I can work this one out of paper, but that would be a tough interview question to answer on the fly.
It would have to have a particular set of questions: 1. Have you smoked in the last 30 days (valid because if they quit a week ago, it may be a temporary quit, reliable because they can probably remember back that far). 2. Have you smoked on a regular basis at anytime in the last 10 years? 3. What is your career (if you're data mining in a large survey with more than just doctors, skip if sent only to doctors). 4. If doctor, what area of medicine do you practice in? (to check that the sample is representative of the medical community). 5. Other demographic questions for the same reason as 4.
Mailing surveys has the problem of non-responders. They may fit certain patterns. One would think that doctors in some specialties would be more likely to not respond than others, and perhaps doctors who smoke may be more ashamed of it than other smokers, and less likely to want to complete a survey specifically about their smoking. Hence question 4, and utilizing an additional source of information:
Data mining is cheap and may tell you how many doctors smoked back then, and how many doctors smoke now, but who knows if that data would be available.
Ideally you'd data mine something large like the General Social Survey (though I suspect that particular one wouldn't be adequate) AND send out surveys. Then you can compare results from the two to see if they are consistent.
Anything anybody would add to or subtract from what I said?
