Individual collision types have different underlying causes and thus the relationships between roadway/traffic characteristics and crash frequency are likely to differ across unique collision types. One way these different influences have been studied is by developing separate statistical models for each collision type. While this is the most straightforward approach, developing collision-specific models can be very tedious and can produce unreliable estimates for collision types that are less frequently observed. Moreover, ignoring correlations between different collision types may result in biased and inefficient parameter estimation. To overcome these limitations, researchers have adopted a multivariate approach that explicitly accounts for the correlation among individual collision types. As an alternative to multivariate approach, two-stage approaches have been proposed in which one model is estimated to predict total crash frequency and its prediction is combined with another model, used to predict the proportions of different collision types. More efficient one-stage joint models, in which both the frequency and proportion models are estimated simultaneously and predictions are provided more directly, have also been proposed for macro-level analysis. This study investigates the performance of this joint model paradigm in analyzing unique collision type frequencies on individual road segments. For this, a joint negative binomial-multinomial fractional split (NB-MFS) model is used. Moreover, this study also proposes the use of a multinomial logit (MNL) model to estimate the proportion of different collision types. As total crash frequency NB model and MNL model utilize different datasets, a two-stage estimation process is required, which leads to the two-stage NB-MNL model proposed here. The performance of proposed model is compared with that of collision-specific NB models, multivariate negative binomial (MVNB) model, and NB-MFS model in predicting crash frequency by collision type on two-way two-lane urban-suburban collector roadway segments in Pennsylvania. The goodness of fit statistics show that the NB-MNL model performs better than collision-specific NB models, MVNB model and joint NB-MFS model and is thus a promising approach in predicting crash frequency by collision type.
All Science Journal Classification (ASJC) codes
- Human Factors and Ergonomics
- Safety, Risk, Reliability and Quality
- Public Health, Environmental and Occupational Health