MODELLING ROAD ACCIDENT DATA IN THE NORTH CENTRAL OF NIGERIA USING POISSON-LOGNORMAL AND NEGATIVE BINOMIAL REGRESSION MODELS

Abstract

This study aims to analyse road accident data to identify significant predictors of fatalities and determine the best-fitting model using statistical criteria. The objectives include visualising road accident data, fitting Poisson-Lognormal and Negative Binomial regression models, assessing model performance using Akaike and Bayesian Information Criteria, and evaluating the contribution of key factors—Driver Error, Faulty Vehicle, and Road Condition—to fatalities. The Poisson-Lognormal model emerged as the best fit, with the lowest AIC (-132.4) and BIC (-118.5), a high R-squared (0.87), and a pseudo-R-squared of 0.9999, indicating strong explanatory power. Key findings reveal that Driver Error is the most significant predictor, contributing to a 2.17% increase in fatalities per unit increase, followed by Faulty Vehicle (1.4%), while Road Condition showed no significant effect. Temporal and seasonal analyses highlighted fluctuating trends, with peaks in 2016-2017 and 2020, and a notable decline by 2024. Regional analysis identified the Federal Capital Territory, Niger, and Kogi as hotspots. The study concludes that targeting Driver Error and Faulty Vehicle issues through enhanced driver training, stricter enforcement, vehicle maintenance programs, and data-driven policies could substantially reduce fatalities. Road infrastructure improvements, while important, require complementary measures to maximise impact. These findings provide critical insights for policymakers to develop effective strategies to mitigate road accident fatalities in North Central Nigeria.