young field of CLA (Deslauriers-Varin & Beauregard, 2013).

Among the most relevant findings for CLA from DLC research is the fact that offenders tend not to specialize in certain offense types, such as property crimes or violence, in their criminal careers, but there is “a small but significant degree of specialize in the midst of a great deal of versatility” (Piquero, Farrington, & Blumstein, 2003, p. 454). In fact, DLC studies have shown that offenders increase their criminal versatility as their careers lengthen, so chronic offenders commit the widest variety of crime types (Farrington & West, 1993; Grubin et al., 2001; Moffitt, Caspi, Harrington, & Milne, 2002; Sullivan, McGloin, Ray, & Caudy, 2009). Other research suggests that offenders are more versatile at the start of their career, and then specialize in crime types that are better suited to their more refined tastes and skills so they become more “successful” offenders (see, for example, Blumstein, Cohen, & Farrington, 1988; DeLisi et al., 2011; Francis, Soothill, & Fligelstone, 2004; Mcgloin, Sullivan, & Piquero, 2009; Sullivan, Mcgloin, Pratt, & Piquero, 2006).

With respect to consistency in offense style rather than crime types, there is limited research. One study on sex offenders found that the offenders were versatile in the type of crimes they committed over their careers, but tended to specialize in the specific types of crimes they committed within sex offenses (e.g., rape, indecent assault; Soothill, Francis, Sanderson, & Ackerley, 2000). On the other hand, a recent study by Deslauriers-Varin and Beauregard (2013) found that sexual offenders are increasingly diverse in offending behaviors between their first and third offenses, but by the fourth crime they begin to specialize in specific criminal behaviors. Together, these findings suggest that there is much to be learned about the development of a “signature” offense style over a criminal career, and it is possible that specialization within a particular type of offending may differ from specialization in crime types over time.

Still, most studies on behavioral consistency have found a significant level of stability in the behavior of offenders across multiple offenses, which is encouraging for CLA, which requires behavioral consistency to work (see, for example, Bennell & Canter, 2002; Bennell & Jones, 2005; Tonkin, Grant, & Bond, 2008; Woodhams & Toye, 2007). However, it has also been shown that not all offenders are equally consistent in their behaviors, and not all offending behaviors are equally stable across crimes (see, for example, Alison, Goodwill, Almond, van den Heuvel, & Winter, 2010; Markson et al., 2010). Specifically, consistency in behavior may change depending on the style of offense an offender commits, and on which methodology is used to assess behavioral consistency. These major sources of variation in past behavioral consistency research will therefore be discussed in detail below.

J Coefficient

The most popular method of measuring behavioral consistency in CLA research has been to use Jaccard’s (1901) similarity coefficient, J (see, for example, Bennell & Canter, 2002; K. Davies et al., 2012; Markson et al., 2010; Tonkin et al., 2008; Woodhams & Toye, 2007). J has been used to evaluate the similarity in behaviors for two consecutive offenses by assessing how many behaviors are consistent between the offenses out of the total number of behaviors witnessed at the crimes (Melnyk et al., 2011). As such, the formula to calculate J for a pair of crimes (A and B) is as follows:

where a equals the number of behaviors common to both crimes, and b and c are equal to the number of behaviors unique to crimes A and B, respectively (Jaccard, 1901). J ranges from 0, indicating complete inconsistency between the two offenses, to 1, indicating complete behavioral consistency among the crime pair (Jaccard, 1901). Consistency is measured by comparing each offense with a subsequent offense for the full length of each offender’s series.

While J coefficient is quite simplistic, and therefore useful to academics and practitioners without advanced crime analysis software or statistical training, it has also been stated that J is a “relatively coarse-grained coefficient and it may be useful in the future to develop a more refined similarity measure” (Bennell & Canter, 2002, p. 157). For example, J is reliant upon the similarities or differences in individual behavior at an offense, rather than relying on broader themes of behavior. This makes J very sensitive to even the smallest variations in offending behavior between crimes (Melnyk et al., 2011). J also tends to be influenced by the frequency of the behaviors being evaluated, and the sample size under consideration.

In addition, the coefficient can only account for the similarity in behaviors in two crimes in an instance, and a crime analyst intending to conduct CLA by using J for an unsolved offense and a host of solved cases would require a new calculation for each pair of individual pair of offenses. However, as the J coefficient does not take joint non-occurrences (i.e., behaviors absent in both crimes) into account, missing data do not skew the results (Bennell & Canter, 2002; Deslauriers-Varin & Beauregard, 2013; Woodhams, Grant, & Price, 2007). But, if two unrelated crimes with missing data initially appear similar and the information on the dissimilarities is missing, then J could mislead us into thinking that there is high behavioral consistency and the crimes are likely linked (Melnyk et al., 2011). Still, some say this may make J an effective method of analysis for CLA, as the police data most commonly used for the analysis may not always be fully complete due to time and organizational limitations, and missing data may be the result (Melnyk et al., 2011).

Jaccard’s similarity coefficient is a rather basic measure of behavioral consistency, but it is quite practical for use by law enforcement, whom CLA is intended for. However, the simplicity and durability of J may also be its weakness, particularly when applied in the field. For those reasons, alternative methods of assessing behavioral consistency, drawn from DLC research, will also be evaluated.

D Index

The D Index was initially created by statisticians Agresti and Agresti (1978) to measure species variation. Due to its ability to examine individual-level patterns of specialization, the D index has increasingly been used as a measure of offending versatility in DLC research (e.g., Mazerolle, Brame, Paternoster, Piquero, & Dean, 2000; Miethe, Olson, & Mitchell, 2006; Piquero, Paternoster, Mazerolle, Brame, & Dean, 1999; Sullivan et al., 2006). This index measures the probability that any two crimes, selected at random from an offender’s criminal career, are different types of offenses (Lussier et al., 2008). As D is an individual-level measure and there is no comparison of behaviors in a sequence of offenses, the D index draws instead on the proportion of crimes that fall into specific categories within the selected set of offenses (Lussier et al., 2008).

Consequently, the D index does not consider chronology in offending, allowing all offenses to be evaluated simultaneously for consistency (Lussier et al., 2008). This is major difference from the FSC, which assesses the degree of consistency between consecutive crimes. A major benefit of the D index is that the individual-level focus allows for the determination of specialization values for a specific offender, instead of a specific behavior or offense style as in J coefficient and the FSC, respectively.

Several studies have capitalized on these differing but complementary features by using both the FSC and D index to measure offense specialization, and they have found highly similar results using the two methods (Lussier et al., 2008; Miethe et al., 2006; Piquero et al., 1999; Sullivan et al., 2006). The D index can be calculated using the following equation:

where k = the number of categories, and p_i = the proportion of an offender’s crimes that fall into each of the i = 1, 2, …, k categories of interest. Like the FSC and J coefficient, the D Index theoretically can range in value from 0 to 1, but a D of 0 represents complete likelihood of specialization while a 1 indicates complete diversity in behavior across offenses. More significantly, the maximum range of the D index scale will vary according to the distribution and number of categories (k) in the analysis, as D m a x = k − 1 k (Lussier et al., 2008). For instance, if four categories of offense styles are considered, as in this case, D will range from a minimum of 0.0 (complete specialization) to a maximum of .75 (complete versatility). Given this fluctuation in the scaling metric of the D index, some concerns have naturally been raised about this measure of behavioral stability. Due to this limitation, the comparability and interpretation of the D index from studies with differing coding schemes and index scales is strongly cautioned (Sullivan et al., 2009). Another issue is that the D index is somewhat confounded with offending frequency, as the number of offenses in a crime series must be at least equivalent to the number of categories being assessed or the maximum diversity score can never be reached, even if every crime is of a different type (Sullivan et al., 2006). While the extent of this limitation remains unclear, it is important to consider in CLA research with rather limited data on serial offenses.

The final limitation of the D index is rather significant, as D provides a quantitative measure of an offender’s degree of specialization, but no information on the nature of the specialization (Sullivan et al., 2009). This means that unless each crime type or style is dichotomized (e.g., violent versus non-violent crime), it is impossible to determine the type of crime or offense behavior that an offender specializes in. Again, this is a major difference between the D index and FSC, as the FSC lacks information on individual-level specialization that the D index provides, but the FSC can show which specific offense styles are most consistently repeated by offenders.

FSC

Farrington (1986) developed the FSC to measure the degree of specialization or versatility in offending across an individual’s criminal career. The FSC has now become one of the favored measures of offense specialization in DLC research due to its appealing features such as its statistical rigor, intuitive interpretation, and ease of calculation (Paternoster, Brame, Piquero, Mazerolle, & Dean, 1998; Sullivan et al., 2009).

While DLC and CLA research have different focuses of analysis (crime types vs. offense styles/behaviors), and use different terms for behavioral stability (specialization vs. consistency), the goal of observing variation or similarity in criminal behavior is shared between both fields. Although DLC research has not yet examined crime specialization as a mechanism for linking crimes, as specialization in criminal offending is very similar to consistency in offending behavior, several analytical techniques developed in DLC research, including the FSC, could easily be applied to evaluate behavioral consistency in CLA (Sullivan et al., 2009).

There are other methods of studying specialization in a criminal career, such as trajectory analysis, but the FSC is highly suited for analyzing consistency in behaviors from one offense to the next (Deslauriers-Varin & Beauregard, 2013). For instance, the FSC is not influenced by sample size or the distribution of offenses in the transition matrices, meaning that rarer types of offenses (e.g., homicide) or styles of offending behavior will not bias the analysis in any way (Farrington, Snyder, & Finnegan, 1988). The FSC can also evaluate the level of specialization in one particular behavior or style of behaviors across more than two offenses, and examine the consistency of offending behaviors across an infinite number of crime types (Deslauriers-Varin & Beauregard, 2013). In a transition matrix of offending behaviors across a series of crimes, the FSC is calculated using the formula:

where O = observed cell count, E = expected number in the cell by chance, and R = row total (Farrington, 1986; Farrington et al., 1988). By applying the FSC to each of the diagonal cells in a transition matrix, the level of specialization and versatility in behaviors may be evaluated (Paternoster et al., 1998). The FSC ranges from 0, indicating perfect versatility in offense behavior (as the observed change is equal to chance), to 1, indicating perfect forward specialization in offending (as every offense of “type x” at time k is followed by another “type x” offense at k + 1; Farrington et al., 1988; Stander, Farrington, Hill, & Altham, 1989, p. 326). The FSC can also take negative values if there was negative specialization (i.e., a tendency for offense “type x” at time k to not be followed by offense “type x” at k + 1; Stander et al., 1989, p. 326). The Adjusted Standardized Residual (ASR) may be used to test the statistical significance of the FSC by indicating whether the observed value is significantly above or below chance expectation (Farrington et al., 1988). The ASR is computed using the formula:

where O = observed cell count, E = expected number in the cell by chance, R = row total, C = column total, T = grand total, * indicates multiplication, and E = R * C T .

While the ease of calculation and interpretation of the FSC has made it one of the most popular measures of specialization in DLC research, it also has its limitations. For instance, some have stated that the interpretation of the FSC is too subjective, and what defines high levels of specialization and versatility may vary between researchers, or the data being analyzed (Britt, 1996). To date, there have not been any specific values set to indicate what constitutes a low, moderate, or strong level of versatility or specialization in the FSC, but due to its similarity to correlation coefficients, the same interpretation used in correlations may be applied to the FSC (Paternoster et al., 1998, p. 138). Also, the FSC is an aggregate measure of offense types, or in this case, offending behavior styles, rather than an examination of the distinct crimes or behaviors of specific individuals (Sullivan et al., 2006). In other words, the FSC is able to describe macro-level patterns in behavioral stability within a particular sample, but making statements about individual-level patterns of specialization or versatility is not possible (Piquero et al., 1999). Consequently, some researchers have turned to a third measure of behavioral consistency, called the D Index (Agresti & Agresti, 1978; Sullivan et al., 2006).

In this article, to enable a more interpretable comparison of results from the D index, FSC, and J coefficient, all D scores are inverted so they are on the same scale (0 indicating diversity, 1 indicating consistency) as the FSC and J coefficient. A summary of the properties, benefits, and limitations of the three analyses just described is presented in Table 1.

OPEN IN VIEWER

Table 1.

Properties and Key Points of Three Methods of Behavioral Specialization Analyses.

Analytical method	Properties	Points to consider
Jaccard’s coefficient (J)	Aggregate measure of similarity in individual behaviors across two offenses (but not more).	Does not account for joint non-occurrences, so missing data does not skew results. Influenced by behavior frequency and sample size.
Forward specialization coefficient (FSC)	Aggregate measure of specialization in themes of behavior across unlimited consecutive offenses. Offense style is the unit of analysis.	Not influenced by sample size or distribution of offenses. Statistical significance of FSC tested using ASR values. Interpretation of results may vary.
Diversity Index (D)	Individual-level measure of specialization that considers the totality of offending patterns over sampled offenses.	Does not consider chronology of offending (or specify the behaviors an individual specializes in). Index varies by categories included in each study, and maximum value can be less than 1.

Aims of the Study

While the three methods of specialization analyses used in this research are quite different in nature, no studies comparing the three approaches have found significant support for the use of one technique over the other (Deslauriers-Varin & Beauregard, 2013; Melnyk et al., 2011). Furthermore, no study has used these methods to directly compare the behaviors of American burglars, and very little research has been conducted to evaluate how behaviors or the consistency of behavior differs between offenders specializing in different styles of offenses. As stated by Tonkin and colleagues (2008), future research must focus not on whether behavioral consistency exists, but when, and in what circumstances it exists, and how these impact the comparability and applicability of research intended for police use in the field.

Therefore, this research aims to examine the following three questions:

Method

Data for this study were obtained from case files and criminal history records on a random sample of convicted serial burglars provided by a police department on the east coast of Florida. A random sample of 405 solved burglaries committed in the police jurisdiction between 2008 and 2009 was drawn, ⁵ and the criminal history for all responsibleoffenders were collected to determine whether the burglars had committed additional burglaries any time before or after the sampled time frame. If an offender committed another burglary, the offense reports and arrest reports for these offense(s) were retrieved to create a sample involving only serial burglars and their full series of offenses. Offenders with no additional burglaries were not included in the final sample.

In total, the final sample consisted of 58 serial burglars who were responsible for 148 solved burglaries. Most offenders in the sample (n = 36) committed two burglaries, 15 committed three burglaries, and 7 offenders committed four or more burglaries. The highest number of recorded offenses by a single offender was six. Although past studies of CLA have included only two or three crimes per offender, to limit the influence of more prolific criminals on the findings (see Bennell & Canter, 2002), recent research has shown that selecting a set number of crimes out of a series does not reflect reality and may underestimate an offender’s level of behavioral consistency (Woodhams & Labuschagne, 2012). To account for the total number of offenses committed by the sampled burglars, while also assessing consistency in a relatively recent and commonly used time period in CLA research, ⁶ the past 5 years of criminal history were included in the present analysis.

The average time span between the solved burglaries was 5 months, with the shortest time between burglaries less than a day, and the longest time frame between burglaries 1,649 days, or approximately 4.5 years. Sampled offenses took place in the same police jurisdiction in the United States.

Offense Styles and Offending Behaviors

Prior research on behavioral consistency has shown that, in general, analyzing a multivariate profile of offending behaviors, rather than each individual crime scene behavior, is most successful in determining the stability of behaviors across multiple offenses (Grubin et al., 2001; Harbers et al., 2012; Markson et al., 2010; Santtila et al., 2004; Woodhams & Toye, 2007; Yokota & Canter, 2004). This is because a set of offending behaviors, or “profile,” may convey a broader picture of the conduct of the offender than a single behavior. Furthermore, some offense behaviors are quite common among all offenders, and therefore little knowledge may be gained from finding consistency in this behavior (Grubin et al., 1997). On the other hand, while highly unusual behaviors are helpful for distinguishing unique offenses, when assessed individually, they may then be too rare to help evaluate behavioral consistency for the majority of offenders who did not commit these behaviors (Grubin et al., 1997). Therefore, the use of an associated group of behaviors (i.e., an offense style), rather than analyzing individual crime scene behaviors, has been considered the most effective method of evaluating consistency (Grubin et al., 1997; Woodhams & Toye, 2007).

Due to the nature of the analyses being tested in this study, both approaches to evaluating behavioral consistency will be utilized. For the J coefficient analysis, consistency in individual behaviors will be examined, but for the FSC and D index, the change in broader offense styles within burglary will be evaluated. Therefore, both the individual burglary behaviors and broader offense styles displayed in all 148 burglaries in the data set will be assessed in this project.

Individual burglary behavior variables

Information on the behaviors present at the burglary crime scenes was acquired from official police records on the reported burglaries, known as “707 files” or arrest reports. These reports contain a narrative of the case from the moment the offense was reported through to arrest, and provide all available details on the offense, the offender, the victim, and other relevant details of the crime. The records were pulled for all 148 offenses, and the 26 most significant behaviors (in terms of uncovering differing themes) committed at a burglary, such as the type and method of entry, type of premises, types of items stolen, target occupancy, time of offense, use of burglary tools, evidence recovered, and state of the crime scene, were coded for analysis.

Burglary offense styles

Multivariate burglary offense styles comprised of various combinations of individual burglary behaviors are used to evaluate offending behavioral consistency in the FSC and D index analyses. The burglary offense styles are based on the previously established and tested set of offending behavior patterns developed by Fox and Farrington (2012) for burglary committed in the United States. ⁷ The four styles of burglary identified in Fox and Farrington’s LCA were the organized, opportunistic, disorganized, and interpersonal style burglaries named after the “theme” of behaviors committed at the crime scene such as low levels of skill and sophistication, or high personal involvement in the offense. The four burglary offense styles, and the specific behaviors that comprise of each, are shown in Figure 1.

OPEN IN VIEWER

Figure 1.

Hierarchy of individual burglary behaviors and burglary offense styles.

Each of the 148 burglaries in the data set was categorized into one of four burglary offense styles using a detailed flow chart that classified each case into an offense style using the crime scene behaviors listed in the police reports. For instance, the organized offense is a highly sophisticated burglary, with behaviors indicating that foresight/care was taken to reduce risks and increase gains in the offense, such as using a tool the offender brought to the crime scene (such as a crow-bar or lock pick) to enter the dwelling, a ruse being used to case a target or gain entry, or evidence being cleaned up or purposely not left behind (as the offender wore gloves or a mask). Conversely, crime scenes characterized by more chaotic and careless traits such as ransacking the target and leaving forensic behind were classified appropriately as disorganized style offenses. All 148 burglaries were coded into a single offense style, with no overlap.

Frequency data for each offense behavior and offense style are shown in Tables 2 and 3, respectively.

OPEN IN VIEWER

Table 2.

Frequency Distribution of Individual Burglary Behavior Variables.

Burglary behaviors	Frequency (%)
Unlawful entry	22.3
Forced entry	73.6
No tool brought to crime	51.4
Tool brought to crime	45.3
Tool left behind at crime	16.2
Tool taken from crime	29.1
Scene left in disarray	32.4
Scene tidy (left as found)	64.9
No evidence left behind	39.9
Evidence left behind	41.2
Utilitarian motivation	76.4
Excitement/thrill motivation	16.9
Anger/dispute motivation	4.1
Unoccupied target	77.7
Occupied target	17.6
Residential target	66.2
Commercial target	31.8
Daytime offense	45.9
Nighttime offense	47.3
High value item stolen	60.8
Low value item stolen	17.6
Drugs stolen	2.7
Nothing stolen	16.2
Escape successful	75.0
Escape unsuccessful	20.3

Note. n = 148 offenses.

OPEN IN VIEWER

Table 3.

Frequency Distribution of Burglary Offense Styles.

Burglary behaviors	Frequency (n)	%
Organized	20	34.5
Disorganized	14	24.1
Opportunistic	22	37.9
Interpersonal	2	3.4

Note. n = 58 offenders.

Analytical Strategy

Our analysis of the behavioral consistency in offending styles for American serial burglars is performed using three different analytical techniques: Jaccard’s similarity coefficient, the D index, and the FSC. An evaluation of the variation in behavioral consistency by an offender’s burglary offense style and length of offense series will be individually presented for each of the three analyses. Results of the these analyses are then compared to determine what, if any, differences in the consistency of the serial burglars’ offending behaviors were found among the techniques.

Results

J coefficient

Table 4 presents the results from Jaccard’s similarity coefficient. The average J value for the 75 sampled offense pairs was .52 (SD = .22), which is mid-way between complete consistency (1) and complete inconsistency (0). This finding suggests no significant inclination toward consistency or versatility in the crime scene behaviors across multiple offenses using the J coefficient.

OPEN IN VIEWER

Table 4.

Jaccard’s Coefficients for Burglary Offense Behavior Specialization.

	Jaccard’s coefficient
Burglary behaviors	M (SD)	Minimum-maximum
Offense series
2 burglaries	.53 (.23)	0.10-1.0
3-4 burglaries	.52 (.20)	0.22-1.0
5-6 burglaries	.30 (.14)	0.22-0.47
Offense style
Organized	.52 (.22)	0.24-1.0
Disorganized	.49 (.23)	0.16-1.0
Opportunistic	.56 (.23)	0.10-0.82
Interpersonal	.38 (.12)	0.29-0.47
Overall Jaccard’s coefficient	.52 (.22)	0.10-1.0

Note. n = 75 pairs of offenses.

Past studies have shown specialization trends vary for infrequent versus prolific offenders, and behavioral consistency research suggests offending behavior consistency takes a non-linear form by decreasing in consistency and then eventually increasing as crimes in an offender’s series increases (Deslauriers-Varin & Beauregard, 2013). J coefficient results show a similar trend in this study, as burglars who committed two offenses were not consistent or inconsistent (J = .53, SD = .23), and similar results were found for burglars with three to four burglaries in their series (J = .52, SD = .20). However, once burglars hit five to six burglaries in their series, their level of specialization in offense behaviors noticeably decreased (J = .30, SD = .14). There was no significant relationship between an offender’s individual behavioral consistency and the number of offenses in their crime series.

While J coefficient is designed to evaluate change or stability in individual behaviors, it is possible to evaluate the general consistency of crime scene behavior by classifying the burglars’ first offenses into offense styles and evaluating change in style between two crimes. When the offenders were grouped according to their burglary offending styles, there was some variation in the behavioral consistency for the burglars in the four different offending styles. The J coefficient was highest for the opportunistic burglars (J = .56, SD = .23), suggesting that the individual crime scene behaviors were most likely to be repeated by the opportunistic style burglars as compared to the offenders with other three crime styles. Conversely, the interpersonal style offenders were the least consistent (J = .38, SD = .12), but there were only two of these. Organized and disorganized burglars were the second and third most consistent in behaviors, respectively (J = .52, SD = .22; J = .49, SD = .23), though, like the opportunistic style burglars, the organized and disorganized offenders appear to be almost as inconsistent in their behaviors as they are consistent. Therefore, J coefficient suggests that crime scene behaviors are, in general, about half-way between complete consistency and complete inconsistency.

D Index

The reverse D Index scores ⁸ for the serial burglars’ offending style are presented in Table 5. The average D score for the sample was .82 (SD = .23), which indicates that, in general, the burglars had a very low level of diversity in their offense style across crimes (as 1 represents complete specialization in the reverse D index). The minimum diversity value (D = .25) was not achieved by any offender, though 24% reached the sample’s highest diversity level of .50, and 60% of burglars had perfect specialization in offense style.

OPEN IN VIEWER

Table 5.

Reverse Diversity Index Scores for Burglary Offense Style Specialization.

	Diversity Index
Burglary behaviors	M (SD)	Minimum-maximum
Offense series
2 burglaries	0.79 (.25)	0.50-1.0
3-4 burglaries	0.82 (.21)	0.56-1.0
5-6 burglaries	1.0 (.0)	1.0-1.0
Offense style
Organized	0.87 (.21)	0.50-1.0
Disorganized	0.83 (.22)	0.50-1.0
Opportunistic	0.76 (.25)	0.50-1.0
Interpersonal	0.50 (.00)	0.50-1.0
Overall Diversity Index	0.82 (.23)	0.50-1.0

Note. n = 75 pairs of offenses.

When D index results are examined by the number of offenses in a crime series, an interesting pattern emerges. Unlike Jaccard’s coefficient which showed that offenders became more versatile as they committed more offenses, results from the D index show that burglars who committed more crimes became increasingly consistent in terms of their offense style behaviors. Specifically, there is an inverse relationship between the offenders’ mean diversity index scores and the number of burglaries they have committed. Furthermore, no burglars who committed three offenses reached the highest potential diversity value, while almost 40% of the burglars with two crimes in their series reached the maximum diversity level. In fact, all of the most prolific burglars with five or more offenses showed 100% perfect specialization, with a D score of 1. The relationship between D score and the number of offenses the offender committed was statistically significant (χ² = 58.40, df = 12, p < .001).

To address whether a burglar’s level of consistency varies depending on the style of behaviors committed at the crime, the average D index scores for offenders in each of the burglary offense styles were calculated for comparison. Results show that burglars who originally committed an organized offense tended to be the most specialized in their series of burglaries (D = .87, SD = .21), while interpersonal style burglars were the most diverse in their burglary offense behavior (D = .50, SD = .00). Offenders who committed a disorganized style burglary were relatively specialized in offending behaviors (D = .83, SD = .22), while opportunistic burglars were slightly more versatile (D = .76, SD = .25). The relationship between a burglar’s offense style and D score was statistically significant (χ² = 21.10, df = 9, p < .01). It is not possible to infer from the D index the follow-up offense styles committed by the burglars that were more versatile, or in what order the changing offense style behaviors took place.

FSC

The FSC values for the four burglary offense styles are presented in the transition matrix shown in Table 6. As the FSC is an offense-focused analysis, the FSC and ASR values were determined by evaluating consistency in offense style for all 148 crimes in the data set. FSC values were not calculated for individual offenders, and no analysis of FSC values by length of crime series was available. However, the FSC was specifically designed for evaluating the specialization in behaviors across a series of offenses, as studied in the present analysis.

OPEN IN VIEWER

Table 6.

Forward Specialization Coefficients (FSC) for Burglary Offense Style Specialization.

	Next offense style (k + 1)
Past offense style (k)	Opportunistic	Organized	Disorganized	Interpersonal
Opportunistic	.35 *	−.27	−.15	—
Organized	−.18*	.78 *	−.23*	—
Disorganized	−.12	−.56*	.68 *	—
Interpersonal	—	—	—	—

Note. n = 75 pairs of offenses. — = not enough values for testing. Bolded cells indicate tests of offense style consistency.

*

A significant FSC value at the p < .05 level.

Results of the ASR tests show that the FSC values for the organized, disorganized, and opportunistic style offenses were statistically significant. No conclusions regarding the consistency of behavior for interpersonal style burglars are able to be drawn using the FSC due to lack of statistical significance resulting from the low number of repeat interpersonal burglars.

Among the remaining offense styles, all three showed moderate to strong FSC values, suggesting that these burglars are rather consistent in their specific style of behaviors across offenses. Specifically, burglars committing the organized style crime were the most consistent in (FSC = .78, ASR = 6.1), and the disorganized style offenders were second most consistent across their offenses (FSC = .68, ASR = 6.8). The FSC scores for the organized and disorganized offense styles are statistically significant at the p < .05 level. Opportunistic style burglars were the least consistent from one crime to the next, as only a moderate, but statistically significant, level of consistency was found (FSC = .35, ASR = 3.9). The remaining statistically significant results were negative values, indicating a significant likelihood to not commit a different style offense after a given offense style. This again indicates the consistency of burglary offending styles.

Comparison of Analytical Methods

In addition to evaluating the consistency of burglars’ offending styles across their crime series, the third aim of this study is to compare results from popular analytical techniques used in CLA and DLC to assess the findings among the methods. To increase comparability, the D index scores were converted to be on the same scale as J coefficient and FSC, and J coefficients were calculated for offense styles (not just individual behaviors) so results from each method are suitable for direct comparison. Results from each analysis are displayed in Table 7.

OPEN IN VIEWER

Table 7.

Comparison of Burglary Offense Style Specialization by Analytic Method.

Offense style	Jaccard’s	Diversity Index	FSC
Organized	.52	.87	.78
Disorganized	.49	.83	.68
Opportunistic	.56	.76	.35
Interpersonal	.38	.50	—
Overall coefficient mean	.52	.82	.60

Note. FSC = forward specialization coefficient.

T tests show that there is a statistically significant difference between the overall mean values for J coefficient and the D index, as well as between the FSC and D index scores (p < .05). The significant variation in results indicates that there are critical differences among the techniques that impact the output. In other words, the three analyses are not measuring the same thing in the same way.

A closer look at behavioral consistency in burglary offense style shows that, in general, the D index is the most skewed toward higher values indicating specialization (M_D = .82), while J coefficient generally has the lowest values and suggests a stronger level of diversity in all offending behaviors (M_J = .52). By comparison, the three offense styles evaluated by the FSC are more evenly distributed across the specialization scale (M_FSC = .60), and no values are at either extreme end of the spectrum. Recall that the minimum value of the D index is not 0, as with the FSC and J coefficient, but .25.

There are also important differences in offense style consistency across the three methods used in this study. In two out of three analyses, the organized offenders were most consistent in crime scene behavior, and most likely to repeatedly commit burglaries in an “organized” fashion. Only when using J coefficient are the opportunistic style burglars slightly more consistent in behavior than the organized offenders (J _op = .56 vs. J _org = .52). In fact, there was a wide variation on the level of consistency among the organized burglars depending on the method used to calculate the offense style specialization. For instance, the D index shows high levels of specialization in organized style burglaries, while J coefficient indicates almost no inclination toward specialization or diversity in behavior for the organized style offenders (D = .87 vs. J = .52). The FSC value for organized style burglars (FSC = .78) fell in between values found using the D index and J coefficient.

There were also variations between the three analyses for the consistency of behavior shown by burglars committing opportunistic style offenses (J = .56, D = .76, FSC = .35), and disorganized offenses (J = .49, D = .83, FSC = .68). A diagram illustrating the variation in behavioral consistency for burglary offense styles depending on the analytical technique is shown in Figure 2.

OPEN IN VIEWER

Figure 2.

Distribution of burglary offense style specialization scores by analytic method.

In short, although it is not possible to infer from this output alone which method is more useful, it is important to consider the variation in findings among the analyses and why such variation may be occurring. These and other considerations are discussed in detail to follow.

Discussion

This study assessed the level of consistency in burglary offense styles shown by serial offenders across a series of crimes, and compared three of the most popular methods of evaluating behavioral consistency using a sample of American serial burglars. Through this study, we hoped to better understand how different styles of behavior and analytical techniques can impact behavioral consistency, and in turn, the ability for police to utilize case linkage analysis in their unsolved investigations.

Conclusion

Due to the distinct and varying patterns of specialization in offense styles across a crime series, there is reason to believe that certain mechanisms may be underlying an individual’s level of behavioral consistency while offending. While there was no data available on the psychological, developmental, or environmental factors that could have influenced the burglars’ decisions to commit their crimes in a similar or different way to past offenses, the results of this research seem to support hypotheses on how certain personality traits, psychological and developmental issues, and situational factors can systematically impact the stability of criminal and non-criminal behavior (Youngs, 2004). It is possible that an offender’s level of consistency in crime scene behaviors is driven in part by internal or external factors that lead offenders to commit crimes in a specialized or versatile way. In other words, it may be that the influence of psychological, developmental, or situational factors is stronger for certain types of offenders than others, leading to different styles of offenses and different levels of consistency in those offending behaviors. These findings, paired with future research on the background of serial offenders with varying styles of offending, could serve as the foundation for the first theory on how, and why, offenders choose to commit crimes in certain ways. The characteristics of consistent and inconsistent offenders should be investigated and compared in the future.

With respect to the methodological comparison conducted in the present research, results indicate that the FSC is the most useful approach of the three methods. In terms of applicability of the methods in varying contexts and intended uses, this study suggests that if a researcher or practitioner wants a summary measure of specialization from offense N to offense N + 1, there is a choice between J and FSC, but our findings suggest that the FSC is the preferred analysis to utilize. However, if a researcher is intending to measure specialization across an offender’s entire crime series, the D index may be best for that purpose.

As no quantifiable measure to determine the accuracy and efficiency of the three analytical methods was available in the present research, we defer to Sullivan and colleagues’(2009) recommendation to utilize multiple analytical techniques when proceeding in future offense specialization research, as each method provides a distinct perspective and reveals a unique aspect of offending consistency that cannot be gained when using one analysis alone.

In terms of the applied uses and implications of this study, there is potential to utilize the knowledge on the variation in burglary offense style consistency to help identify the most likely suspects in unsolved burglaries due to a highly similar style of behavior in past offenses, or limit suspect pools by lowering the priority of known offenders with burglary offense styles in stark contrast to the behavior a new unsolved offense. For instance, crime analysts and detectives searching for a lead on a new case could look to solved burglaries and, depending on the style of offense that was committed, see if any crimes of a similar style were committed in the recent time frame and geographic location to the unsolved offense. As forensic evidence is so rarely available to help identify a burglar, the ability to draw a lead based upon behavioral evidence seen at a crime scene may make a critical difference in cracking a substantial number of burglaries that would otherwise go unsolved.