These data don't account for
These data don't account for economics (who is more likely to buy a 4WD car); geography; buying habits, and risk-tolerance (are more safety-conscious people buying 4WD, therefore driving more safely?) Correlation is not causation. Too many confounding variables to be meaningful.