This is a difficult question to answer accurately. Surveys are unreliable because people are not likely to be truthful about behavior that is considered socially unacceptable. Surveys under-report the true percentage.
Even if it is difficult, estimating the percentage of people who engage in a specific illegal or frowned-upon behavior is critical as it has important uses in policy-making and enforcement. Research studies have shown some improvement in accuracy by asking participants to estimate the percentage themselves. People are more likely to be open about what others are doing than they are about themselves. Obviously this works best for populations that are well-known by the participants, such as employees in a company.
Survey researchers have come up with a randomized response technique (RRT) that encourages participants to be more open. The experimental design is as follows:
- Before answering a question, the participant throws a single dice which the researcher cannot see.
- The participants answers ‘yes’ if a 1 comes up and ‘no’ if it is a 6, regardless of the actual answer to the question.
- For all other numbers, the participant answers the question truthfully.
Because a ‘yes’ response does not necessarily mean the person actually engaged in the undesirable behavior, people are more honest. Admittedly, the forced yes and no means at least of 1/6 of the answers are incorrect but, despite introducing this statistical noise, the overall results give better answers.
Introducing error to increase accuracy seems counterintuitive but it works in multiple domains. As an example, comparing RRT results to drug screening on hair samples shows that it is 30% better than traditional techniques.
Now go wash your hands. I bet you forgot.