A comparison of the mixed logit and latent class methods for crash severity analysis

•We compared the latent class and mixed logit methods for crash severity analysis.•We used a large sample of crash data on multiple vehicle crashes involving a heavy truck.•The comparison lied on model fit, marginal effects, and predicted outcome probabilities.•The latent class model had a slightly better fit but both models' inferences were fairly similar.•Mixed logit average predicted probabilities were closer to the observations.

While there have been many studies analyzing crash severity, few studies have accounted for unobserved heterogeneity and compared different crash severity models. The objective of this paper is to investigate the differences between two preferred methods for accommodating individual unobserved heterogeneity, the mixed logit and latent class methods, in exploring the relationship between heavy truck crash severity and its contributing factors. To achieve this, a large sample of crash data on multiple vehicle crashes involving a heavy truck on public roadways in Iowa from 2007 to 2012 was collected. The comparison of the two methods lied on model fit, inferences, and predicted crash severity outcome probabilities. The results suggested a slight superiority of the latent class method in terms of model fit; however, the mixed logit predicted probabilities for all three levels of injury severities were closer (on average) to the observations than the ones predicted by the latent class model. Only a few notable differences in the inferences were found between the two models.