Does All This Really Make a Difference?
Yes, it does. Thurman et al. (1993) were interested in the attitudes and self-reported behaviors of people who admit to drunk driving. Using Dillman’s TDM, they sent out questionnaires to a national sample of 1,310 and got back 765, or 58%. Not bad for a first pass, since you can generally expect about 25% to 30% from the first wave. Unfortunately, for lack of time and money, Thurman et al. couldn’t follow through with all the extra mailings.
Of the 765 respondents, 237 said they were nondrinkers. This left 525 eligible questionnaires for analysis. Of the 525 respondents who said they were consumers of alcohol, 133 admitted driving while drunk in the past year. Those 133 respondents provided data of intrinsic interest, but the 765 people who responded from the nationally representative sample of 1,310 may be a biased sample on which to base any generalizations. I say ‘‘may be’’ a biased sample because there is no way to tell. And that’s the problem.
The bottom line: The last interview you get in any survey—whether you’re sending out questionnaires, doing a phone survey, or contacting respondents for face-to-face interviews—is always the most costly and it’s almost always worth it. If you really care about representative data, you won’t think of all the chasing around you have to do for the last interviews in a set as a nuisance but as a necessary expense of data collection. And you’ll prepare for it in advance by establishing a realistic budget of both time and money.