Statistical Modelling of Escalation in Crime Seriousness: Through Survival Analysis, Mixed-effects and Mixture Modelling Approaches.

Liu, Jiayi (2012) Statistical Modelling of Escalation in Crime Seriousness: Through Survival Analysis, Mixed-effects and Mixture Modelling Approaches. PhD thesis, UNSPECIFIED.

PDF (11003744.pdf)
11003744.pdf - Published Version
Available under License Creative Commons Attribution-NoDerivs.
Download (6MB)

Abstract

Escalation/de-escalation of offending is an important topic for criminal justice policy, but has been comparatively neglected in criminal careers research. This thesis introduces the Offenders Index (OI) dataset from the Home Office in England and Wales which is the preferred dataset for assessing escalation in this thesis. Three major studies are then reported under two main research focuses. One research study focuses on 'serious offender escalation'. This study examines offenders who had been convicted of arson, blackmail, threats to kill, or kidnapping, and assesses whether they will be convicted of the most serious crime - homicide. This study suggested that 1 in 100 kidnapping offenders are likely to have a subsequent homicide conviction over a 20-year follow-up period, which doubles the risk of homicide conviction compared to the other three types of offenders. Moreover, offenders can double their risk of homicide conviction by being involved in multiple serious offences (among the four serious offences). The second research focuses on 'general escalation'. This includes two studies: the first study examines the effects of two temporal scales, both age and order of convictions on escalation of seriousness by using a linear mixed-effects model. The results suggested that ageing is associated with de-escalation whereas the number of conviction occasions is associated with escalation, with the two processes pulling in different directions. This is followed by the last study which examines the hypothesis that there are different types of underlying criminal development in escalation across offenders. Therefore, a combination of mixed-effects and mixture modelling methodology has been developed to understand both individual crime growth curves and to distinguish latent types of crime development. A three-class solution has been identified by growth mixture modelling approach. The first class consists of the majority (88%) of offenders who are relatively stable in their seriousness in crimes, and have some tendency to de-escalate with age and some tendency to escalate with experience. The second class consists of 6.4% of offenders who have average high seriousness (7.6) at age 10, and have a strong de-escalation effect with age. The third class consists of 5.6% of offenders who have shown more diversity in crime seriousness, and also are involved with more high seriousness crimes. The last study also provides a comparison framework to compare the linear mixed-effects model, group-based trajectory model, and growth mixture model through graphical investigation and proposed statistical diagnostic measures. For the particular data used in this thesis, the growth mixture model with three classes is preferred as the best fitting model compared to various other fitted models.