Abstract

Crime time series data can often be naturally disaggregated based on various attributes of interest, such as crime type or geographical location. When modelling this type of data, the current practice in crime science is to model each series at the most disaggregated level, as it helps to identify more subtle changes. However, authorities and stakeholders often focus on the bigger picture, leading researchers to either simply sum the fitted value series or model the aggregated series independently. This practice often leads to poorer performance at the higher levels of aggregation as the most disaggregated series typically exhibit a high degree of volatility, while the most aggregated series tends to be smoother and less noisy. In this presentation, we will demonstrate how the hierarchical and grouped time series structure can be utilised to provide “coherent” estimates for all disaggregate and aggregate series while also “reconciling” them to enhance forecast and criminal justice program evaluation by using all the available information. We will utilise some US crime data alongside the COVID lockdown as the intervention effect for illustrative purposes.