A study can be designed to account for embarrassing and/or incriminating questions.
Say the question is about "illegal file-sharing" and Yes means the subject has participated in this behavior in the past and No means the subject has not. Under private conditions, have the subject perform the following:
1. Flip a coin.
2. Have the subject answer Yes if they have participated in the illegal sharing of files.
3. If the subject has not done this activity, have them answer No only if they flipped 'tails' in step #1. Otherwise, if the coin came up 'heads', they answer Yes.
The true Yes proportion in the survey population of size n can then be determined by (Y_count - N_count) / n.
While that "works", what fraction of survey participants will actually do that?
I suspect that many/most folks who are reluctant to admit that they've done something won't admit it even if you tell them that other people "will" falsely admit to doing said thing so based on a coin flip. That's completely rational because I suspect that many people who haven't done said thing won't say that they have just because the coin tells them to.
Say the question is about "illegal file-sharing" and Yes means the subject has participated in this behavior in the past and No means the subject has not. Under private conditions, have the subject perform the following:
1. Flip a coin.
2. Have the subject answer Yes if they have participated in the illegal sharing of files.
3. If the subject has not done this activity, have them answer No only if they flipped 'tails' in step #1. Otherwise, if the coin came up 'heads', they answer Yes.
The true Yes proportion in the survey population of size n can then be determined by (Y_count - N_count) / n.