Reliability of demographic variables

Reliability reflects the extent to which subjects provide consistent responses to identical or similar questions posed at different times. Reinterviews in a panel study are an ideal way of assessing the reliability of data. The 17-year follow-up interviews with teenage mothers in Baltimore afford an opportunity to check the consistency of selected items from the 1984 interview with data collected in earlier interviews.

An analysis of reliability must focus on the comparability of both the questionnaire items themselves and the respondents' answers to these items. Ideally, questions from the different interviews are identical. If a question asks about a single event occurring at one point in time, reports of that event should not change retrospectively. Parallel questions will produce inconsistent data only because of respondent or interviewer error. Since interviewer discrepancies were usually corrected in the cleaning and coding process, our main concern in this appendix is respondent error.

Lower reliability may result from a respondent's error due to problems of recall, motivational factors, communication, or knowledge (Sudman and Bradburn, 1982). Recall is an important factor in any question about behavior; over time, a respondent will become less certain of the timing or detail of events. Some degree of uncertainty due to problems of recall should be expected and is evident in our most recent data. Motivation is an important cause of error in responding to questions that are threatening to a respondent.