Differences in reported car weight between fatality and registration data files

Two national-level data sources are commonly used together to estimate and compare fatality rates by car weight. The weight of each car in a fatal crash is available on the automated files of the National Highway Traffic Safety Administration's Fatal Accident Reporting System; weight is derived by interpreting the Vehicle Identification Number of each car using a computer algorithm developed and maintained by R. L. Polk & Co. Counts of cars in use, by weight, are available on R. L. Polk & Co.'s National Vehicle Population Profile files; weights are coded from information in state vehicle registration files. However, it appears that there are systematic differences in car weight coding that complicate the use of these two sources together for calculating fatality rates (fatalities per registered car). Overall, the registration data appear to describe a car (of a particular make, model, and model year) as about one hundred pounds heavier than that car is described in the fatality data. The effect is to bias the comparison of fatalities per registered vehicle against lighter cars. Failure to consider this difference can lead to very misleading results. For example, the uncorrected data produce an estimate that the number of occupant fatalities per registered minicompact car (those under 1,950 pounds) was five times the rate in the largest cars (those weighing at least 3,950 pounds). Correcting for differences in car weight reporting produces estimates that the fatality rate in minicompact cars was twice that in the largest cars. Differences by car weight remain, but they are much less than would be concluded from the biased comparison.