The randomness of accident counts.

Road accident occurrence is generally assumed to be governed by a random process. Experience analysing accident count series indicates a need for testing the randomness of such a series, and a simple statistical test is described. The test is a modified one-sample runs test, which can be used to assist the identification of three distinct patterns constituting non-randomness (namely, a trend, a discontinuity and over-correction). The test is designed for the preliminary analysis of accident count data, and its use is demonstrated, using both artificial and actual accident count series.