It is shown how statistical methods based on directed graphs may be useful in modeling traffic fatalities by comparing models specified using directed graphs to a model originally developed by Peltzman (work published as: The effects of automobile safety regulation in: Journal of Political Economy 1975 v83 pp 677-725 and further work in Accident Analysis and Prevention 1976 v8 pp 139-142). The comparison uses Peltzman's original data, as well as up-dated data (and coefficients) through 1993. Out-of-sample forecasts of traffic fatalities from Peltzman's model are compared with those from a model constructed using directed graphs over data for the more recent period. The directed graphs model outperforms Peltzman's model in root mean squared forecast error.
Abstract