For thousands of years inferences of a person's characteristics from their thoughts, motives, and actions have been made (Canter & Alison, 1997) although the application of these inferences to criminal investigations was first done in the 1970's by the FBI (Egger, 1999). Offender profiling, now known as behavioural investigative advice, is the process of providing inferences of likely offender characteristics based on crime scene information (Canter & Alison, 1999).
Case linkage is now more commonly known as linkage analysis which is the process of identifying whether or not a series of crimes have been committed by the same offender using crime scene behaviours (Hazelwood & Warren, 2003). Case linkage produces similar fact evidence in legal cases so that offenders can be charged with numerous offences (Ormerod, 1999). Case linkage can also allow all relevant crime scene information to be pooled thus making the use of investigative resources more efficient (Grubin, Kelly, & Burnsden, 2001).
The practices of case linkage and offender profiling have overlapping as well as separate assumptions. Case linkage and offender profiling both use theories from personality psychology and the assumption underlying the practices of both offender profiling and case linkage is consistency (Woodhams & Toye, 2007). Regarding offender profiling, individuals need to be consistent in their actions for there to be similarity associations between an individual's behaviour and characteristics (Mokros & Alison, 2002). In reference to case linkage, this consistency refers to the offender exhibiting similar behaviours each time he or she offends (Canter, 1995; Grubin et al., 2001; Woodhams, Hollin, & Bull, 2007). If offenders were not consistent in their offending behaviour, then it would not be possible to determine which single offender committed the crimes on the basis of their behavioural similarity. An additional assumption underlying the practice of offender profiling is homology (Alison, Bennell, Mokros, & Ormerod, 2002). This assumption states that offenders who commit crimes in a similar style will have similar background characteristics (Woodhams & Toye, 2007). Therefore, offenders who have different crime scene behaviour will have different characteristics. According to Woodhams and Toye (2007) an additional assumption underlying the practice of case linkage is behavioural distinctiveness. This assumption refers to the actions of an individual serial offender being different to other offenders committing similar crimes.
This paper will now discuss the evidence for the assumptions underlying the practices of case linkage and offender profiling. Then methods to combat the problems of the assumptions underlying the practices of case linkage and offender profiling will be presented. Finally, a conclusion will be reached as to whether there is enough evidence to support the assumptions underlying the practices of case linkage and offender profiling to produce valid and useful results.
Case linkage and offender profiling: Behavioural consistency
Behavioural consistency is an assumption that underlies the practices of case linkage and offender profiling. Psychological research has defined behavioural consistency as the repetition of particular aspects of behaviour if the same offender engages in the same type of offence again (Canter, 1995). Conversely, criminological research has defined behavioural consistency as the probability that an individual will repeatedly commit similar types of offences (Farrington, 1997). With regard to case linkage, it refers to the hypothesis that people are consistent in their behaviour across situations and the consistency assumption of offender profiling implies that the behaviour of one offender must be more consistent than the behaviour between all other offenders (Woodhams & Toye, 2007).
Behavioural consistency has been investigated by personality psychology studies and the hypothesis has been confirmed that people produce similar behaviour when they are faced with situations with similar psychological meaning to them (Mischel & Shoda, 1995). There is also supporting evidence that offenders commit crime in a consistent manner. For example, Farrington and Lambert (1997) showed that offenders are consistent in the choice of crime type. Furthermore, offenders have been shown to be consistent in their behaviour across robbery (Woodhams & Toye, 2007), arson (Santtila, Fritzon, & Tamelander, 2005), burglary (Bennell & Canter, 2002), homicide (Salfati & Bateman, 2005), and sexual assault (Grubin et al., 2001).
Therefore, there is evidence of the consistency assumption that underlies the practices of offender profiling and case linkage.
Case linkage: Behavioural consistency and distinctiveness
A necessity in proving consistency for investigative purposes is that an individual's behavioural variation must be less than others to be investigatively useful which Woodhams and Toye (2007) named 'distinctiveness'. An example of this behavioural consistency is that Jack the offender always robs a bank at gun point which is 100 percent consistent. Although if the majority of other bank robbers also rob banks at gun point then it cannot be concluded that Jack has committed the burglary in question. This consistency can be tested by predicting whether two crimes are committed by the same offender (linked) or by two different offenders (unlinked). Grubin et al. (2001) combined behavioural consistency and distinctiveness and found that linked crime and unlinked crimes are able to be reliably distinguished using behavioural consistency and distinctiveness theory. Using this approach, research with robberies and burglaries by Bennell and Jones (2005), have also shown that linkage status can be accurately predicted.
Cervone and Shoda (1999) have researched personality psychology and found that people consistently exhibit individual differences on some personality traits in similar situations which provides support of case linkage (Bennell & Canter, 2002). Although a review by Biesanz, West, and Kwok (2003) showed that personality psychologists and investigative psychologists examine consistency and distinctiveness differently. Shoda (1999) states that personality psychologists see individuals as demonstrating similar trait-related behaviour in different situations and although people's behaviour does change in different situations they remain relatively stable compared to other people. For example, if person B is more aggressive in one situation than person A, then in different situations, person B will always exhibit more aggressive behaviour than person A although they both exhibit different levels of aggression in different situations. Although personality psychologists are able to observe behaviour in highly controlled experimental conditions, the same cannot be done by investigative psychologists. Instead they use behavioural signatures, modus operandi, or psychological themes to determine the support of the behavioural distinctiveness and consistency assumptions which will be discussed next.
Hazelwood and Warren (2003) state that signatures are ritualised actions that are not necessary to commit or conceal the crime but are unique fantasies and drives of the offender to commit the crimes. These actions aid case linkage as they will be consistently exhibited in all of an offender's crimes (Hazelwood & Warren, 2003). There is some evidence for the existence of signatures in serial homicide and rape (Hazelwood & Warren, 2003), although there is no empirical evidence. One reason for this is that signatures may not be identified at crime scenes due to victim reactions or other situational factors (Douglas & Munn, 1992). Furthermore, Labuschagne (2006) states that foraging animals and body decomposition may disrupt the crime scene. Finally, it is difficult to separate the offender's behavioural signature from the modus operandi (Hazelwood & Warren, 2003).
From the evidence presented above, behavioural signatures may only exist in particular types of relatively rare crimes such as fantasy-driven, interpersonal, and serial violent crimes in specific situations. Therefore there is not enough evidence to support the use of behavioural signatures to determine behavioural consistency and distinctiveness as part of case linkage for all crimes.
Modus Operandi (MO)
Douglas and Munn (1992) state that the MO of offenders are the behaviours that an offender exhibits to commit the crime. Although greatly used by police in the past (Douglas & Munn, 1992), MO's can change over the course of a crime series (Hazelwood & Warren, 2003) due to factors such as unavailability of victims, victim reactions, interruptions, education, experience, and maturation (Hazelwood & Warren, 2003).
Despite the above findings, depending on the MO behaviour being observed there are sufficient levels of distinctiveness and consistency to make case linkage possible beyond chance levels e.g. rape (Grubin et al., 2001) and commercial robbery (Woodhams & Toye, 2007). Bennell and Jones (2005) found that although an analysis of property stolen during serial burglaries had a moderate level of linking accuracy, an analysis of distances between crimes produced much higher accuracy scores. The authors found that the closer together the burglaries, the more likely that they were committed by the same offender.
Bennell and Canter (2002) have found that MO's which are brought to the scene by the offender such as where the offence took place (offender-driven) versus what was stolen (situation-driven), are the most stable and distinct. Seeing as some behaviours are more consistent and are therefore more accurate at predicting case linkage allows analysts to be more efficient in only using the most consistent behaviours and also more accurate at predicting linkage (Woodhams & Toye, 2007).
Consistency can found in an offender's crimes despite differences in MO as when using the thematic approach consistency is defined at a more general level related to the function of the behaviour as opposed to discrete behaviours. For example, Salfati and Bateman (2005) state that when an offender cuts or tortures the victim, this behaviour suggests an expressive theme and the murder was part of an emotionally fuelled assault. Whereas when the victim is sexually assaulted or the body is posed, this behaviour demonstrates an instrumental theme and the aim of the murder was to achieve sexual gratification or something else beyond the murder.
The research into consistency and distinctiveness regarding psychological themes is promising (Salfati & Bateman, 2005) although when more stringent criteria to assign crimes to predominant themes, the results are not as convincing (Bateman & Salfati, 2007).
Woodhams et al. (2007) have argued that measuring consistency using the psychological themes approach to case linkage has limited use when offenders are not able to be easily categorised into predominant themes. The authors also state that this approach may be too general to be used for identifying different offenders although it can be argued that this depends on the type of crime as it may be more applicable for rare crimes compared to more common crimes.
Even though there is evidence that offenders' crime series demonstrate behavioural distinctiveness and consistency as discussed above, crime analysts and police investigators may not be able to link crimes. For example, Canter et al. (1991) state that it is difficult to correctly identify crime scene behaviours appropriate for case linkage as attention may focus on behaviours that are not distinct or not stably exhibited.
Methods to combat the problems of case linkage
Cluster analysis, across-crime similarity coefficients, logistic regression analysis, and Receiver Operating Characteristic (ROC) analysis are analytical strategies employed to combat the problems of the assumptions underlying the practice of case linkage. Each of these strategies will be detailed next.
Cluster analysis places the crimes on a plot where the shorter the distance between plots the greater the similarity there is between the crimes (Green, Booth, & Biderman, 1976). The closer the crimes are together the more likely they are to be linked. The study by Green et al. (1976) found this analytical approach to be highly successful at linking the crimes of three burglars. Although there are some drawbacks to this method as it is not yet determined as to what specific distance suggests that the crimes are linked.
Across-crime similarity coefficients
Woodhams and Toye (2007) state that across-crime similarity coefficients quantify the degree of behavioural distinctiveness and consistency between two crimes. Woodhams and Toye (2007) used the Jaccard coefficients to see whether pairs of robbery crimes were linked based on the similarity of offence behaviours. Seeing as this coefficient ranges from 0 to 1, a large coefficient reflects a greater amount of across-crime similarity. In relation to the assumptions underlying the practice of case linkage, small similarity coefficients associated with different offender crimes suggests behavioural distinctiveness. Conversely, large similarity coefficients are associated with same offender crimes suggests consistency. Woodhams and Toye (2007) found that compared with unlinked crime pairs, linked pairs had greater similarity in behaviour exhibited between offences committed by the same criminal than between offences committed by different offenders. In addition, the difference between linked and unlinked crimes was significant for overall and within each behaviour. These findings support the offender behavioural consistency hypothesis underlying the practices of case linkage as well as offender profiling.
Logistic regression analysis
Bennell and Canter (2002) state that logistic regression analysis can be used to produce predicted probabilities that crime pairs are linked. This is done by combining the values of predictor variables such as across-crime similarity scores related to entry method, characteristics of the target, and the property stolen. Woodhams and Toye (2007) used logistic regression analysis to determine whether robberies were linked and produced accurate linking decisions.
Receiver Operating Characteristic (ROC) analysis
The measures of linking accuracy described above are threshold specific in that the accuracy score is only relevant when a specific decision threshold is used. To counter this problem, ROC analysis can be used in addition to these measures (Woodhams & Toye, 2007). ROC analysis quantifies the level of linking accuracy associated with lining cues (inter-offence distance) or case linkage tools (e.g., cluster analysis, across-crime similarity coefficients, and logistic regression analysis). This method is not affected by the decision threshold that was used to determine whether or not the crimes were linked.
Studies using ROC analysis have set suitable decision thresholds for linking purposes as well as producing valid measures of linking accuracy (e.g., Bennell & Jones, 2005) demonstrating that methods such as logistic regression used in conjunction with ROC analysis can accurately link serial crimes.
Offender profiling: Homology
Despite different methods of profiling being multi-disciplinary and overlapping, the approaches can be segregated into three broad categories: clinical practitioner, statistical, and criminal investigative (Wilson, Lincoln, & Kocsis, 1997). The FBI model is a criminal investigative approach and it is the most common. It is used in USA, Canada, Australia, UK, Germany, Sweden, Norway, and Holland amongst others. The work of Babcock, Boon, and Britton (Copson, Babcock, Boon, & Britton, 1997) is an example of the clinical approach whereas David Canter's (Canter & Heritage, 1990) work is representative of the statistical approach.
In the past, the FBI model has been criticised for being based upon opinion and intuition rather than fact and solid methodology which lacks the element of psychological theory (Canter & Alison, 1999). Now it takes more of a scientific approach by publishing empirical findings. The clinical practitioner approach emphasises its use of psychological, psychiatric, clinical, or mental health research or knowledge. Finally, the statistical approach infers psychological processes and characteristics of offenders using the multivariate analysis of behaviour at the crime scene as well as other crime scene information (Canter & Alison, 1999).
Pinizzotto and Finkel (1990) state that professional profilers first assess the type of criminal act with reference to individuals who have committed similar acts previously. Then they analyze the crime scene, the victim's background, and any possible suspects. Next profilers establish the possible motivations of all the individuals involved. Finally, a description of the offender is created from the characteristics identified with such a psychological disposition.
Homology is an assumption that underlies the practice of offender profiling and suggests that similar offence styles are associated with similar background characteristics (Woodhams & Toye, 2007). Thus offenders with similar offence behaviours have similar background characteristics. The complex relationships between a given action and a given characteristic are made to profile clusters of background features from crime scene actions to develop a psychological picture of the suspect (Alison et al., 2002). These relationships involve multivariate forms of prediction in which specific sets of actions are linked to specific sets of characteristics. Although the results of these analyses are not convincing when tested (Alison, 2002).
Alison et al. (2002) state that current offender profiling methods rely on an outdated understanding of personality and the trait approach. Alison et al. (2002) found that the homology assumption is not supported as it uses global traits, or broad personality types, which are unlikely to be useful in predicting criminal behaviour. These traditional personality trait theories suggest that all offenders' behaviours are affected in predictable ways (deterministic) and that behaviour remains stable despite different environmental influences (nonsituationist) (Alison et al., 2002). Furthermore, Mischel and Peake (1982) demonstrated that global trait theories fail to allow predictions of behaviour across specific situations.
Alison et al. (2002) suggest that it is unlikely that the classification of offenders into broad personality types would enable the profiler to relate clusters of sociodemographic characteristics to different types. Profilers often make inferences about offender characteristics that are not appropriate for a psychological definition e.g. age, ethnicity, marital status, gender, degree of sexual maturity, and possible reactions to police questioning (Alison et al. 2002).
Woodhams and Toye (2007) tested the homology assumption by identifying three different styles of robbery behaviour and investigated whether commercial robbers with different offence behaviour styles were significantly different to each another on ethnicity, employment, previous convictions, age, and distance to crime. No significant differences were found which suggests no support for the homology assumption.
The homology assumption has also been investigated using regression analyses to determine if certain offence behaviours can reliably predict offender characteristics although there has been little support found for the homology assumption (Tonkin, Bond, & Woodhams, 2009). For example, Mokros and Alison (2002) investigated rapists and defined homology as a positive linear relationship between the similarities of crime scene actions and similarities in background characteristics of the offenders. The researchers investigated whether similarity in personal characteristics (e.g., ethnicity, age, employment status, marital status, criminal record, and education) was correlated with similarity in offence behaviour. No linear relationships were observed with similarity in behaviour. Therefore there was no evidence found for the homology assumption.
Despite this lack of evidence for the homology assumption, there has been success with homicide (Salfati & Canter, 1999), burglary, assault, robbery, actual bodily harm and violent disorder (Farrington & Lambert, 2000), and sexual assault (Santtila, Ritvanen, & Mokros, 2004) offences at determining offender characteristics from crime scene behaviour.
In summary, using bivariate predictive analyses, relationships between specific characteristics and actions have been found (Aitken et al., 1996) although it has been argued that these are simplistic compared to the profiles that have been produced over the years (e.g. Pinizzotto & Finkel, 1990). Alison et al. (2002) state that a direct link between offence behaviour and offender characteristics is unlikely to exist because of the influence of the situation effecting offence behaviour. In addition, the authors state that profilers often make inferences regarding offender characteristics (e.g., gender, ethnicity, and age) from behaviour that is not appropriate according to Bem and Allen's (1974) personality theory. Therefore, the research into the homology assumption has been less supportive (Woodhams & Toye, 2007). Interestingly, there have been many reviews of profiling based on the opinions of detectives about the usefulness of the advice that they received, the overall opinion was that the advice was of use (Woodhams & Toye, 2007).
Methods to combat the problems of offender profiling and case linkage
A method to combat the problems of the assumptions underlying the practice of offender profiling and case linkage is for police forces to enter crime scene information (including motivational and behavioural characteristics) into databases. The Violent Crime Case linkage System (ViCLAS: Collins, Johnson, Choy, Davidson, & Mackay, 1998) is the most popular of the databases and was first used by the Royal Canadian Mounted Police to profile offenders and identify serial offenders. This system can record 200 items of information from each crime scene which analysts can search when a new crime is committed. This system is widely used in the UK and assists offender profiling as well as case linkage by providing a way to manage the mass of information although there is little research on its effectiveness.
Beauregard (2005) states sharing information between investigators and researchers is a way to develop offender profiling as it standardises data reliability. Alison et al. (2004) state that this pragmatic profiling (sharing information, investigative advice, and profiles) is the way forward for profiling and case linkage. Yokota, Fujita, Watanabe, Yoshimoto, and Wachi (2007) have even developed a computerised profiling system that takes in the details of an offence and produces the offender characteristics.
Finally, Alison et al. (2002) suggest that a theoretical framework that emphasizes the importance of Person X Situation interactions in generating behaviour may be more productive. Recent research by Goodwill and Alison (2007) has included these aspects into predictive analysis of offender characteristics from crime scene information and stated that the relationship between offender characteristics and the crime scene may be influenced by other aspects of the crime. Therefore, using the contemporary trait perspective would include contextual details that effect behaviour and thus make inferences from crime scene behaviours more valid and useful.
In summary, there is evidence highlighting problems with the consistency, homology, and behavioural distinctiveness assumptions underlying the practices of offender profiling and case linkage. Regarding offender profiling, ignoring contextual and situational variables of offending makes homology and consistency unreliable and possibly invalid. The simple behaviour and characteristics approach to offender profiling is not adequate. It may be argued that a significant component of offender profiling is based on intuition rather than scientific research (Pinizzotto & Finkel, 1990). On conclusion of the evidence for homology, the practice of offender profiling does not include current psychological theory and therefore would not be appropriate for use in court as expert witness evidence. The homology assumption has not seen as much research as the behavioural consistency and distinctiveness assumptions. There is greater support for the consistency and distinctiveness assumptions underlying the practice of case linkage when additional measures are used to improve validity and usefulness.
Methods to improve the practices of case linkage and offender profiling include the use of databases, pragmatic profiling, and automation. Methods to improve case linkage include cluster analysis, across-crime similarity coefficients, and logistic regression analysis. These analytical techniques improve the practices of case linkage and offender profiling as they are able to accurately handle large amounts of data. The drawback to many of these approaches is that they do not specifically state what probability level should be used to determine whether or not crimes and behaviours are linked. These measures of linking accuracy are threshold specific in that the accuracy score can only relevant when a specific decision threshold is used. To counter this flaw, ROC analysis has proven to accurately link crimes when used in conjunction with these methods.
It is important to have valid assumptions underlying the practices of case linkage and offender profiling as incorrect advice could mislead a police investigation and be costly in financial and human terms. Greater research is desperately needed into the effectiveness of the use of databases to assist case linkage and offender profiling. These databases are in widespread use despite a lack of research on their effectiveness.
Word Count: 3814