I'm looking at the alcohol use questions on the WLS*. A question asked in 1992 and 2004 is whether respondents have ever gone to anyone for help about their drinking. A little more than 3% of respondents said "yes" to this question in 1992. But here's the thing that makes survey research so fun: of these people who said "yes" to the question in 1992, almost a third (31%) say "no" in 2004. In other words, almost a third of the people who say they have sometime sought help for their drinking in 1992 deny that they have ever done so in 2004.
One thing I wondered is if these new-deniers are more likely to have relapsed since 2004 and so are drinking more than the people who were consistent in saying they had sought help in 1992 and 2004. Sure enough, the new-deniers are more likely to be drinking and, among those who are drinking, are drinking more. But wait, they were already drinking more in 1992, by almost exactly the same amount. So there goes that idea.
And then all this raises the puzzle of how you should think about people who said "no" they had never sought help for their drinking in 1992, but say "yes" now in 2004. The logically consistent conclusion, of course, is that these are people who got help in the last 12 years. Some of them probably are. But some are probably people who will, apparently, tell you sometimes in surveys that they have sought help for their drinking in the past, and on other occasions will deny this. If the number of no-then-yeses was much greater than the yes-than-nos, you could draw the happy conclusion perhaps that MOST of the no-then-yeses are people who have recently sought help. Alas, as things stand now, the number of no-then-yeses is actually fewer than the number of yes-then-nos, which seems almost to border on the absurd and is the kind of thing that causes the conscientious survey data analyst to bang his head on his desk.
* [W]isconsin [L]ongitudinal [S]tudy, a survey that has followed a 1/3 sample of all 1957 Wisconsin high school graduates.