Spatiotemporal and random parameter panel data models of traffic crash fatalities in Vietnam

• This paper investigates factors associated with traffic crash fatalities in Vietnam.
• Random effect and random parameter negative binomial panel data models are used.
• A spatiotemporal model with conditional autoregressive priors (ST-CAR) is adopted.
• The ST-CAR model outperforms random effect and random parameter models.
• The safety impact of level crossings at a province-level is highlighted.

This paper investigates factors associated with traffic crash fatalities in 63 provinces of Vietnam during the period from 2012 to 2014. Random effect negative binomial (RENB) and random parameter negative binomial (RPNB) panel data models are adopted to consider spatial heterogeneity across provinces. In addition, a spatiotemporal model with conditional autoregressive priors (ST-CAR) is utilised to account for spatiotemporal autocorrelation in the data. The statistical comparison indicates the ST-CAR model outperforms the RENB and RPNB models. Estimation results provide several significant findings. For example, traffic crash fatalities tend to be higher in provinces with greater numbers of level crossings. Passenger distance travelled and road lengths are also positively associated with fatalities. However, hospital densities are negatively associated with fatalities. The safety impact of the national highway 1A, the main transport corridor of the country, is also highlighted.