Markov switching multinomial logit model: An application to accident-injury severities

Area of Science:

• Statistical Modeling
• Transportation Safety
• Econometrics

Background:

• Unobserved heterogeneity in accident data can bias traditional models.
• Markov switching models offer a framework to capture unobserved state dynamics.

Purpose of the Study:

• To propose and estimate two-state Markov switching multinomial logit models for accident-injury severities.
• To account for unobserved heterogeneity in roadway safety states.
• To assess the statistical fit and applicability of these models.

Main Methods:

• Development of two-state Markov switching multinomial logit models.
• Application to accident severity data from Indiana roads (four-year period).
• Utilized Bayesian inference and Markov Chain Monte Carlo (MCMC) simulations for model estimation.

Main Results:

• Markov switching models demonstrated a superior statistical fit compared to standard multinomial logit models.
• Identified two distinct states of roadway safety.
• The more frequent state correlated with favorable weather; the less frequent state correlated with adverse weather.

Conclusions:

• Markov switching models effectively capture unobserved dynamics in accident-severity outcomes.
• Weather conditions are significantly associated with different states of roadway safety.
• The proposed models provide a more robust approach to traffic safety analysis.