Identifying values for missing crime severity using constraints
20th April 2018, 2:00 pm – 3:00 pm
Main Maths Building, SM3
A long-standing issue in criminology is that sentence types come in two flavours -- custodial sentences measured in days and non-custodial sentences measured as factor levels -- making it difficult to compare the two types of outcomes. We describe a method to extend a severity score based on sentence length to non-custodial outcomes. We use data from the Criminal Council Sentencing Survey (CCSS) a UK based survey containing detailed information on each crime. First we assume that the relationship between the covariates of interest (e.g. prior convictions) are the same across the two types of sentence. These can be estimated from the custodial outcomes. Second we impose constraints on the imputed non-custodial severities. These include an ordering and also: imposing a range of thresholds and resampling "incorrectly" imputed sentences until the fall within the correct range and resampling from the correct match's distribution amongst others. We apply these constraints to two types of Bayesian missing not at random models. We present results from the CCSS data where we use one subset to develop the model and another subset to test it.