Box-Tiao Time Series Models for Impact Assessment

Abstract

Deutsch and Alt (1977) use a time series quasi-experiment to assess the impact of the Massachusetts gun control law on gun-related crime in Boston. They conclude that the new law resulted in a statistically significant drop in both armed robberies and gun assaults, though not in homocides. However, our reanalysis of their data suggests that these conclusions are extremely tenuous. First, Deutsch and Alt have seriously mis specified the stochastic components for the three time series. We report alternative models which, in all three cases, have better statistical properties. Second, Deutsch and Alt conduct their analysis only six months after passage of the new law. This postintervention time series segment is obviously too short a period to permit specification of the inter vention component. Our reanalysis demonstrates that there is inconclusive evidence for the effect claimed by Deutsch and Alt.