Correcting sample selection in FARS data to estimate seatbelt use.

INTRODUCTION In this paper, we show that FARS data can be a comparable alternative to observational NOPUS data in estimating seat belt use in the United States once we correct for sample selection bias. RESULTS Based on assumptions of independence for seatbelt choice, we establish a lower and upper bound for seatbelt usage rates, and find that once we correct for sample selection bias, the seatbelt usage estimates from the corrected FARS emerge at least as a comparable alternative to NOPUS estimates. IMPACT ON INDUSTRY This implies that researchers can use corrected FARS to complement NOPUS, thus being able to utilize the rich cross-sectional details available in FARS data to analyze various relevant research questions.