Lay's Potato Chips recently had a flavor contest. People could vote on one of the four finalists flavors: wavy mango salsa, wasabi ginger, bacon mac & cheese, and cappacino. (The voting closed October 21st; wasabi ginger won.) Lay's naturally has an incentive to find out who is likely to win; that way they can ramp up the production of the new chip as soon as possible. For fun, let's assume they don't have access to all the data; they can only rely on samples.

Suppose, based on a sample of 100 people, wasabi is in the lead with 40% in favor. It's very important you capture the true result in your range or you'll look really stupid at the meeting after the votes are counted (this means you should set p to 0.5). At 99% confidence, what is the margin of error?