C-LINK's Research Projects

The C-LINK members worked together on two collaborative research projects. Network members and their partner police organisations collated datasets of unsolved (linked-by-DNA) and convicted serial and one-off sexual offences/homicides. Standard protocols were used to anonymise, code, and store data securely, and data sharing agreements were agreed and implemented. The datasets from each country were collated by Dr. Amy Burrell at the University of Birmingham and then subject to statistical analysis by C-LINK members.

Project 1: Main findings

The first study was focused on testing the underlying assumptions of crime linkage (i.e., behavioural consistency and behavioural distinctiveness) with the most ecologically-valid dataset amassed through the collaborative efforts of the group. We succeeded in producing a dataset that more closely reflected the types of data held on police databases used for crime linkage which was also the largest sample of sexual offences studied, to date. We tested the assumptions using a standard statistical protocol that we and others have used in existing studies of the crime linkage assumptions that was first articulated by Bennell (2002) which involves contrasting the behavioural similarity observed in linked crime pairs (same offender) and unlinked crime pairs (different offender). We have the additional step nowadays of including a leave-one-out cross validation (see Woodhams & Labuschagne, 2012). The main findings were:

• That the underlying assumptions are, generally, supported as per previous studies in that it is possible to differentiate linked from unlinked crime pairs on the basis of behavioural similarity to an excellent level of accuracy.
• As per previous studies, our sample contains evidence of serial offenders who are sufficiently consistent and distinctive in their behaviour for their crimes to be linked accurately using behaviour; however, there are also examples of serial offenders who are far less consistent and/or distinctive.
• The levels of discrimination accuracy are, however, reduced when testing the assumptions solely with crime pairs that were linked by DNA. For some time, researchers have highlighted that convicted crime series might be characterised by greater behavioural consistency and distinctiveness which is why they were identified as a series in the first place. These findings suggest these concerns are valid.  

Project 2: Main findings

Using the same dataset as for Project 1, each network member applied the statistical technique(s) for which they have particular expertise. The relative performance of the different statistical techniques was assessed by the C-LINK team.

For the full articles – now published or submitted for publication – please see the Resource Centre.