This report presents the results of a first attempt to combine detailed road geometry, road surface condition, carriageway characteristics and crash data information to develop a statistical crash prediction model for application to rural New Zealand state highways. Such a study was made possible because high-speed surveys generating simultaneously measured road condition and road geometry data for the entire 22,000lane-km of New Zealand's state highway network have been undertaken annually since 1997.
Four road crash subsets were investigated: all reported injury and fatal crashes; selected injury and fatal crashes for loss-of-control events; reported injury and fatal crashes in wet conditions; and selected injury and fatal crashes in wet conditions.
One- and two-way tables and Poisson regression modelling were employed to identify critical variables and their relationship with crash risk. Horizontal curvature, traffic flow, skid resistance and, to a lesser extent, lane roughness were critical variables common to all investigated crash subsets. The resulting Poisson regression model uses 2nd- or 3rd-order polynomial functions of critical variables to allow for observed non-linear responses, enabling the model to be incorporated into existing road asset management systems. A comparison of observed and predicted crash numbers shows that the model can provide estimates of crash numbers to sufficient accuracy for safety management purposes.