Adjusted Crash Odds Ratio Estimates of Driver Behavior Errors: a Re-analysis of the Shrp 2 Naturalistic Driving Study Data

Dingus and colleagues recently estimated crash odds ratios (ORs) for “driver behavior errors” (hereafter, “Behaviors”) in the Strategic Highway Research Program Phase 2 naturalistic driving study. Behaviors are illegal, improper, aggressive, and/or reckless driving maneuvers. For example, the Dingus study OR estimate for “Speeding over limit and too fast for conditions,” (hereafter, “Speeding”), was 12.8, with a 95% confidence interval (CI) from 10.1 to 16.2. The current study identified four issues in the Dingus study. First, heterogeneous Behaviors were pooled; e.g., “Exceeded speed limit,” and “Exceeded safe speed but not speed limit” were apparently improperly pooled to form Speeding. Second, exposed drivers often had other Behaviors in the same time window, but unexposed drivers had none, a selection bias that inflated Behavior ORs by 30%. Third, impairments were not filtered out. Fourth, secondary tasks were not filtered out, creating a confounding bias that deflated Behavior OR estimates by 50%. To correct these issues, the current study stratified the heterogeneous categories, then filtered out other Behaviors, impairments, and secondary tasks. “Pure Behavior” (no other Behaviors, secondary tasks, or impairments) was thus compared to “Pure Driving” (no Behaviors, secondary tasks, or impairments). The Pure OR estimate for “Exceeded speed limit” was 5.4 (CI 2.7-10.1), and for “Exceeded safe speed but not speed limit” was 71.5 (CI 36.0-136.2), both substantially different than the Dingus study Speeding estimate. All Behavior OR estimates in the Dingus study should be similarly corrected and adjusted to improve their validity.