Disentangling age, period and cohort effects in criminal typologies

A classic problem in the statistical analysis of longitudinal data is the ability to separate out age, period and cohort effects. This is particularly relevant in criminology, where it is important to disentangle social change (different generations behaving differently) from year effects (the immediate effect of world events, economics and government criminal justice policy)while taking account of thewell-known age-crime relationship. We explore this problem in the context of patterns of offending behaviour. In the analysis of the conviction histories of six cohorts of offenders in England and Wales,we carry out a latent class analysis of five-year periods for all cohorts, identifying patterns of offences which co-occur. We then examine the estimated proportions of each cohort and each five year period, with the intention of identifying generational effects in cluster sizes. We explore statistical modelling approaches to this problem, and present preliminary findings.