Investigating Offending Consistency of Geographic and Environmental Factors Among Serial Sex Offenders

Abstract

Crime linkage analysis constitutes a tool to help investigators prioritize suspects, but a scarcity of research and methodological issues limits our knowledge on behavioral consistency in sexual offenses. The current study identifies geographic and environmental factors that are useful in examining offending consistency across series of sexual assaults using different specialization coefficients. The current study draws on criminal career research and methodology as a way to improve the study of behavioral consistency. The sample includes 72 serial stranger sex offenders who have committed a total of 361 sexual assaults. Three methods are used (i.e., diversity index, forward specialization coefficient, and Jaccard’s coefficient) and reveal a high degree of offending consistency. All three methods also highlight promising factors to rely on for crime linkage of serial sexual offenses. Empirical and methodological implications for behavioral consistency research are discussed as well as practical implications for police investigations and crime linkage.

Keywords

crime linkage behavioral consistency criminal career sexual offending versatility specialization

Suspect prioritization is a central component of criminal investigations as it helps to narrow down the overwhelming number of potential suspects while reducing the direct and collateral costs associated with an investigation (e.g., length of investigation, number of police officers working on the case). The most reliable way to identify a suspect is through forensic evidence (e.g., DNA, fingerprint) found at the crime scene (Grubin, Kelly, & Brunsdon, 2001) or through a confession by the offender. In the absence of a confession, eyewitness, or physical evidence, however, other methods must be used to assist police investigators in identifying potential suspects. One such method, crime linkage, involves the identification of similarities between offenses of the same type to help identify the individual responsible for the crime being investigated. More specifically, with the use of police databases, crime linkage helps to determine if a crime for which the offender is not yet known presents evidence of similar offender behaviors with another (previous) crime for which the offender is already known (Woodhams, Hollin, & Bull, 2007). In other words, it is believed that offenders will repeat the same crime but will also commit these crimes in a consistent way across crime events.

Associated with the emergence of crime linkage, researchers from the field of investigative psychology started to question whether offenders are in fact consistent in the way they commit their crimes. As such, they recently began to examine behavioral consistency in offending using different analytical methods. However, this line of research has emerged and evolved almost independently of a long tradition of empirical research on offending specialization in criminology: the criminal career paradigm. The current study aims to bridge this gap by presenting findings and analytical methods from this field as a way to complement the investigation of offending consistency in sex offending.

Empirical Findings and Methodological Issues in Behavioral Consistency Studies

Most of the studies on behavioral consistency have used Jaccard’s similarity coefficient to quantify the degree of similarity and consistency that exists between crimes committed by the same offender (e.g., Bennell & Canter, 2002; K. Davies, Tonkin, Bull, & Bond, 2012; Markson, Woodhams, & Bond, 2010; Tonkin, Grant, & Bond, 2008; Woodhams, Hollin, & Bull, 2008; Woodhams & Toye, 2007). This coefficient provides information on the degree of consistency across consecutive offenses for each behavior examined. Jaccard’s coefficient is widely used in crime linkage studies as it does not take into account joint nonoccurrences (i.e., 0/0), meaning that if a particular behavior is absent across two crimes, the level of similarity between those crimes will not increase (Bennell & Canter, 2002). As such, this coefficient has a great advantage when using official or police data, as the information recorded about a crime is not always entered in a consistent manner and the absence of a behavior in an offense report does not mean that the behavior did not occur (Woodhams, Grant, & Price, 2007). Other coefficients have been proposed and used (e.g., taxonomic similarity index; Woodhams, Grant et al., 2007), but recent studies have not found significant support for the use of one coefficient over the other (Melnyk, Bennell, Gauthier, & Gauthier, 2011).

Studies analyzing offending consistency have typically examined modus operandi behaviors and have shown evidence that offenders commit crimes in a consistent manner across different types of crime (Grubin et al., 2001; Santtila, Fritzon, & Tamelander, 2004; Santtila, Junkkila, & Sandnabba, 2005; Woodhams & Toye, 2007; Yokota & Canter, 2004). In recent years, however, studies have examined the geographic and spatial information of the offense (e.g., average distance travelled, range of distance travelled, spatial patterns) and found that these factors could outperform more traditional modus operandi behaviors (Bennell & Canter, 2002; Bennell & Jones, 2005; Bernasco, 2008; Goodwill & Alison, 2006; Markson et al., 2010; Snook, Wright, House, & Alison, 2006; Tonkin et al., 2008). This is especially true for the intercrime distance, which was recently found as a promising aspect to rely on to facilitate crime linkage across crime types and categories (Burrell, Bull, & Bond, 2012; Tonkin, Woodhams, Bull, Bond, & Palmer, 2011). In a recent study, Lundrigan, Czarnomski, and Wilson (2010) also examined the consistency displayed by serial sex offenders in the choice of crime location and the characteristics of the crime site selected (referred to as environmental consistency). Their results indicated that these offenders showed high environmental consistency across their crime series, suggesting that serial offenders are not randomly selecting environments to commit their crimes but that whatever might be influencing the selection of one environment is also influencing the selection of following environments.

Although evidence of behavioral consistency was previously found, the level displayed by offenders fluctuates throughout the studies. The unstable levels of behavioral consistency found appear to be partially due to the diversity of variables and behaviors investigated by previous studies. The time period between offenses forming a series might have played a role as well. Indeed, certain behaviors seem to be temporally less stable and more susceptible to situational influence, especially so for sexual offenses (Alison, Goodwill, Almond, van den Heuvel, & Winter, 2010; Markson et al., 2010). First, on the one hand, prior studies have shown that crimes committed in greater temporal proximity showed greater behavioral consistency and that the likelihood of stranger sex offenders displaying behavioral consistency across their series decreased when looking at longer series of crimes (Alison et al., 2010; Grubin et al., 2001; see, however, Harbers, Deslauriers-Varin, Beauregard, & van der Kemp, 2012). On the other hand, prior studies show that greater levels of consistency are found in offenses committed later in the series investigated as the offenders have achieved a certain offending knowledge and expertise to commit their crime more successfully (Grubin et al., 2001; Sorochinski & Salfati, 2010; Woodhams & Labuschagne, 2011).

Second, due to the interactional nature of this offense, sexual assault is subject to situational and contextual influence and more prone to crime switching, as offenders might be unable to successfully perform their initial cognitive script and reach their goal. More specifically, previous studies have shown that the target selection process of sex offenders depends heavily on the social, physical, and geographic environment as well as the victim’s behaviors and location prior to the crime (e.g., Beauregard, Proulx, Rossmo, Leclerc, & Allaire, 2007; Canter & Larkin, 1993; Deslauriers-Varin & Beauregard, 2010; Rossmo, 1997). Variation in the level of behavioral consistency found for this offending aspect is thus expected. However, prior studies have shown that behaviors that are less dependent on situational factors and over which the individual exerts control present higher level of consistency (Bennell & Canter, 2002; A. Davies, 1992; Markson et al., 2010; Sorochinski & Salfati, 2010). For example, Bennell and Jones (2005) argued that the location chosen to commit a crime is a decision on which the offender has control. As such, spatial behaviors and the crime site selection should present a higher level of consistency comparatively to other traditional modus operandi behaviors, making them a promising aspect to rely on for crime linkage purposes.

Empirical studies on behavioral consistency have used different methodologies, which could explain the unstable levels found and make it difficult to draw firm conclusions about the main assumptions of crime linkage analysis. For instance, previous studies have used different sampling procedures and criteria, which are a pivotal part of the research process because they determine the generalization of the findings. In previous studies examining behavioral consistency, all crimes committed during a specific period for a specific crime type (e.g., burglary, homicide, sexual assault) were selected. Also, when specified, the (follow-up) period used varied considerably from 2 (e.g., Markson et al., 2010) up to more than 30 years (e.g., Lundrigan et al., 2010; Santilla et al., 2008). However, as their samples were not individual-based (but rather crime-event-based; see, however, Sjöstedt, Långström, Sturidsson, & Grann, 2004), the selection of a longer period does not necessarily imply a long follow-up period for each offender included. Hence, in previous studies, the same start and end date was used for each offender, which means that at the start date of the study period, some offenders might have been at the beginning, while others might have been at the end of their crime series. As a result, most studies examined consistency for partial crime series only. Since prior studies found that the likelihood of behavioral consistency could fluctuate as the series gets longer, it is reasonable to think that biased and unstable results might be found when investigating behavioral consistency for partial crime series.

Another sampling procedure used in behavioral consistency studies was to include the same number of offenses per offender. Most of the time, these studies have restricted the number of crimes per offender to two or three offenses (see, however, Woodhams et al., 2008) to limit the influence that prolific offenders may have on the study findings (Bennell & Canter, 2002). In these cases, different procedures have been used to select the crimes to be analyzed. Some studies have used the most recent offenses committed by the offender (Tonkin et al., 2008; Woodhams & Toye, 2007), whereas others have selected the first known offenses (Bateman & Salfati, 2007; Salfati & Bateman, 2005; Sorochinski & Salfati, 2010). The advantage of these procedures is that the temporal sequence of the series of crimes committed by the offender is taken into account. Other studies have simply randomly selected a predetermined number of crimes among the series, which did not take into account the sequence of offending (Bennell & Canter, 2002; Bennell & Jones, 2005; Markson et al., 2010). Regardless of the selected procedure, only a sample of crimes in the series was examined, which is especially the case for the prolific offenders. In a recent study, however, Woodhams and Labuschagne (2011) found that selecting a number of crimes per series to limit the effect of more prolific offenders does not reflect reality and could in fact lead to an underestimation of the behavioral consistency. Researchers should thus examine all the crimes (of the same nature) committed by the offender rather than selecting a fixed number of crimes per offender.

The analytical strategies used in previous studies might also have influenced the findings. Two analytical strategies were generally used in past studies to test the behavioral consistency assumption of crime linkage. A few studies have investigated behavioral consistency by measuring specific single behaviors (e.g., rare behaviors) among offenses, a procedure that is also known as the signature approach (e.g., Bateman & Salfati, 2007; Harbers et al., 2012; Schlesinger, Kassen, Mesa, & Pinizzotto, 2010). A more common analytical strategy, however, was to measure similarities between themes or domains of behaviors (e.g., Bennell & Canter, 2002; Burrell et al., 2012; A. Davies, 1992; Grubin et al., 2001; Santtila et al., 2004, 2005; Woodhams, Grant et al., 2007). The idea behind this procedure is that consistency in offense behaviors is more likely to be expressed through the common themes underlying different combinations of actions rather than through a single salient action. In combining behaviors under themes, however, methodological issues can arise. For example, the way those themes are operationalized fluctuates from one study to the other and can most likely affect the consistency level found (see, for example, K. Davies et al., 2012). Moreover, by measuring themes or groups of behaviors, rather than individual behaviors, one can expect to find more behavioral consistency.¹ The consistency level shown by specific behaviors is also left unknown, and the identification of reliable salient components of offending behavior that could be used to link individual crimes as part of a series remains elusive (Sorochinski & Salfati, 2010). The insufficient knowledge about consistency of individual behaviors and the somewhat changing nature and composition of themes/domains of behaviors used in prior studies do not contribute to the development of sound knowledge in crime linkage.

Overall, the review of the scientific literature reveals the scarcity of research and knowledge regarding the behavioral consistency assumption of crime linkage analysis. Moreover, these studies are based on various conceptual, methodological, and analytical procedures, which limit the conclusion that can be drawn. Despite this, findings from previous studies indicate that, when examining suitable behavioral domains or individual behaviors, high stability exists among offenders’ behaviors and that these behaviors can be used to successfully link serial crimes (Bennell & Jones, 2005). More specifically, geographic and environmental factors recently appeared as reliable and stable aspects to look at for crime linkage purposes. Considering its importance, research is still needed to understand if behavioral crime linkage is an effective and feasible procedure to help in suspect prioritization. More specifically, as pointed out by Tonkin et al. (2008), future research needs not only to assess whether behavioral consistency exists but also when and, in this instance, where it exists. This is where knowledge from the criminal career field becomes valuable.

The Value of the Criminal Career Paradigm

Crime linkage analysis and the investigation of behavioral consistency have mainly been conducted in psychology, more specifically in the field of investigative psychology. This line of research, however, has emerged and evolved almost independently of a long tradition of empirical research in criminology. Hence, the presence of crime specialization and the extent to which offenders specialize in their offending has been a key question in criminal career research (see Nieuwbeerta, Blokland, Piquero, & Sweeten, 2011), and behavioral consistency researchers could significantly benefit from the scientific knowledge, methods, and findings coming from this field. Criminal career research can be described as a scientific paradigm that focuses on the longitudinal sequence of crime committed by individuals (Blumstein, Cohen, Roth, & Visher, 1986). Thus, instead of focusing on the crime event, the individuals and their criminal behaviors throughout their career are examined. This longitudinal sequence or “criminal career” has been conceptualized along various crime parameters such as the age of onset, the offending frequency (or lambda), the seriousness (or aggravation), the length of the criminal career, and the desistance (or termination of offending) (Blumstein et al., 1986).

Crime specialization and crime switching

Criminal career researchers have also stressed the importance of examining two other aspects of the longitudinal sequence of offending: (a) crime specialization (i.e., the tendency to repeat the same crime type) and (b) criminal versatility or crime switching (the tendency to commit a wide array of crime types). Based on these parameters, “specialist” offenders are thus those who predominantly commit a specific type of crime and tend to engage in that behavior repeatedly and frequently, whereas, on the opposite, versatile offenders tend to commit many different types of crime over time without specializing in any of them specifically (Harris, Mazerolle, & Knight, 2009).

In the traditional view of criminal careers, according to Blumstein, Cohen, and Farrington (1988), offenders are thought to “sample” a wide variety of offenses during the early phase of their career before becoming more specialized, with time, in particular crime types that are more suited to their tastes and skills. In other words, crime switching should be expected in the first few arrests, whereas crime specialization should be found for more prolific offenders. In line with this hypothesis, Brennan, Mednick, and John (1989) found evidence of specialization in violent and property offenses for more prolific offenders (e.g., with more extensive records). More recently, researchers have used a more circumscribed approach to analyze specialization (e.g., life-course perspective) and found that, although generality is the norm when offending is viewed over the long-term criminal career, offenders tend to specialize in the short term (i.e., months or years rather than developmental stages or the full career) (e.g., DeLisi et al., 2011; Francis, Soothill, & Fligelstone, 2004; McGloin, Sullivan, & Piquero, 2009; Soothill, Francis, Sanderson, & Ackerley, 2000). For example, Sullivan, McGloin, Pratt, and Piquero (2006) found that the level of specialization progressively decreased as the time window of focus grew broader (i.e., month to year to 3 years).

Although it has been a common tendency in criminology to dichotomize offenders as specialists or generalists, Soothill et al. (2000) suggest that such distinction is not necessary and that sex offenders can, in fact, be both. Hence, results of their study show evidence that although sex offenders tend to be versatile in terms of their participation in crime over their criminal career, they do specialize in specific types of sex offending (e.g., indecent assault, rape) among their sexual criminal career. Crime specialization, however, has been examined for various crime types but not within specific crime type, which has been completed in behavioral consistency and crime linkage studies (see, however, Leclerc, Lussier, & Deslauriers-Varin, in press; Lussier, Leclerc, Healey, & Proulx, 2008). For example, criminal career researchers have examined the tendency to specialize in violent crimes, whereas crime linkage researchers have looked at the tendency for violent offenders to commit their violent crimes in the same way. Although the research question is different, the analytical approach is the same (i.e., examining whether offending behaviors are similar). Criminal career researchers have not examined crime specialization and criminal versatility for the purpose of crime linkage but their research has produced various methodological options to analyze these crime parameters (Sullivan, McGloin, Ray, & Caudy, 2009) that are suited for the examination of behavioral consistency. For the purpose of this study, crime specialization is not used in its traditional meaning but is rather defined here as equal to the term consistency used in the behavioral consistency and crime linkage fields—that is, the tendency to commit the same crime type (i.e., sex crime) in the same way.

Crime specialization and the forward specialization coefficient (FSC)

A recent review of crime specialization techniques by Sullivan et al. (2009) has shown that the transition matrix has been one of the most widely used methods to examine the level of crime specialization over time. These initial transition matrices, however, were quickly replaced by the FSC developed by Farrington (1986), which became the favored measure in specialization research and is still used by researchers today (Sullivan et al., 2009). The FSC was presented as an improvement over initial transition matrices because the FSC allowed adjusting for sample size as well as the frequency for each offense type analyzed (Farrington, Snyder, & Finnegan, 1988). Indeed, offense types that occur rarely (e.g., homicide) may have initially produced biased findings due to the small cell count. Compared to the previous transition matrix measures, the FSC reduces the effect of the frequency of the offense type by taking into account the total number of offenses for the cell of interest.

The FSC also provide information not only on the degree of specialization or versatility for each offense type but also on the specialization for one specific category of a variable. In other words, this coefficient indicates crime switching or specialization of the offender for one particular variable (behavior), which can be beneficial for the investigation of behavioral consistency. Moreover, the FSC allows for the investigation of the degree of specialization across consecutive offenses within different types (i.e., sequential specialization) and provides insight into which crimes are most often repeated. This coefficient thus indicates the extent of sequential specialization for each offense type by providing an aggregate measure across offenses rather than individuals. However, due to its aggregate nature, concerns were raised regarding its utility when investigating offending specialization at the individual level (Osgood & Schreck, 2007; Piquero, Paternoster, Brame, Mazerolle, & Dean, 1999; Sullivan et al., 2006).

Crime specialization and the diversity index (D index)

To address the issue of aggregate data and to allow the examination of individualized patterns of crime specialization, researchers have proposed using the D index. The D index was developed by Agresti and Agresti (1978) to measure species variation. In the last decade or so, however, this index has been increasingly used as a measure of offending versatility (e.g., Mazerolle, Brame, Paternoster, Piquero, & Dean, 2000; Miethe, Olson, & Mitchell, 2006; Piquero et al., 1999; Sullivan et al., 2006). The D index is a continuum where, on the one end, it represents complete specialization and, on the opposite end, it represents complete diversity. The D index has attracted a lot of attention from criminal career researchers because it allows investigators to examine the individual factors that may increase or decrease crime specialization (or crime versatility) (e.g., Mazerolle et al., 2000; Miethe et al., 2006; Piquero et al., 1999; Sullivan et al., 2006).

Another aspect of the D index is that it provides researchers with an opportunity to determine the level of crime specialization (or crime switching) by simultaneously considering all crimes committed by the individuals, regardless of the chronological order of the offenses. This is an important distinction from the FSC, which only provides an overview of specialization for two consecutive crimes. Therefore, while the D index allows the researcher to determine the overall level of specialization across an individual’s offending series, the FSC provides an estimate of specialization for specific parts of it. As a result, recent empirical studies have used both the FSC and the D index due to the complementary insight they provide about offending specialization (Lussier et al., 2008; Miethe et al., 2006; Piquero et al., 1999; Sullivan et al., 2009). It is thought that these aforementioned methods could improve knowledge on offending consistency and crime linkage.

Aim of the Study

Crime linkage analysis constitutes a tool to help investigators prioritize suspects, but a scarcity of research limits our knowledge on behavioral consistency in sexual assault cases. Furthermore, previous studies are characterized by methodological issues that could bias the results found. Despite this, prior studies have shown that geographical behaviors and environmental aspects of crime events are more consistent over time than those related to the offender’s modus operandi. The current study thus aims to address the limited knowledge in this field while attempting to address some of the methodological limitations of previous studies. In doing so, the current study draws on research from the criminal career literature as a way to improve the study of behavioral consistency in the context of crime linkage analysis. More specifically, criminal career research has identified several analytical strategies that can help estimate the level of behavioral consistency and that can be used for the analysis of geographic and environmental aspects of sexual assault cases. Taken together, the goal of this study is to determine the level of consistency of specific geographical behaviors and environmental factors among serial sex offenders using analytical strategies coming from both the behavioral consistency and criminal career fields.

Method

Participants

The initial sample for the study consisted of all male sex offenders convicted of a sentence of 2 years or more between 1995 and 2004 in one province of Canada. This list of more than 1,000 offenders was examined to identify all serial sex offenders of stranger victims. Ninety-two individuals matched the criteria, and 72 of these agreed to participate in the study. Together, these men were responsible for a total of 361 sexual assaults for which they were charged and convicted. The participants were all incarcerated in a Correctional Service of Canada penitentiary. The sample included individuals who had committed two or more sexual assaults involving a victim of any age and any gender with whom the offender had no personal relationship prior to the day the offense was committed. Offenders included in this study had sexually assaulted adult women (n = 33), children (n = 17), or both (n = 22), and 80.0% (n = 291) of the victims were female. The victim’s mean age was 18.7 years (SD = 9.6). The majority of the offenders were White (91.3%; n = 63), and the average age at the beginning of the crime series was 30.7 years (SD = 9.4). The participants had committed an average of 5 sexual crimes in their series (ranging from 2 to 37 sexual assaults each), and the average crime series’ length was 1,718 days (approximately 5 years).

Procedures

A specially constructed instrument was built to collect information from police investigative reports (present in the institution’s case file) and to guide in-depth semistructured interviews with the offenders. This questionnaire includes five sections that allow the collection of information on precrime factors, target selection processes, modus operandi, postcrime factors, and geographic behavior. The reliability of responses in our study was monitored by checking for and questioning inconsistencies. Self-reported information was then compared with official data (i.e., police reports). In the case of a discrepancy on factual information (e.g., location of the crime, day of the crime), information from the official police data was used to limit memory bias. To minimize response distortion, offenders were also promised complete anonymity and confidentiality and a guarantee that their information provided could not be used in any way against them by the Correctional Service of Canada. Interviews were conducted in a private office, isolated from correctional staff and other inmates. They lasted from 2 to 12 hr, depending on the number of crimes committed and the participants’ verbosity. Due to the sensitive nature of the conversations, permission was not requested to tape record the interviews, although extensive verbatim notes were taken whenever possible. No participant was paid for participating in the study.

Analytical Strategy

The current study compared the utility of three analytical strategies for the purpose of crime linkage: (a) Jaccard’s similarity coefficient, coming from the crime linkage and behavioral consistency literature and (b) the standardized D index and (c) the FSC, both coming from the criminal career literature. This procedure made it possible to determine if one of the strategies used was better to investigate offending consistency or if the information provided by each strategy was complementary to one another. Three steps were taken to examine offending consistency. First, the level of consistency across indicators within the geographic and the environmental domains were compared to determine if one was characterized by a higher level of consistency than the other. Second, the overall consistency of geographic behaviors and environmental aspects of the crime were determined. Third, the consistency of the environmental aspects at two different stages of the offense was investigated (i.e., victim encounter and victim release) to investigate if one of the offense stages was more reliable than the others for crime linkage purpose. These two stages were specifically selected as they represent the two most important locations for police investigations (Rossmo, 2000).

Jaccard’s coefficient

Jaccard’s (1901) coefficient was used as a similarity measure for each of the variables included in the study. This coefficient provides information on the degree of consistency across consecutive offenses for each behavior examined. This coefficient ranges from 0, indicating complete inconsistency, to 1, indicating complete behavioral consistency. Jaccard’s coefficient can be calculated using the following formula:

α (a + b + c),

where a equals the number of behaviors present in both crimes (1/1), and b and c equal the number of behaviors present in one crime but not the other (1/0 and 0/1).

Consistency was measured by comparing each offense with the subsequent offense for the full length of each offender’s series. The coefficient thus reflects whether two consecutive crimes are similar or not. However, for police investigation, the importance is to know if it is possible to rely on behaviors to link crimes together without knowledge of the offender’s prior criminal behavior. Indeed, the chronology of the offender’s crimes will usually be determined only after his arrest. It is thus important to rely on methods that can help to establish if the offenders are consistent in terms of their behaviors throughout their whole career (or series of crimes) rather than chronologically. This is where methods used in the criminal career literature become useful.

D index

The D index provides a probability that any two offenses randomly selected from an individual’s criminal history are in different categories. The D index theoretically ranges from 0 to 1, where 0 represents complete specialization—that is, all of the offender’s crimes were of the same type. As the value approaches 1, there is greater evidence of diversity in the types of offenses across an offender’s criminal career. For example, an index of 0.6 means that the probability is of 0.6 that two randomly selected offenses are of the same nature. This index is an individual-level measure and draws on offense proportions rather than sequence (i.e., successive offenses). The larger the number of categories and the more uniformly distributed the observations are over the categories, the higher the index tends to be (Agresti & Agresti, 1978). As such, the maximum value (indicating complete diversity) will vary according to the number of offense categories considered.

Considering the fluctuating nature of the D index and due to the different scaling of each variable included in the current study, the standardized D index (Agresti & Agresti, 1978) was used. The standardized D index can be computed using the following equation, in which p is the proportion of crimes committed for each of the i = 1, 2, … k categories identified:

I = \frac{1 - \sum_{i = 1}^{k} p_{i}^{2}}{1 - 1 / k}

Concerns have been previously raised regarding this standardized index and its validity (Agresti & Agresti, 1978). Some have suggested that it does not capture criminal diversity as well as the unstandardized index since it only measures the dispersion of the population among the categories, regardless of the number of categories.² In the context of behavioral consistency, however, it becomes essential to standardize this index if one wants to compare the consistency level of specific offending behaviors that include an uneven number of categories. By putting all of the variables on the same scale (0 to 1), it is possible to compare them among each other. In the same way, this allows for the three different coefficients used in the current study (all three on a 0 to 1 scale) to then be compared to one another.

FSC

This coefficient ranges from 0, indicating complete versatility (i.e., the frequencies in the diagonal cells are what one would expect by chance), to 1, indicating complete specialization (i.e., every “k + 1” offense is of the same type as the “k” offense; Sullivan et al., 2009). An indication of specialization is obtained by computing the FSC_jk for each of the diagonal cells of an offense transition matrix³ (where jk refers to a cell on the diagonal) using the following formula:

(O_{j k} - E_{j k}) / (R_{j} - E_{j k}),

where O_jk represents the observed number of cases in the diagonal cell of interest, E_jk is the number of cases in that cell expected by chance, and R_j is the total number of cases for the row of interest (Paternoster, Brame, Piquero, Mazerolle, & Dean, 1998; Sullivan et al., 2009). As the FSC focuses on the transition from one crime to the following one, rather than all of the crimes committed by the offender, a fixed number of crimes per offender was needed. The current study focuses on the crime transitions for the first four sexual crimes committed by the offender (i.e., three crime transitions) because as the number of crimes considered increases, fewer offenders could be included in the analyses. For instance, 72 offenders committed two crimes (Transition 1, from Crime 1 to Crime 2), 49 committed three crimes (Transition 2, from Crime 2 to Crime 3), and 27 committed at least four crimes (Transition 3, from Crime 3 to Crime 4). The Adjusted Standardized Residual (ASR) was also computed to test the statistical significance of the deviation of the observed number from the expected number for any given FSC (Farrington, 1986; Farrington et al., 1988). The ASR was computed using the following formula:

\frac{O - E}{S Q R T (E) \times S Q R T {[1 - (R / T)] [1 - (C / T)]}},

where O = the observed number, E = the expected number by chance, R = the row total, C = the column total, T = the grand total, and E = RC/T.

Variables

Two main dimensions of the offender’s sexual criminal activity were considered in the present study: (a) geographic behaviors and (b) environmental indicators. The frequency data for the 12 variables included in the study are presented in Table 1.

Table 1:

Frequencies for the Variables of the Geographic and Environmental Dimensions

Variables	Frequencies % (n)	Variables	Frequencies % (n)
Environmental Factors
Encounter land area		Victim release land area
Residential	61.8 (223)	Residential	68.4 (247)
Commercial	25.2 (91)	Commercial	15.5 (56)
Industrial	3.0 (11)	Industrial	2.2 (8)
Institutional	1.4 (5)	Institutional	1.7 (6)
Park	6.9 (25)	Park	4.2 (15)
Wilderness/rural	1.4 (5)	Wilderness/rural	8.0 (29)
Encounter location		Victim release location
Inside	52.9 (191)	Inside	55.7 (201)
Outside	47.1 (170)	Outside	44.3 (160)
Encounter site		Victim release site
Private	32.1 (116)	Private	50.1 (181)
Public	67.9 (245)	Public	49.9 (180)
Encounter site familiarity		Victim release site familiarity
Not familiar	0.0 (0)	Not familiar	8.9 (32)
Offender	16.6 (60)	Offender	39.9 (144)
Victim	21.6 (78)	Victim	24.1 (87)
Both	61.8 (223)	Both	27.1 (98)
Offense timing
Week	68.4 (247)
Weekend	18.3 (66)
Mixed	13.3 (48)
Geographic Behaviors
Attack method		Hunting style
Raptor	42.4 (153)	Hunter	54.8 (198)
Stalker	8.0 (29)	Poacher	5.5 (20)
Ambusher	49.6 (179)	Troller	18.8 (68)
		Trapper	20.8 (75)
Target’s mobility
One location	43.5 (157)
Multiple locations (two +)	56.5 (204)

Note. n = 361.

Geographic behaviors

Three variables associated with geographic-related behaviors committed by the offender during the offense were included in the study. The first two variables are based on the hunting pattern typology developed by Rossmo (1997), relevant to the identification of spatial patterns of serial predators: (a) the offender’s hunting style, referring to the victim search methods used by the offenders to commit their offenses (0 = hunter; 1 = poacher; 2 = troller; 3 = trapper) and (b) the offender’s attack method (0 = raptor; 1 = stalker; 2 = ambusher).⁴ The third variable represents the crime set location, or target’s mobility, which designates if the offense occurred all in one location or if the victim was moved to different locations (1 = one location; 2 = multiple locations) throughout the offense.

Environmental indicators

The environmental aspects of crime events refer to the “static” characteristics of the offense’s environment and includes nine variables associated with the physical, temporal, and contextual features of the offense. The variables included examine the offense’s environment for two stages of the offense: (a) the encounter with the victim and (b) the victim release. The nine variables are as follow: (a) offense timing (0 = week; 1 = weekend; 2 = mixed, week and weekend), (b) offense land area use for the two stages of the offense (i.e., encounter and victim release) (0 = residential area; 1 = commercial area; 2 = industrial area; 3 = institutional area; 4 = park; 5= wilderness/rural area). For the FSC analyses, this variable has been recoded into four categories for statistical power purposes (0 = residential area; 1 = commercial area; 2 = park/wilderness area; 3= other area): (b) offense location (0 = inside; 1 = outside) for the two stages of the offense; (d) types of site for the two stages of the offense, referring to whether the offense was committed on a private (e.g., home, backyard) or a public/semipublic site (e.g., park, business/shopping site, street) (0 = private; 1 = public/semipublic) and (e) offender and victim familiarity with the offense site (0 = not familiar to both of them; 1 = familiar to the offender; 2 = familiar to the victim; 3 = familiar to both the offender and the victim) for the two stages of the offense.

Results

Jaccard’s Coefficient

Jaccard’s coefficient for the geographic behaviors and the environmental aspects of the offense are presented in Table 2. Overall, the 72 offenders included in the study show a mean Jaccard’s coefficient of .77 (SD = .24), with the coefficients for the variables ranging from .68 to .86. At the dimension level, the geographic behaviors dimension shows a slightly higher Jaccard’s coefficient than the environmental dimension, with a mean of .80 (SD = .26) and .76 (SD = .25) respectively. For the geographic behaviors dimension, the attack method used by the offenders is the variable showing the highest Jaccard’s coefficient with a mean of .86 (SD = .31). In fact, this variable shows the highest Jaccard’s coefficient among all the 12 variables included in the study. For the environmental indicators dimension, the encounter site shows a slightly higher Jaccard’s coefficient compared to the other environmental indicators (Jaccard’s coefficient = .80, SD = .36). At the other end of the spectrum, victim release site familiarity is the indicator with the lowest Jaccard’s coefficient (M = .65, SD = .42).

Table 2:

Jaccard’s Coefficient for the Geographic Behaviors and the Environmental Dimensions by Number of Crimes Committed

	Total		Two Offenses		Three Offenses		Four+ Offenses		Ten+ Offenses
	(n = 72)		(n = 23)		(n = 22)		(n = 27)		(n = 8)
	M (SD)	Range^a	M (SD)	Range^a	M (SD)	Range^a	M (SD)	Range^a	M (SD)	Range^a
Geographic behaviors
Hunting style	.75 (.39)	.00 to 1.00	.74 (.45)	.00 to 1.00	.68 (.42)	.00 to 1.00	.83 (.30)	.00 to 1.00	.98 (.07)	.80 to 1.00
Attack method	.86 (.31)	.00 to 1.00	.87 (.34)	.00 to 1.00	.82 (.36)	.00 to 1.00	.87 (.25)	.00 to 1.00	.96 (.07)	.83 to 1.00
Target’s mobility	.79 (.34)	.00 to 1.00	.78 (.42)	.00 to 1.00	.68 (.36)	.00 to 1.00	.88 (.19)	.25 to 1.00	.99 (.02)	.94 to 1.00
Environmental
Offense timing	.79 (.35)	.00 to 1.00	.74 (.45)	.00 to 1.00	.77 (.34)	.00 to 1.00	.86 (.25)	.00 to 1.00	.98 (.07)	.80 to 1.00
Encounter land area	.78 (.36)	.00 to 1.00	.87 (.34)	.00 to 1.00	.57 (.42)	.00 to 1.00	.88 (.22)	.00 to 1.00	.92 (.14)	.60 to 1.00
Encounter location	.75 (.37)	.00 to 1.00	.70 (.47)	.00 to 1.00	.66 (.39)	.00 to 1.00	.88 (.19)	.25 to 1.00	.92 (.10)	.70 to 1.00
Encounter site	.80 (.36)	.00 to 1.00	.78 (.42)	.00 to 1.00	.68 (.39)	.00 to 1.00	.90 (.21)	.00 to 1.00	.98 (.03)	.93 to 1.00
Encounter site familiarity	.75 (.38)	.00 to 1.00	.70 (.47)	.00 to 1.00	.68 (.34)	.00 to 1.00	.85 (.25)	.00 to 1.00	.99 (.02)	.94 to 1.00
Victim release land area	.74 (.38)	.00 to 1.00	.65 (.49)	.00 to 1.00	.68 (.36)	.00 to 1.00	.86 (.25)	.00 to 1.00	.92 (.14)	.60 to 1.00
Victim release location	.76 (.34)	.00 to 1.00	.70 (.47)	.00 to 1.00	.68 (.29)	.00 to 1.00	.87 (.19)	.25 to 1.00	.93 (.11)	.70 to 1.00
Victim release site	.77 (.35)	.00 to 1.00	.74 (.45)	.00 to 1.00	.68 (.33)	.00 to 1.00	.88 (.23)	.00 to 1.00	.98 (.03)	.93 to 1.00
Victim release site familiarity	.68 (.42)	.00 to 1.00	.61 (.50)	.00 to 1.00	.61 (.41)	.00 to 1.00	.80 (.33)	.00 to 1.00	.98 (.04)	.89 to 1.00
Geographic behavior coefficient	.80 (.26)	.00 to 1.00	.80 (.30)	.00 to 1.00	.73 (.27)	.00 to 1.00	.86 (.20)	.25 to 1.00	.98 (.03)	.92 to 1.00
Environmental coefficient	.76 (.25)	.15 to 1.00	.72 (.28)	.22 to 1.00	.67 (.25)	.22 to 1.00	.86 (.19)	.15 to 1.00	.96 (.06)	.82 to 1.00
Individual total coefficient	.77 (.24)	.17 to 1.00	.74 (.27)	.17 to 1.00	.68 (.23)	.25 to 1.00	.86 (.18)	.25 to 1.00	.96 (.05)	.85 to 1.00

Note. n = 72.

Observed range.

For Jaccard’s coefficient analyses, four groups were created based on the number of sex crimes included in their series: (a) two offenses (n = 23), (b) three offenses (n = 22), (c) four or more offenses (n = 27), and (d) 10 or more offenses (n = 8). On the one hand, offenders who committed three sexual offenses tend to present lower Jaccard’s coefficients than the other groups of offenders. For example, the total mean Jaccard’s coefficient for offenders who committed three sex crimes is .68 (SD = .23), compared to .74 (SD = .27) for those who committed two offenses, .86 (SD = .18) for those who committed four sex offenses or more, or .77 (SD = .24) for the total sample (n = 72). On the other hand, more prolific offenders who committed at least four offenses (n = 27) generally show higher Jaccard’s coefficients compared to the other groups or to the total sample (M = .86, SD = .18; range for all variables = .80 to .90). In fact, almost all the variables used in the study show really high levels of consistency for this group of offenders, none of them showing significantly more or less consistency than the others. This is especially true for offenders having committed at least 10 offenses (M = .96, SD = .05; range for all variables = .91 to .99).

Finally, Jaccard’s coefficients for the geographic and environmental indicators by offense stages (i.e., encounter and victim release) were examined. Findings reported in Table 3 show that environmental indicators present relatively stable levels of consistency across the offense stages, with the encounter stage presenting a slightly higher coefficient (M = .77, SD = .27) than the victim release stage (M = .74, SD = .30). Relative consistency is also found when grouping by offense variables (i.e., land area, location, site, site familiarity), the coefficients varying between .78 (SD = .31) for the offense site and .72 (SD = .36) for the site familiarity.

Table 3:

Jaccard’s Coefficients, Standardized D Index, and Forward Specialization Coefficients by Offense Stages and Offense Variables

	Coefficient’s M (SD)
Variables	Jaccard’s Coefficient	D Index	FSC
Offense stages
Encounter	.77 (.27)	.25 (.25)	.67
Victim release	.74 (.30)	.30 (.29)	.63
Offense variables
Land area	.76 (.32)	.22 (.26)	.68
Residential	—	—	.71
Commercial	—	—	.70
Park/wilderness	—	—	.68
Others	—	—	.63
Location	.76 (.32)	.29 (.31)	.61
Inside	—	—	.67
Outside	—	—	.54
Type of site	.78 (.31)	.33 (.38)	.63
Private	—	—	.65
Public	—	—	.60
Site familiarity	.72 (.36)	.27 (.28)	.66
Not familiar	—	—	.77
Offender	—	—	.78
Victim	—	—	.55
Both	—	—	.59

Note. n = 72. D Index = diversity index; FSC = forward specialization coefficient.

D Index

Table 4 presents the D index of offending for the geographic behaviors and the environmental indicators of the offense. The unstandardized D index was also examined for comparison purposes. Although the standardized D indexes obtained were all of higher values than their unstandardized counterparts, the results and patterns discussed here were still similar. It is noteworthy, however, that the standard deviations of the standardized indexes were higher, indicating that a lot of variation exists among the offenders.

Table 4:

Standardized D Indexes for the Geographic Behaviors and the Environmental Dimensions by Number of Crimes Committed

	Total		Two Offenses		Three Offenses		Four+ Offenses		Ten+ Offenses
	(n = 72)		(n = 23)		(n = 22)		(n = 27)		(n = 8)
	M (SD)	Range^a	M (SD)	Range^a	M (SD)	Range^a	M (SD)	Range^a	M (SD)	Range^a
Geographic behaviors
Hunting style	.21 (.30)	.00 to .85	.17 (.30)	.00 to .67	.24 (.30)	.00 to .59	.22 (.31)	.00 to .85	.05 (.14)	.00 to .40
Attack method	.17 (.30)	.00 to .86	.10 (.26)	.00 to .75	.18 (.30)	.00 to .67	.23 (.33)	.00 to .86	.20 (.32)	.00 to .86
Target’s mobility	.33 (.43)	.00 to 1.00	.22 (.42)	.00 to 1.00	.44 (.46)	.00 to .89	.32 (.41)	.00 to 1.00	.22 (.42)	.00 to 1.00
Environmental
Offense timing	.20 (.31)	.00 to .75	.20 (.34)	.00 to .75	.24 (.33)	.00 to .67	.17 (.26)	.00 to .75	.03 (.09)	.00 to .25
Encounter land area	.21 (.28)	.00 to .80	.08 (.21)	.00 to .60	.34 (.30)	.00 to .80	.21 (.27)	.00 to .75	.24 (.28)	.00 to .59
Encounter location	.28 (.34)	.00 to .75	.23 (.35)	.00 to .75	.33 (.34)	.00 to .67	.28 (.33)	.00 to .75	.38 (.37)	.00 to .75
Encounter site	.30 (.43)	.00 to 1.00	.22 (.42)	.00 to 1.00	.40 (.45)	.00 to .89	.29 (.41)	.00 to 1.00	.25 (.41)	.00 to 1.00
Encounter site familiarity	.23 (.29)	.00 to .83	.20 (.31)	.00 to .67	.27 (.30)	.00 to .59	.22 (.28)	.00 to .83	.15 (.28)	.00 to .67
Victim release land area	.23 (.29)	.00 to .80	.21 (.29)	.00 to .60	.29 (.31)	.00 to .80	.21 (.29)	.00 to .79	.19 (.30)	.00 to .79
Victim release location	.29 (.34)	.00 to .75	.23 (.35)	.00 to .75	.39 (.34)	.00 to .67	.27 (.33)	.00 to .75	.29 (.36)	.00 to .75
Victim release site	.35 (.45)	.00 to 1.00	.26 (.45)	.00 to 1.00	.49 (.45)	.00 to .89	.32 (.43)	.00 to 1.00	.25 (.41)	.00 to 1.00
Victim release site familiarity	.31 (.34)	.00 to .96	.29 (.34)	.00 to .67	.36 (.35)	.00 to .89	.29 (.34)	.00 to .96	.18 (.27)	.00 to .67
Geographic behavior coefficient	.24 (.24)	.00 to .81	.16 (.24)	.00 to .81	.29 (.22)	.00 to .72	.26 (.26)	.00 to .77	.16 (.23)	.00 to .55
Environmental coefficient	.27 (.23)	.00 to .78	.21 (.21)	.00 to .60	.35 (.24)	.00 to .70	.25 (.23)	.00 to .78	.22 (.23)	.00 to .59
Individual total coefficient	.26 (.22)	.00 to .72	.20 (.21)	.00 to .65	.33 (.21)	.00 to .65	.25 (.23)	.00 to .72	.20 (.23)	.00 to .58

Note. n = 72.

Observed range.

Examining the individual mean D index, the 72 offenders included in the study show a relatively low level of diversity (M = .26, SD = .22). The maximum diversity value (1.00) is only obtained for 3 of the 12 variables included in the study (i.e., target’s mobility, encounter site, and victim release site). At the dimensional level, the geographic behaviors dimension shows a slightly lower D index (M = .24, SD = .24) than the environmental dimension (M = .27, SD = .23). For the geographic behaviors dimension, the attack method used by the offenders is the variable with the lowest D index with a mean of .17 (SD = .30). In fact, this variable shows the lowest D index of the 12 variables included in the study. On the other hand, the target’s mobility is the geographic behavior with the highest D index (M = .33, SD = .43). For the environmental indicators, the offense timing (M = .20, SD = .31) and the encounter land area (M = .21, SD = .28) are two variables with lower D indexes compared to the other environmental indicators examined. At the other end of the spectrum, victim release site is the indicator with the highest D index (M = .35, SD = .45).

Four groups were also created based on the number of sex crimes included in their series: (a) two offenses (n = 23), (b) three offenses (n = 22), (c) four or more offenses (n = 27), and (d) 10 or more offenses (n = 8). On the one hand, offenders who committed three sexual offenses tend to present higher D indexes than the other groups of offenders. More precisely, the total mean D index for offenders who committed three sex crimes is .33 (SD = .21), compared to .20 (SD = .21) for those who committed two offenses or .26 (SD = .22) for the total sample (n = 72). Despite this, the variables attack method (M = .18, SD = .30) and victim release site (M = .49, SD = .45) still show, respectively, the lowest and highest D index obtained among the 12 variables analyzed. On the other hand, more prolific offenders who committed at least 10 offenses (n = 8) generally show lower D indexes than the total sample with an individual total mean D index of .20 (SD = .23). More specifically, these offenders commit their sexual offenses consistently at the same time of the week (offense timing; M = .03, SD = .09), using consistently the same hunting style (M = .05, SD = .14), for all their offenses.

Finally, the D indexes for the geographic and environmental indicators by offense stages (i.e., encounter and victim release) were examined. Findings reported in Table 3 show that environmental indicators present a relatively stable level of consistency across the offense stages, with the encounter stage presenting a slightly lower mean D index (M = .25, SD = .25) than the victim release stage (M = .30, SD = .29). When examining the mean D indexes by offense variables (combining variables for the two stages analyzed—that is, land area use, location, site, site familiarity), the land area indicator shows the lowest D index (M = .22, SD = .26), whereas the offense site shows the highest D index (M = .33, SD = .38).

Forward Specialization Coefficient

Table 5 presents the FSCs for the geographic behaviors and the environmental indicators. The coefficients tend to show a somewhat low overall consistency level. It is also noteworthy that a lot of fluctuation exists among each variable across the three crime-transitions analyzed. Thus, there is a common trend for each of the categories of each variable to show consistency at one transition, while indicating a trend for inconsistency at another. For example, offenders using the hunter hunting style at Crime-Transition 1 (T₁) show high stability, most often using this style to commit their first and second crime (.86), while demonstrating lower consistency when committing their following crimes (T₂ = .67 and T₃ = .72). On the other hand, offenders using the poacher hunting style appear to use this style in a somewhat inconsistent manner for their first and second crime (T₁ = .39), while demonstrating perfect consistency at Crime-Transitions 2 and 3 (T₂ = 1.00, n = 3; T₃ = 1.00, n = 5).

Table 5:

FSC by Crime-Transition for the Geographic Behaviors and the Environmental Dimension

	T₁	T₂	T₃
Transition Variables	(n = 72)	(n = 49)	(n = 27)	Mean T₁ to T₃	Total Mean Variable	Dimension Mean
Geographic Behaviors
Hunting style					.73
Hunter to hunter	.86***	.67***	.72***	.75
Poacher to poacher	.39***	1.00***	1.00***	.80
Troller to troller	.36***	.48**	1.00**	.61
Trapper to trapper	.79***	.75***	.72***	.75		.70
Attack method					.72
Raptor to raptor	.73***	.63***	1.00***	.79
Stalker to stalker	.69***	1.00***	.28⁺	.66
Ambusher to ambusher	.71***	.67***	.74***	.71
Target’s mobility					.65
One to one	.51***	.56***	1.00***	.69
Multiple to multiple	.57***	.52***	.74***	.61
Environmental Indicators
Offense timing					.70
Week to week	.57***	.68***	.67**	.64
Weekend to weekend	.74***	.86***	.59***	.73
Mixed to mixed	.55***	.84***	.79***	.73
Encounter land area					.73
Residential to residential	.73***	.56***	.81**	.75
Commercial to commercial	.49***	.84***	.65***	.80
Park to park	.64***	.57***	.63**	.61
Others to others	.48***	1.00***	.64***	.75		.64
Encounter location					.61
Inside to inside	.47***	.51***	1.00***	.66
Outside to outside	.62***	.51***	.55***	.56
Encounter site familiarity					.67
Offender to offender	.88***	.76***	.81***	.82
Victim to victim	.51***	.48***	.71***	.57
Both offender/victim to both	.42***	.78***	.69***	.63
Victim release location					.60
Inside to inside	.58***	.47**	1.00***	.68
Outside to outside	.52***	.54**	.46***	.51

Note. FSC = forward specialization coefficient.

p < .10. *p < .05. **p < .01. ***p < .001.

At the dimensional level, the geographic behaviors, here again, show a higher mean FSC (M = .70) than the environmental indicators (M = .64). For the variables included under the geographic behaviors dimension, the hunting style (M = .73) and the attack method (M = .72) used by the offenders show the highest FSC across transitions, whereas the target’s mobility shows the lowest coefficient (M = .65). More specifically, for the hunting style, the offenders using the poacher style show the highest FSC (M = .80), whereas the troller style is characterized by the lowest FSC (M = .61). As for the attack method, the FSC indexes show that offenders using the raptor attack method are more consistent (M = .79) than the offenders using one of the other two attack methods, the ambusher (M = .71) and the stalker (M = .66).

For the environmental indicators dimension, the encounter land area (M = .73) and the offense timing (M = .70) show the highest FSC across transitions. More specifically, offenders using a commercial area to encounter their victim show higher consistency (M = .80) than offenders using a residential area (M = .75), a park (M = .61), or another land area (M = .75). In addition, offenders committing their offenses during the week show a lower FSC (M = .64) than those committing their offenses on the weekend (M = .73) or during the week and the weekend (M = .73). All the other environmental indicators analyzed show rather low FSC means ranging from .60 to .67. It is noteworthy, however, that offenders encountering their victim(s) on a site that is familiar to them show the highest FSC obtained across transitions (M = .82).

Finally, the FSCs for the geographic and environmental indicators by offense stages (i.e., encounter and victim release) were examined. Findings reported in Table 3 show that environmental indicators present relatively low levels of consistency across the offense stages, with the encounter stage still presenting a slightly higher FSC (M = .67) than the victim release stage (M = .63). Also, somewhat low consistency is found when grouping by offense variables (combining variables for the two stages analyzed—that is, land area, location, site, site familiarity), the coefficients varying between .61 for the offense location and .68 for the offense land area.

Discussion and Conclusion

The current study aimed to address the limited knowledge in the crime linkage and behavioral consistency fields while attempting to address some of the methodological limitations of previous studies. Using multiple analytical strategies coming from different fields of research, the current study identified geographic behaviors and environmental factors that appeared promising to rely on for the purpose of crime linkage of sexual offenses. The use of different analytical methods to examine behavioral consistency in sexual offenses, two of them coming from the criminal career field, also provided a new and interesting way to analyze offending consistency while allowing us to better understand the impact of the analytic method used on the results found.

Geographic and Environmental Consistency

First, results found with the three methods show that the geographic behaviors are characterized by a somewhat higher stability and consistency than the environmental variables included in the study. Second, the three methods used also help to highlight specific variables showing higher or lower behavioral consistency in sexual offenses of serial rapists. Among the geographic behaviors investigated, results show that the attack method is used in a consistent manner (sequentially or overall) by serial sex offenders. More specifically, looking at results for the FSC, it appears that offenders using the raptor attack method are more consistent throughout their series of offenses, followed by those using the ambusher method. The raptor, on the one hand, attacks almost immediately after having encountered his victim (Rossmo, 1997). The ambusher attack method, on the other hand, refers to actions taking place in locations where the offender is familiar and feels in control (e.g., residence, workplace; Rossmo, 1997). These two offending types might represent two opposite ends of a spectrum where on the one end the offender is careful about the location of the offense and where to lure his victim (i.e., the ambusher), whereas at the other end such considerations are less relevant than the presence of a suitable target (i.e., raptor). For crime linkage purposes, these two methods (if used by the offender to commit an offense under investigation) could help in suspect prioritization, as chances are that the offender has used the same method previously.

On the other hand, the target’s mobility appears to be the geographic behavior showing the lowest consistency level among serial sex offenders. In other words, sex offenders are less likely to follow a consistent pattern or modus operandi when it comes to moving (or not moving) the victim from one place to the other throughout the offense (encounter → attack → crime → victim release). Among the environmental factors investigated, the offense timing appears to be a promising factor to rely on for crime linkage. Hence, with both methods used in the criminal career literature (i.e., D index and FSC), results show that sex offenders are consistently committing their crimes at the same time of the week. More specifically, offenders who commit their offenses on weekdays are less likely to commit their subsequent offenses at the same time.

Third, the encounter stage shows a tendency for higher (sequential) consistency than the victim release stage. Within the encounter stage, the type of land area, the type of site, the location, and the offender/victim’s familiarity with the site are analyzed (combined). Among these, the encounter land area shows the highest consistency. This is not to say that the encounter land area is significantly different from the other aspects of the encounter stage but that this aspect might be more promising with respect to crime linkage. This is an important finding as the literature on geographic profiling suggests that the encounter location where the offenders first “contact” their victim is often close to where they live (Canter & Larkin, 1993; A. Davies & Dale, 1996; Rossmo, 1997). Investigators and crime analysts could thus use the encounter site characteristics as a starting point to narrow down the pool of potential suspects, knowing that the offender responsible for the offense under investigation most likely lives close by and will also most likely have encountered his previous victims in a similar environment. The relative stability of the encounter site and the fact that offenders tend to target victims that are close by are two potentially important factors that should be analyzed simultaneously in future studies.

Finally, Jaccard’s coefficient and D index values obtained show interesting results suggesting that the relationship between the number of crimes committed and offending consistency might not be linear. More specifically, findings indicate that offenders who committed three sexual offenses tend to show somewhat more overall versatility in their offending as opposed to those who committed less than three or four or more offenses. Moreover, the most prolific offenders of the study sample are characterized by high consistency. Previous studies, mostly in the criminal career literature, have found that it was usually the other way around, a higher level of versatility found for persistent offenders (Grubin et al., 2001; Heil, Ahlmeyer, & Simons, 2003; Smallbone & Wortley 2004; Sullivan et al., 2009). It is unclear, however, how many prolific offenders were included in these studies.

In line with a more traditional view of criminal career (Blumstein et al., 1988) and more recent studies in the behavioral consistency field (Harbers et al., 2012; Sorochinski & Salfati, 2010), this result could suggest that offenders having committed only three offenses are still evaluating the different and most successful ways, locations, and moments to commit their crimes, whereas the persistent offenders have achieved a certain offending knowledge and expertise. This result is also consistent with the learning process hypothesis coming from the personality psychology literature suggesting that most offenders will try something different (i.e., committing a crime in a different way) after the first offense but, by their third offense, they will determine what strategy works best for successfully achieving their goal (Sorochinski & Salfati, 2010). However, it is important to keep in mind the small sample size for the current study, specifically for the more prolific offenders (n = 8). Future analyses should further examine the impact of the number of crimes committed by the offender by specifically analyzing this variable in a continuous rather than categorical form.

Methodological Considerations and Results

Although the methods used in the study are not nested and therefore not fully comparable, all three approaches reveal that some degree of geographic behaviors and environmental consistency is present. More specifically, Jaccard’s coefficients and the D indexes show very similar results indicating strong overall and variable-specific tendency for consistency. For the FSC, however, although still moderately high, values found are somewhat closer to the versatile offending end of the spectrum. Moreover, for this coefficient, it appears easier to highlight variables or categories characterized by higher versatility rather than higher consistency values. It is noteworthy, however, that the FSC has more conceptual and measurement constraints than the D index or Jaccard’s coefficient. For example, the FSC is based on hierarchical and sequential offending (rather than proportions) and only reflects offending consistency patterns for the first four crimes committed by the offenders (rather than the overall sexual “criminal career”). These constraints are thus thought to provide the FSC less freedom to uncover patterns of consistency in the data and can account for the somewhat divergent results found between the FSC and the other methods used (e.g., behaviors and environmental factors used by the offender more consistently with the FSC are not necessarily the same as those highlighted for Jaccard’s coefficient and D index).

As found by previous studies (Grubin et al, 2001; Woodhams & Labuschagne, 2011), greater levels of consistency are found in offenses committed later in the series investigated. Researchers need to be aware that by using a method that restricts the number of crimes analyzed per series, a greater or lower consistency will be found. In accordance with the criminal career paradigm, this result shows the importance of selecting all crimes of the same nature in the series instead of a specific number of crimes per offender, which will permit a more realistic understanding of offending consistency. In the same way, the somewhat lower consistency level found using the FSC might also be an indication that, as found by Grubin et al. (2001), the time span between the crimes of the series investigated needs to be analyzed and controlled for as it might influence the consistency observed. Future analyses should further look at the impact of the number of crimes committed and the length of the crime series.

Interestingly, similar results are found using Jaccard’s coefficient and the D index. For both of these methods, all of the crimes that were part of the series are analyzed for consistency. However, Jaccard’s coefficient takes into consideration the sequence of the offense, whereas the D index does not. The similar results found by both of these methods thus suggest that the chronological order of the crimes might not be that important when investigating the offending consistency of sex offenders, as long as all of the crimes of the offender’s series are included.

Overall, results obtained by the three methods suggest that there is no clear indication that one method is better than the other. As suggested by Sullivan et al. (2009), it is thought that the best course of action for knowledge building is to use multiple methods; each distinct approach reveals a different aspect of the question under investigation and provides depth and comprehensive understanding of offending consistency. However, methods used prominently in the criminal career literature (i.e., D index and FSC) help to provide a new and interesting way to examine offending consistency for the purpose of crime linkage. More specifically, the FSC provides a more in-depth and detailed analysis of consistency by taking into account each category of the variables, something not achieved by the consistency/similarity coefficients used thus far in the crime linkage literature (e.g., Jaccard’s coefficient). In the same way, it is noteworthy that the results found using Jaccard’s coefficient and the FSC separately are found when using the D index. The D index might then be a more integrative and complete measure to use when investigating behavioral consistency.

Having considered the main findings of the current study and their implications, the limitations of the current study also need to be reviewed. First, the sample only includes crimes committed by incarcerated stranger sex offenders for which the offenders were charged and convicted. Therefore, the results of the current study might reflect behaviors of offenders who were not able to avoid detection and were apprehended by the police. Second, the study is based on a relatively small sample of serial sex offenders that precluded the use of more sophisticated statistical analyses. Although the sample is relatively small, it is important to stress that it is composed of a very specific group of offenders having committed a relatively rare type of offense (i.e., serial stranger rapists). Different results might have been found if the sample had been composed of more offenders as well as different types of sex offenders.

Despite these limitations, the current study provides complementary methods to measure behavioral consistency and shows the importance of the analytical strategy used on the results found. The notion of consistency in the crime linkage literature has typically been applied to crime scene behavior rather than to the geographical aspect of the offense (Grubin et al., 2001). Moreover, when taken into account, studies have usually looked at the journey to crime (e.g., Snook et al., 2006) rather than broader elements of the environment where an offense takes place. Little research exists on the extent or nature of serial offenders’ consistency in geographic behaviors and environmental and offense site selection. This lack of empirical research is especially problematic for the investigation of sexual crimes between strangers where crime linkage becomes a necessity as no connection between the victim and the offender can help to identify potential suspects. The current study is thought to have filled some gaps left by previous studies and to have provided evidence of the importance and reliability of geographic behaviors and environmental factors for successful crime linkage of stranger sexual offenders. Future studies should further investigate the stability and reliability of these factors. More specifically, additional research needs to be conducted to investigate the heterogeneity and stability of sexual offense environments and crime site selection considering that they are less prone to situational influence. It is also believed that future studies should rely more closely on the rich and well-established criminal career literature and methodology to improve current practices and knowledge on offending consistency and crime linkage.

Footnotes

Authors’ Note:

The lead author wishes to thank the Vanier Canada Graduate Scholarships program, administered by the Social Sciences and Humanities Research Council of Canada, for its financial support. The authors also wish to thank the anonymous reviewers for their comments on an earlier version of this article. The views expressed are those of the authors and are not necessarily those of the Correctional Service of Canada. An earlier draft of this article was presented at the 2011 American Society of Criminology Meeting in Washington, DC.

Notes

Nadine Deslauriers-Varin is an assistant professor of criminology at Université Laval in Canada. Her research interests include police investigations, offending consistency, offending process, modus operandi, and police interrogations. Her work has also been published in Justice Quarterly, Police Quarterly, and Sexual Abuse: A Journal of Research and Treatment.

Eric Beauregard is an associate professor in the School of Criminology at Simon Fraser University, and he is the director of the Centre for Research on Sexual Violence. He has specific research interests in understanding the offending process of sex offenders. His research has appeared in Sexual Abuse: A Journal of Research and Treatment, Justice Quarterly, Journal of Criminal Justice, Violence and Victims, Journal of Research in Crime and Delinquency, Police Quarterly, and Journal of Interpersonal Violence.

References

Agresti

B. F.

(1978). Statistical analysis of qualitative variation. Sociological Methodology, 9, 204-237.

Alison

Goodwill

Almond

van den Heuvel

Winter

(2010). Pragmatic solutions to offender profiling and behavioral investigative advice. Legal and Criminological Psychology, 15, 115-132.

Bateman

Salfati

(2007). An examination of behavioral consistency using individual behaviors or groups of behaviors in serial homicide. Behavioral Sciences & the Law, 25, 527-544.

Beauregard

Proulx

Rossmo

Leclerc

Allaire

J.- F.

(2007). Script analysis of hunting process in serial sex offenders. Criminal Justice and Behavior, 34, 1069-1084.

Bennell

Canter

(2002). Linking commercial burglaries by modus operandi: Tests using regression and ROC analysis. Science and Justice, 42, 153-164.

Bennell

Jones

N. J.

(2005). Between a ROC and a hard place: A method for linking serial burglaries by modus operandi. Journal of Investigative Psychology and Offender Profiling, 2, 23-41.

Bernasco

(2008). Them again? Same-offender involvement in repeat and near repeat burglaries. European Journal of Criminology, 5, 411-431.

Blumstein

Cohen

Farrington

D. P.

(1988). Criminal career research: Its value for criminology. Criminology, 26, 1-35.

Blumstein

Cohen

Roth

J. A.

Visher

(1986). Criminal careers and “career criminals” (2 vols.). Washington, DC: National Academy Press.

10.

Brennan

Mednick

John

(1989). Specialization in violence: Evidence of a criminal subgroup. Criminology, 27, 437-453.

11.

Burrell

Bull

Bond

J. W.

(2012). Linking personal robbery offenses using offender behaviour. Journal of Investigative Psychology and Offender Profiling, 9, 201-222.

12.

Canter

D. V.

Larkin

(1993). The environmental range of serial rapists. Journal of Environmental Psychology, 13, 63-69.

13.

Davies

(1992). Rapist’s behaviour: A three-aspect model as a basis for analysis and the identification of serial crime. Forensic Science International, 55, 173-194.

14.

Davies

Dale

(1996). Locating the stranger rapist. Medicine, Science and the Law, 6, 146-156.

15.

Davies

Tonkin

Bull

Bond

J. W.

(2012). The course of case linkage never did run smooth: A new investigation to tackle the behavioural changes in serial car theft. Journal of Investigative Psychology and Offender Profiling, 9, 274-295.

16.

DeLisi

Beaver

K. M.

Wright

K. A.

Wright

J. P.

Vaughn

M. G.

Trulson

C. R.

(2011). Criminal specialization revisited: A simultaneous quantile regression approach. American Journal of Criminal Justice, 36, 73-92.

17.

Deslauriers-Varin

Beauregard

(2010). Victims’ routine activities and sex offenders’ target selection scripts: A latent class analysis. Sexual Abuse: A Journal of Research and Treatment, 22, 315-342.

18.

Farrington

D. P.

(1986). Age and crime. In Tonry

Morris

(Eds.), Crime and justice (Vol. 7, pp. 189-250). Chicago: University of Chicago Press.

19.

Farrington

D. P.

Snyder

H. N.

Finnegan

T. A.

(1988). Specialization in juvenile court careers. Criminology, 26, 461-485.

20.

Francis

Soothill

Fligelstone

(2004). Identifying patterns and pathways of offending behaviour: A new approach to typologies of crime. European Journal of Criminology, 1, 48-87.

21.

Goodwill

A. M.

Alison

L. J.

(2006). The development of a filter model for prioritizing suspects in burglary offenses. Psychology, Crime & Law, 12, 395-416.

22.

Grubin

Kelly

Brunsdon

(2001). Linking serious sexual assaults through behavior. London: Home Office, Research, Development and Statistics Directorate.

23.

Harbers

Deslauriers-Varin

Beauregard

van der Kemp

J. J.

(2012). Testing the behavioural and environmental consistency of serial sex offenders: A signature approach. Journal of Investigative Psychology and Offender Profiling, 9, 259-273.

24.

Harris

D. A.

Mazerolle

Knight

R. A.

(2009). Understanding male sexual offending: A comparison of general and specialist theories. Criminal Justice and Behavior, 36, 1051-1069.

25.

Heil

Ahlmeyer

Simons

(2003). Crossover sexual offenses. Sexual Abuse: A Journal of Research and Treatment, 15, 221-236.

26.

Jaccard

(1901). Étude comparative de la distribution florale dans une portion des Alpes et des Jura (A comparative study of the floral distribution in Alps and Jura). Bulletin de la Société Vaudoise des Sciences Naturelles, 37, 547-579.

27.

Leclerc

Lussier

Deslauriers-Varin

(in press). Offending patterns over time: An examination of specialization, escalation and de-escalation in the commission of sexual offenses. In Blokland

Lussier

(Eds.), Sex offenders: A criminal career approach. Oxford, UK: Wiley.

28.

Lundrigan

Czarnomski

Wilson

(2010). Spatial and environmental consistency in serial sexual assault. Journal of Investigative Psychology and Offender Profiling, 7, 15-30.

29.

Lussier

Leclerc

Healey

Proulx

(2008). Generality of deviance and predation: Crime-switching and specialization patterns in persistent sexual offenders. In Delisi

Conis

(Eds.), Violent offenders: Theory, research, public policy, and practice (pp. 97-118). Sudbury, MA: Jones and Bartlett Publishers.

30.

Markson

Woodhams

Bond

J. W.

(2010). Linking serial residential burglary: Comparing the utility of modus operandi behaviours, geographical proximity, and temporal proximity. Journal of Investigative Psychology and Offender Profiling, 7, 91-107.

31.

Mazerolle

Brame

Paternoster

Piquero

A. R.

Dean

(2000). Onset age, persistence, and offending versatility: Comparisons across gender. Criminology, 38, 1143-1172.

32.

McGloin

J. M.

Sullivan

C. J.

Piquero

A. R.

(2009). Aggregating to versatility? Transitions among offender types in the short term. British Journal of Criminology, 49, 243-264.

33.

Melnyk

Bennell

Gauthier

D. J.

Gauthier

(2011). Another look at across-crime similarity coefficients for use in behavioural linkage analysis: An attempt to replicate Woodhams, Grant, and Price (2007). Psychology, Crime & Law, 17, 359-380.

34.

Miethe

Olson

Mitchell

(2006). Specialization and persistence in the arrest histories of sex offenders: A comparative analysis of alternate measures and offense types. Journal of Research in Crime and Delinquency, 43, 204-229.

35.

Nieuwbeerta

Blokland

A. A. J.

Piquero

A. R.

Sweeten

(2011). A life-course analysis of offense specialization across age: Introducing a new method for studying individual specialization over the life course. Crime & Delinquency, 57, 3-28.

36.

Osgood

D. W.

Schreck

C. J.

(2007). A new method for studying the extent, stability, and predictors of individual specialization in violence. Criminology, 45, 273-312.

37.

Paternoster

Brame

Piquero

A. R.

Mazerolle

Dean

C. W.

(1998). The forward specialization coefficient: Distributional properties and subgroup differences. Journal of Quantitative Criminology, 14, 133-154.

38.

Piquero

A. R.

Paternoster

Brame

Mazerolle

Dean

C. W.

(1999). Onset age and specialization in offending behavior. Journal of Research in Crime and Delinquency, 36, 235-274.

39.

Rossmo

D. K.

(1997). Geographic profiling. In Jackson

J. L.

Bekerian

D. A.

(Eds.), Offender profiling: Theory, research and practice (pp. 159-175). Chichester, UK: John Wiley & Sons.

40.

Rossmo

D. K.

(2000). Geographic profiling. Boca Raton, FL: CRC Press.

41.

Salfati

G. C.

Bateman

A. C.

(2005). Serial homicide: An investigation of behavioral consistency. Journal of Investigative Psychology and Offender Proﬁling, 2, 121-144.

42.

Santtila

Fritzon

Tamelander

A. L.

(2004). Linking arson incidents on the basis of crime scene behaviour. Journal of Police and Criminal Psychology, 19, 1-16.

43.

Santtila

Junkkila

Sandnabba

N. K.

(2005). Behavioural linking of stranger rapes. Journal of Investigative Psychology and Offender Profiling, 2, 87-103.

44.

Santtila

Pakkanen

Zappal

Bosco

Valkama

Mokros

(2008). Behavioural crime linking in serial homicide. Psychology, Crime & Law, 14, 245-265.

45.

Schlesinger

L. B.

Kassen

Mesa

V. B.

Pinizzotto

A. J.

(2010). Ritual and signature in serial sexual homicide. Journal of the American Academy and Psychiatry Law, 38, 239-246.

46.

Sjöstedt

Långström

Sturidsson

Grann

(2004). Stability of modus operandi in sexual offending. Criminal Justice and Behavior, 31, 609-623.

47.

Smallbone

S. W.

Wortley

R. K.

(2004). Criminal diversity and paraphilic interests among adult males convicted of sexual offenses against children. International Journal of Offender Therapy and Comparative Criminology, 48, 175-188.

48.

Snook

Wright

House

J. C.

Alison

L. J.

(2006). Searching for a needle in a needle stack: Combining criminal careers and journey-to-crime research for criminal suspect prioritization. Police Practice and Research, 7, 217-230.

49.

Soothill

Francis

Sanderson

Ackerley

(2000). Sex offenders: Specialist, generalists—or both? A 32-year criminological study. British Journal of Criminology, 40, 56-67.

50.

Sorochinski

Salfati

G. C.

(2010). The consistency of inconsistency in serial homicide: Patterns of behavioural change across series. Journal of Investigative Psychology and Offender Profiling, 7, 109-136.

51.

Sullivan

C. J.

McGloin

J. M.

Pratt

T. C.

Piquero

A. R.

(2006). Rethinking the “norm” of offender generality: Investigating specialization in the short-term. Criminology, 44, 205-240.

52.

Sullivan

C. J.

McGloin

J. M.

Ray

J. V.

Caudy

M. S.

(2009). Detecting specialization in offending: Comparing analytic approaches. Journal of Quantitative Criminology, 25, 419-441.

53.

Tonkin

Grant

Bond

J. W.

(2008). To link or not to link: A test of the case linkage principles using serial car theft data. Journal of Investigative Psychology and Offender Profiling, 5, 59-77.

54.

Tonkin

Woodhams

Bull

Bond

J. W.

Palmer

E. J.

(2011). Linking different types of crimes using geographical and temporal proximity. Criminal Justice and Behavior, 38, 1069-1088.

55.

Woodhams

Grant

T. D.

Price

A. R. G.

(2007). From marine ecology to crime analysis: Improving the detection of serial sexual offenses using a taxonomic similarity measure. Journal of Investigative Psychology and Offender Profiling, 4, 17-27.

56.

Woodhams

Hollin

C. R.

Bull

(2007). The psychology of linking crimes: A review of the evidence. Legal and Criminological Psychology, 12, 233-249.

57.

Woodhams

Hollin

C. R.

Bull

(2008). Incorporating context in linking crimes: An exploratory study of situational similarity and if-then contingencies. Journal of Investigative Psychology and Offender Profiling, 5, 1-23.

58.

Woodhams

Labuschagne

(2011). A test of case linkage principles with solved and unsolved serial rapes. Journal of Police and Criminal Psychology, 27, 85-98.

59.

Woodhams

Toye

(2007). An empirical test of the assumptions of case linkage and offender profiling with serial commercial robberies. Psychology, Public Policy, and Law, 13, 59-85.

60.

Yokota

Canter

(2004). Burglars’ specialization: Development of a thematic approach in investigative psychology. Behaviormetrika, 2, 153-167.