Previous studies, though not specific to juveniles, have identified gender, age, offence and criminal history as the key correlates of reconviction (Tarling, 1993). Cain (1995) in his study on juvenile recidivism found recidivism to be associated with similar variables in addition to type of court and first penalty. Criminal history was approximated from the number of previous proven court appearances at the time of the first proven appearance during the 1992-93 period.
A log-normal specification proved to be the most adequate to describe the relationship between time to reappearance and its explanatory variables. Two separate data sets were randomly created from the records in the cohort. The first data set known as the training sample was used to identify the model that provided the best fit to the data. Selection of covariates and determination of their levels was made in terms of theoretical relevance and contribution to goodness of fit. A first model was fitted with dummy variables defined for all the categories for each of the candidate covariates (refer to Table 1).