Résumé
In the absence of physical evidence, investigators must rely on offense behaviours when determining whether several crimes are linked to a common offender. This is known as behavioural linkage analysis (BLA). A variety of factors may influence the degree to which it is possible to link serial crime, including the crime type under consideration, the similarity coefficient used to assess across-crime similarity, and the frequencies of the behaviours used to conduct the analysis. The current thesis consists of two studies designed to examine these factors. Study 1 compares the linking accuracy of two similarity coefficients - Jaccard's coefficient (J) and the taxonomic similarity index (As)- across three crime types - serial sexual assault (n = 126), serial homicide (n = 237), and serial burglary (n = 210). Study 2 compares the linking accuracy that can be achieved across the three crime types when using various subsets of behaviours defined by their frequency of occurrence. The degree to which these factors influence linking accuracy is determined through the use of receiver operating characteristic (ROC) analysis. The results of Study 1 indicate that no significant differences exist between the linking ability of J and As across a variety of conditions, although J slightly outperforms As when relying on larger sample sizes. Both coefficients lead to much higher levels of linking accuracy in cases of serial homicide and sexual assault compared to serial burglary. The results of Study 2 indicate that no significant differences emerge when four behavioural frequency recommendations are compared to one another, indicating that it might be most productive to base linkage analyses on all crime scene behaviours included in a crime sample. These various findings are discussed, along with their implications. The thesis ends with a presentation of some limitations with the research and some suggestions for how these limitations can be overcome in the future.