ABSTRACT
Social surveys typically have missing data, because the intended recipients cannot be reached or refuse to respond. How should such data be analyzed?
One response, all too common, I fear, is simply to ignore the issue, and assume that those who answered are typical of those who did not. There is usually no warrant for this belief, and indeed it is rarely even broached, let along discussed. Another extreme might be to refuse to extend the claims from the population that responded to the population to whom the survey was made available. Neither of these extremes seems adequate to the needs of social science.
