Regression analysis of aggregate accident data: some methodological considerations and practical experiences.

Abstract

In empirical traffic safety studies regression analysis is perhaps the most widely used statistical technique. In the great majority of cases the regression model is applied in its standard form. If, however, highly aggregate accident data are analyzed, usually most of the assumptions of the standard model are violated. Referring to a practical study it is demonstrated how it may become difficult to decide whether or not a certain variable can be regarded as truly "explanatory." It appears that even in seemingly unproblematic situations contradictory results are obtained from only slightly different model specifications. It is recommended that the use of highly aggregate accident data be avoided whenever possible. Some hints are given concerning the use of appropriate extensions of the classical linear regression model in road accident studies.