Abstract
Despite rigorous safety standards, school buses in the United States still experience around 26,000 crashes annually, resulting in approximately 10 fatalities, with a fatality rate persistently static despite advances in vehicle safety. Such crashes often have significant implications, particularly for the young passengers involved. This study utilized a novel approach by analyzing Texas Crash Records Information System (CRIS) data from 2017 to 2022 through a Random Parameter Logit Model with Heterogeneity in Means (RPLHM) and Correlated RPLHM (CRPLHM). This method allows for a detailed examination of unobserved heterogeneity and specific variances within the data, enhancing the understanding of the complex dynamics influencing crash severity. The analysis revealed that crashes on state highways typically presented a lower likelihood of fatal and severe crash outcomes. Additionally, demographic attributes such as age significantly impacted crash outcomes, with middle-aged drivers (25-54) often experiencing less severe injuries. Additionally, driver inattention was associated with an increased occurrence of no-injury crash outcomes. While daylight is associated with less moderate and possible injury crashes, clear weather was associated with higher no-injury crashes. The transferability tests revealed temporal instability in yellow school bus crash severity patterns across 2017-2022. Key variables such as intersections, daylight, and driver characteristics demonstrated varying effects over time. While morning and afternoon crashes increasingly reduced the likelihood of fatal and severe injuries in later years, factors like divided roadways and clear weather saw greater variability in their impact on no-injury and moderate injury outcomes. These findings highlight the importance of year-specific modeling and support data-driven policymaking to improve school bus safety.
