Count time series models for road traffic accidents in Tanzania Mainland

Abstract

A pairwise analysis was conducted to assess the trends and factors associated with road traffic accidents in Tanzania. The Poisson and Negative Binomial Autoregressive Models were used to extend log linear functions by accounting time-varying components. A total of 85,514 road traffic accidents in Tanzania mainland that occurred from 2012 to 2017 were extracted from Tanzania Police Office records. Eleven factors were grouped into a human, vehicle, physical/environmental and pedestrian-related factors. The Likelihood ratio test, Akaike Information Criterion, Bayesian Information Criterion and residual ACF plots were used to evaluate the performance of the models in Dar es Salaam and other combined regions. The trend analysis indicated a declining pattern in all factors and human-related factors appeared higher than the other three factors. The highest number of road traffic accidents was observed in Dar es Salaam Region compared to other combined regions. The models, including its past values and time-varying factors, were in favour-of other models. In both, Dar es Salaam and other combined regions, non-linear pattern and Negative Binomial Autoregressive Models fitted the data well. The implementation of collective actions in recent years seems positive on road traffic accidents. Nevertheless, more emphasis is needed to monitor trends on the number of accidents and related fatalities.
 Keywords: Road Traffic Accidents, Poisson, Negative binomial, Autoregressive Models, Tanzania.