What is the minimum sample size necessary to estimate the proportion of automobile accidents that are caused by an intoxicated driver, if we want to be 95% confident to within 2% of the true proportion?