A latent class analysis of single-vehicle motorcycle crash severity outcomes

• We examined the factors affecting single-vehicle motorcycle crash severity outcomes.
• We estimated a latent class multinomial logit model to address unobserved heterogeneity.
• Two distinct classes of crashes with homogeneous attributes were identified.
• The estimated parameters differed in sign, magnitude or significance across classes.
• The model fit and estimation results underline the need for segmentation of crashes.

Unobserved heterogeneity has been recognized as a critical issue in traffic safety research that has not been completely addressed or often overlooked, and can lead to biased estimates and incorrect inferences if inappropriate methods are used. This paper uses a latent class approach to investigate the factors that affect crash severity outcomes in single-vehicle motorcycle crashes. Motorcycle crash data from 2001 to 2008 in Iowa were collected with a total of 3644 single-vehicle motorcycle crashes occurring during that time period. A latent class multinomial logit model is estimated that addresses unobserved heterogeneity by identifying two distinct crash data classes with homogeneous attributes. The estimation results show a significant relationship between severe crash injury outcomes and crash-specific factors (such as speeding, run-off road, collision with fixed object and overturn/rollover), riding on high-speed roads, riding on rural roads, riding on dry road surface, riding without a helmet, age (riders older than 25 years old) and impaired riding (riders under the influence of drug, alcohol or medication). The model fit and estimation results underline the need for segmentation of crashes, and suggest that the latent class approach can be a promising tool for modeling motorcycle crash severity outcomes.