Temporal Dynamics and Statistical Modelling of Road Crash Frequency, Severity Composition, and Vehicle Involvement in Dhaka City of Bangladesh

Abstract

Dhaka's road-safety burden is shaped not only by the number of reported crashes but also by changes in severity, pedestrian involvement, and the number of vehicles participating in each event. This study develops an internally validated annual time series from road-accident, casualty, and vehicle cross-tabulations for Dhaka City for 1998-2015, supplied as data from the Accident Research Institute (ARI), Bangladesh University of Engineering and Technology (BUET). The harmonised record contains 10,959 reported crashes, 12,470 casualty records, and 15,697 involved-vehicle records. Data quality was assessed by reconciling independent table totals and reconstructing crash counts from the vehicle-distribution table. Temporal behaviour was evaluated using descriptive statistics, ordinary least squares with HC3 robust standard errors, Mann-Kendall and Theil-Sen trend estimators, Poisson and negative binomial NB2 count models, grouped binomial logistic models, segmented regression, Spearman correlation, and principal component analysis (PCA). Reported crashes decreased from 1,202 in 1998 to 391 in 2015 (-67.5%), casualty records decreased by 65.2%, and involved-vehicle records decreased by 73.8%. Strong Poisson overdispersion (8.22-27.06) favoured NB2 models; the crash-frequency NB2 incidence-rate ratio was 0.525 per decade (95% CI: 0.459-0.602). However, the fatal-crash share rose from 33.2% to 74.2% (grouped-binomial OR = 3.044 per decade), the pedestrian casualty share rose from 35.9% to 59.4%, and the pedestrian-collision share rose from 41.0% to 71.4%. Conversely, the multi-vehicle crash share fell from 52.8% to 23.8%, and mean vehicles per crash declined from 1.562 to 1.256. The first two principal components explained 84.2% of standardised annual variation. Overall, the findings indicate a transition from a higher-frequency, multi-vehicle crash profile toward a lower-frequency but more fatal- and pedestrian-concentrated reported crash structure. These results support prioritising pedestrian protection, speed management, safer junctions and public-transport interfaces, and stronger integrated crash-data systems in Dhaka. Because exposure denominators and crash-level covariates were unavailable, the results should be interpreted as temporal associations in reported events rather than causal or exposure-adjusted risk estimates.