ON THIS LINK

Easy, says Sapolsky. Let's say 10% have done this taboo sexdeed, 90% haven't. Tell your subjects that those who have done the taboo sexdeed should answer the question honestly, ie confess to it. Add that those who haven't done the taboo sexdeed should toss a coin. If it comes up heads, they should say they have done the sexdeed, whether they really have or not. If it comes up tails, they should say they haven't, again regardless of what they've actually done.

This randomizing process will allow the people who have done the taboo sexdeed to confess it, knowing that they have an alibi, should they be challenged, in the shape of the coin. As long as they understand the rules, everyone who did the taboo sexdeed should confess. The researcher then just needs to double the number of people who say they haven't done the non-taboo deed -- everyone else has done the taboo deed.