Unobserved Heterogeneity and Temporal Instability in the Analysis of Work-zone Crash-injury Severities

Document Type


Publication Date



work zone crashes, temporal stability, injury severity, mixed logit model with heterogeneity in means and variances

Digital Object Identifier (DOI)


In the state of Florida, work-zone related crashes and their resulting injury severities have been increasing recently, particularly over the 2015 to 2017 time period. In the current study, we seek to provide insights into the factors that have been influencing this trend. Using work zone data from the 2012 to 2017 time period, resulting driver-injury severities in single-vehicle work zone crashes were studied using random parameters logit models that allow for possible heterogeneity in the means and variances of parameter estimates. The available data included a wide variety of factors known to influence driver injury severity including data related to the crash characteristics, vehicle characteristics, roadway attributes, prevailing traffic volume, driver characteristics, and spatial and temporal characteristics. The model estimates produced significantly different parameters for each of the year from 2012 to 2017, and a fundamental shift in unobserved heterogeneity, suggesting statistically significant temporal instability. In addition, in several key instances, the marginal effects of individual parameter estimates show marked differences between one year and the next. However, these findings may not be the sole result of variations in driver behavior over time as has been argued in past research that has found temporal instability. This is because each work zone has a unique set of characteristics and, with the sample of work zones changing from one year to the next as highway maintenance and construction is undertaken in different locations, this work-zone sample variation could be a substantial source of the observed temporal instability.

Was this content written or created while at USF?


Citation / Publisher Attribution

Analytic Methods in Accident Research, v. 28, art. 100130