Classifying patterns of criminal activity: a latent class approach to UK longitudinal conviction records

This paper is concerned with classifying longitudinal event data for a set of individuals. However, rather than classifying individuals, we are interested in identifying homogeneous periods of types of activity within such a longitudinal time series, and in classifying these periods of activity. The focus is on males measured through official criminal conviction histories. We examine a cohort of 10,000 offenders in England and Wales born in 1953. For any individual, we expect migration through different types of activity during their criminal career. For each age, we construct a time window of fixed size centred on that age. We summarise the criminal activity in a time window by constructing a set of binary indicator variables for a large number of offences. A latent class analysis of all time windows provides a classification of types of offending which co-occur in time. The approach is analogous to a uniform kernel density smoothing with fixed bandwidth, and can be used for classifying any set of longitudinal event data where changes in activity are expected over time.