Correcting Nonresponse Bias in Mail Questionnaires

A randomly selected sample of men was drawn f~om telephone directories in the New York metropolitan area. Two hundred were contacted by telephone and interviewed. Three months later, a four-page questionnaire was mailed to these 200 men. It repeated 14 items they had been asked on the telephone and included an additional eight filler questions on topics of general interest. Seven of the 14 repeated items became the variables on which the study is based. These included three demographic items (years in school, family size, age) and four nondemographic items (supervisory level-is the respondent employed at the supervisory level or abov.e? XYZ readership-does he read XYZ, a popular daily newspaper? car ownership-does he own an automobile? ABC's ad awareness-does he recall, aided, the advertising for ABC, a national shoe manufacturer?). A covering letter asked for cooperation, and a stamped self-addressed return envelope was enclosed. Respondents were not asked to identify themselves, so the questionnaires were blind-coded. Twenty-four questionnaires were returned for incorrect address. Of the 176 subjects left, 59 responded (34%) and 117 did not (66% ). For the purposes of the study, then, the group of 176 is defined as the population and the 59 respondents as a sample drawn from that population. The telephone survey provides benchmark statistics of the population. The answers tabulated from the 59 respondents were then compared with known characteristics of the population to determine the accuracy of the mail survey. One