Temporal Stability of Factors Influencing Driver-Injury Severities in Single-Vehicle Crashes: A Correlated Random Parameters with Heterogeneity in Means and Variances Approach

Undoubtedly, single-vehicle crashes remain a major concern for roadway users and highway
administrators, especially in low- and middle-income developing countries, where
accident death rates remain extremely high. This study investigated the temporal instability
of contributing factors of driver-injury severities in single-vehicle crashes using data in
Thailand, a developing country, from 2011 to 2017. The uncorrelated and correlated random
parameters model, which enable a possible heterogeneity in means and variances
approaches, were estimated for individual year model using two levels of driver-injury
severities, namely, no/minor injury and severe/fatal injury. The models considered a wide
range of factors, such as driver, roadway, vehicle, crash, environmental and temporal, and
spatial characteristics. The series of likelihood ratio test and the marginal effect of significant
factors were computed to explore the temporal stability of the year models and to
investigate the temporal instability of the effect of each parameter estimate on the probability
of driver-injury severities within given time periods, respectively. The result indicates
that a substantial temporal instability exists in the model specifications and
estimated parameters (temporally unstable factor included male driver, driving using
exceeding speed limit, crashes on asphalt pavement, crashes on weekends, and crashes
on weekend during nighttime with present of road lighting) across the time periods under
study (despite insignificant in particular year models, some factors were stable but marginal
effects varied across time). The findings may be used to assist and guide decision
makers in policy generation for plans to mitigate driver-injury severities. Despite the
unclear source of temporal instability, the finding emphasizes the importance of the temporal
instability of the factors that influence the outcomes of driver-injury severities.
Alternatively, ignoring temporal instability in studies on crash severity may lead to high levels of bias and inaccurate conclusions. With regard to methodologies, both random
parameters with heterogeneity in means and variances and correlated random parameters
with heterogeneity in means approaches are promising methods with ability to offer
another layer of insight into unobserved heterogeneity in injury severities research.