we want to construct a 99% confidence interval for the proportion of people who will vote 'yes' on a ballot question in the next election. how many people should we sample to construct such an interval with a margin of error of 3%?