12,000

We have over 12,000 students, from over 100 countries, within one of the safest campuses in the UK

93%

93% of Lancaster students go into work or further study within six months of graduating

Home > Research > Publications & Outputs > Jointly modeling time-varying offending profile...
View graph of relations

Download

« Back

Jointly modeling time-varying offending profiles and criminal career trajectories in a sample of sexual offenders: A multivariate approach to modelling criminal careers

Research output: ThesisMaster's Thesis

Published

Publication date2012
Number of pages57
Awarding Institution
Supervisor(s)/Advisor
Date of Award12/12/2012
Original languageEnglish

Abstract

The study begins by summarising three key topics of research in criminal careers of sexual offenders: explaining the observed bimodality of the age-crime curve for sexual offenders; determining whether rapists and child molesters follow well-separated, or overlapping criminal career paths; and assessing whether non-contact sexual offending is associated with serious sexual crime, and if so, which serious sexual crimes it is associated with. It is argued that these and other salient questions in the study of criminal careers of sexual offenders cannot easily be answered using trajectory models that aggregate crimes across types, or profile models that account for mix of crimes but do not take account of criminal career dynamics.

Using data on 824 male sexual offenders from the Massachusetts Treatment Center, the study investigates four methods for the joint analysis of sexual offender trajectories and profiles. The four methods are: optimal matching with non-probabilistic clustering; ``constrained'' multivariate group based trajectory models; ``unconstrained'' multivariate group based trajectory models; and a Poisson-log normal factor model with posterior trajectory analysis of the factor scores.

Optimal matching is found to be a useful way to summarise and visualise complex sequences of events, with the advantage that clustering can be performed without aggregating the data. However, without the ability to make inferences from findings its usefulness is limited to an exploratory role.

The constrained and unconstrained multivariate trajectory models are compared, and it is found that neither one dominates the other in all contexts, and that both correspond to interesting theoretical hypotheses. It is suggested that the unconstrained models should be a starting point for modeling, since they make fewer assumptions and allow the appropriateness of the constrained model to be tested.

It is demonstrated that the factor model can produce parsimonious trajectories for different types of criminal activity that are close to independent at each time point, but which are nevertheless highly associated (positively or negatively) over the life course. In applying the model, it is shown that a three factor model distinguishes between trajectories characterised by general crime, rape and child molestation.

With regards to the research questions, all of the methods lend evidence to indicate the existence of bimodal trajectories, and that these trajectories exist at the individual level and are not an artifact of aggregation. However, it is shown that it is not possible to answer questions relating to associations between the occurrences of certain types of crime, using a dataset in which certain combinations of occurrence and non-occurrence are structurally missing.