Lawrence Berkeley National Laboratory
Assessment of DRI’s Two-Stage Logistic Regression Model Used to Simultaneously Estimate the Relationship between Vehicle Mass or Size Reduction and U.S. Fatality Risk, Crashworthiness/Compatibility, and Crash Avoidance:
- Author(s): Wenzel, Tom, P
- et al.
This report summarizes an effort to replicate the results from a 2-stage regression model developed by Dynamic Research Inc. (DRI) to simultaneously estimate the effect of mass or footprint reduction on the two components of societal fatality risk per vehicle miles of travel, crashes per VMT (crash frequency) and fatality risk once a crash has occurred (crashworthiness/ compatibility). Lawrence Berkeley National Laboratory (LBNL) was not able to exactly replicate the results from DRI’s simultaneous 2-stage regression model. This may be because of discrepancies in how DRI and LBNL classified the state police-reported crash data into crash types. LBNL’s analysis of four alternate regression models suggests that the results from DRI’s method are sensitive to changes in what data are used in the analysis, or even the particular vehicles included in the decimation sample; in some cases the sign of the estimated relationship from DRI’s results changes under an alternate LBNL regression. However, for the most part LBNL’s alternate regressions confirm the general results from DRI’s simultaneous model, and LBNL’s analysis in its Phase 2 report: that mass reduction is associated with an increase in crash frequency (crashes per VMT), but a decrease in fatality risk once a crash has occurred, across all vehicle types. Similar results were obtained after using stopped rather than non-culpable vehicles as the induced exposure records, and replacing footprint with wheelbase and track width.