Modeling road traffic fatalities in India: Smeed's law, time invariance and regional specificity

Mathematical formulations linking road traffic fatalities to vehicle ownership, regional population, and economic growth continue to be developed against the backdrop of Smeed and Andreassen models. Though a few attempts were made, Smeed's law has not been fully tested in India. Using the 1991–2009 panel data from all states, this work (a) developed the generalized Smeed and Andreassen models; (b) evaluated if traffic fatalities were impacted by structural changes; and (c) examined if – in relation to the generalized model – the individual (time and regional) models are more relevant for application. Seven models (Smeed: original, generalized, time-variant, state-variant; and Andreassen: generalized, time-variant, state-variant) were developed and tested for fit with the actual data. Results showed that the per vehicle fatality rate closely resembled Smeed's formulation. Chow-test yielded a significant F-stat, suggesting that the models for four pre-defined time-blocks are structurally different from the 19-year generalized model. The counterclockwise rotation of the log-linear form also suggested lower fatality rates. While the new government policies, reduced vehicle operating speeds, better healthcare, and improved vehicle technology could be the factors, further research is required to understand the reasons for fatality rate reductions. The intercept and gradients of the time-series models showed high stability and varied only slightly in comparison to the 19-year generalized models, thus suggesting that the latter are pragmatic for application. Regional formulations, however, indicate that they may be more relevant for studying trends and tendencies. This research illustrates the robustness of Smeed's law, and provides evidence for time-invariance but state-specificity.

► Smeed and Andreassen-based time-series and region-specific models were developed.
► Time-based models showed high stability indicating that they can be safely applied.
► Regional models were unstable suggesting that they cannot be used universally.
► Structural analysis showed that road fatality rates dropped during four time-blocks.
► Smeed’s model is robust; panel data help improve the predictive power of models.