Abstract
The common and convenient assumption in accident count analysis, that accidents are Poisson-distributed, is re-examined. Two statistical tests, for evaluating the assumption are described and compared. It is shown that a test based upon a combinatorial analysis is much more accurate than the alternative chi-square test when accident counts are expected to be small. The more accurate test is used to reinterpret data on accident count variability, the results indicatingthat the Poisson distribution is appropriate for the analysis of accidents at individual sites. (A)