Some statistical issues in the assessment of crime reduction interventions

Paul Marchant, Paul D. Baxter & Stuart Barber

This paper points out deficiencies in the methods that are used to judge the effectiveness of crime reduction interventions. The problem is that the variance of the number of crimes recorded, within an area in a period of time, is many times the mean rather than simply being equal to the mean number; that is there is overdispersion. The work here shows how a number of studies might be combined to estimate how the claimed variance needs to be inflated to get a more realistic, yet still uncertain, estimate of the true variability in the number of crimes and thereby tell whether there is any sound evidence for claiming that the intervention has the desired effect.

Keywords Crime reduction, Non-randomised studies, Overdispersion, Quasi-Poisson Models, Scientific Methods Scale , `What Works'