Truck Drivers and Traffic Fatalities: Estimating the Value of a Statistical Life Using Panel Data

The value of a statistical life (VSL) literature primarily utilizes cross-sectional analyses that are susceptible to bias from unobserved time-invariant factors. Further, many occupational death data have considerable non-classical measurement error. This paper addresses these two shortcomings by using truck driver panel data from 1979 to 2002 to bound VSL estimates. Measurement error will attenuate Fixed Effects and First Differences estimates. It will also anti-attenuate Instrumental Variable (IV) estimates when another mismeasured risk measure is used as an instrument. Both strategies control for unobserved time-invariant factors, and examining wages within a competitive occupation controls for occupation-specific productivity shocks, worker risk tolerance and worker heterogeneity. This study also uses a time-aggregated cross-section estimator to investigate bias due to within-occupation aggregation. Risk measures are created from the National Highway Traffic Safety Administration—Fatality Analysis Reporting System (NHTSA-FARS), the census of all traffic fatalities on public U.S. roadways. The OLS-IV bounding strategy returns a VSL range of $0.4 million to $4.1 million. The timeaggregated cross-section estimator returns $8.2 million and $8.8 million, in the middle of the VSL literature’s between-occupation estimates. This suggests upward bias of at least $4.1 million when aggregating heterogeneous workers. Although not externally valid, this paper illustrates the need to control for time-invariant factors. When estimating a VSL, future research should reduce sample measurement error and control for time-invariant factors with panel data.