[msg]

One interesting thing you can do with such a questionnaire is a coin flip survey (sorry, I don't know the real name). Give the survey participants a list of sensitive true/false questions and a coin. Instruct them to flip the coin twice, concealing the results from the survey-giver. Then have them follow this algorithm:

  if (first flip == heads):
    return second flip == heads
  else:
    return truthful T/F answer to sensitive question

When results are aggregated, roughly half the participants will have answered truthfully, and the other half are uniformly distributed. None of the individual participants have tipped off whether they did something bad, but in aggregate they have truthfully answered the survey.

[delinka]

While this operation may be statistically sound, it requires work. Humans are notoriously lazy when it comes to anything other than what they want to do. You'd be hard pressed to get each respondent to actually do this repetitively for an entire survey. Even if it had only three questions.

[msg]

For $50? College students have done much weirder and stupider stuff for less money than flipping a coin twice and answering a straightforward T/F question, even if it were twenty questions. The technique has been around for a long time. You should consider it a solved problem.

Finally googled the names. I also discovered the original formulations do it with only one coin flip (or one roll of a die).

http://en.wikipedia.org/wiki/Randomized_response

http://en.wikipedia.org/wiki/Unmatched_count

[delinka]

If you're offering $50 in cash to answer some questions and flip some coins, sure you'd get people to answer loads of them. But you're not offering an incentive to answer the questions.