Martin, Steven P., and Lawrence Wu
Working paper no. 1999-10
Abstract
In this study we estimate error rates in two event calendars in the 1995 NSFG – a five year contraceptive use calendar and a five-year sexual nonintercourse calendar. We superimpose each event calendar over respondents’ birth histories and count implausible combinations of fertility events, sexual activity, and contraceptive use. We then evaluate how similar errors might affect summary measures based on the event calendars. Finally, we identify characteristics of respondents, interviewers, and interviews that correlate with high detectable error rates.
Our results suggest that researchers should use the NSFG event histories cautiously. We find a small but significant rate of detectable errors in both of the event calendars we study. Unfortunately, events in the event calendars are distributed so that even a few response errors can severely impair some types of studies (such as studies of contraceptive failure). In addition, response errors correlate with some important background characteristics of the respondents; this may hamper some comparative studies of contraceptive use or sexual nonintercourse. Lastly, we identify characteristics of interviews and interviewers that are associated with detectable errors. Such associations may suggest ways to reduce response errors in the future.