Fitting probability distributions to accident frequency data

Previous research has demonstrated that the truncated Negative Binomial distribution can be used successfully to model accident frequency data. Recent work, however, on accident data from Melbourne, apparently refutes this, and suggests that the log series distribution should be used instead. The Melbourne data are re-analyzed in this paper, and it is shown that the Negative Binomial does in fact fit better than the log series distribution. For measuring the lack of fit between model and data, an alternative statistic is proposed which is easily interpretable in terms of the typical calculations which might be done once a usable model has been found. Language: en