This document discusses the importance of controlling for self-reported exposure in traffic accident prediction studies. It notes that accident risk increases with increased driving exposure. However, most studies do not control for exposure beyond calendar time. Self-reported exposure data is unreliable and can create spurious associations. The study analyzed data from over 24,000 drivers, controlling for self-reported mileage. It found a 20-30% reduction in the explained variance of crashes and offenses when controlling for mileage. The conclusion is that exposure has a strong effect on crash prediction but is often not properly controlled for, leading to overestimation of the effects of other variables.