ARSRPE Conference Paper Database

Using logit modelling to measure the safety of New Zealand drivers

Veltman, R

Data Analysis


Measuring Safety is akin to quantifying risk. A logit model has been developed that determines the percentage probability of a driver being involved in a crash incident in the following year, conditional upon certain driver characteristics and driving behaviour in the current year. The logit model was regressed using the total New Zealand driver population, and is applied to each and every individual current New Zealand driving license holder. The model is designed to be applied annually, and a five class histogram of driver probabilities is derived from the results. The probability risk score is conditional upon the drivers' gender, age, driving experience and traffic offending. The "most at Risk" class contains those drivers whose probability scores exceed 0.7632%. That driver group is categorized as being "Recidivist" and is analysed as such to determine common characteristics that can be utilized to inform policy. In applying the model to drivers and tracking the resulting scores to "at fault" drivers involved in serious crashes in the following year, a robust "predictive" validation of the model was discovered. Crashes in the main involved both band 1 and band 2 "recidivist" drivers already identified in the year prior. Analysis of the recidivist group reveals that the predominant problem cohort is young males aged generally between 16 and 24 years of age, and speed is the predominant behaviour.