Heckman's Methodology for Correcting Selectivity Bias: An Application to Road Crash Costs

Abstract

Aggregate road crash costs are traditionally determined using average costs applied to incidence figures found in Police-notified crash data. Such data only comprise a non-random sample of the true population of road crashes, the bias being due to the existence of crashes that are not notified to the Police. The traditional approach is to label the Police-notified sample as 'non-random' thereby casting a cloud over data analyses using this sample. Heckman however viewed similar problems as 'omitted variables' problems in that the exclusion of some observations in a systematic manner (so-called selectivity bias) has inadvertently introduced the need for an additional regressor in the least squares procedures. Using Heckman's methodology for correcting for this selectivity bias, Police-notified crash data for Western Australia in 1987/88 is reconciled with total (notified and not notified) crash data in the estimation of the property damage costs of road crashes.