Models using survey data typically assume that respondents report their behavior accurately. In fact, however, these models may be flawed due to bias and imprecision in the survey data collection process. The sources of error of self-reports are numerous, reflecting a combination of the characteristrics of individual respondents and the nature of the survey instrument and the way it is administered. This research focuses on gender differences in self-reporting on several topics, including sexual behavior, where primary emphasis is put, illicit drug use and abortion. It will be carried out in conjunction with the investigators' National Center for Health Statistics project entitled, "Cognitive Research on Health Promotion and Disease Prevention". Numerous surveys, both in the United States and in other countries, have shown that men report more sex partners than women do. This difference persists even when homosexual partners are excluded and when adjustments are made for extreme values, item non-response, population size, and related matters. The differences become larger as the time frame covered by the questions grow longer. The most plausible explanation for the difference is that the two genders differ in the direction and/or magnitude of their response errors -- either the men are overreporting, the women underreporting, or both. A series of related studies will be carried out to examine these gender differences in the reports of sexual partners as well as reports on other sexual behaviors, drug use, and abortions. They consist of 1) cognitive interviews exploring the understanding of sexual terms and willingness to discuss sexual matters, 2) a bogus pipeline experiment to test whether assumed validation of self-reports will change the reports of men and women, 3) a survey experiment to test differences in response format, recall cues, and interviewing staff, 4) a second survey experiment to test differences in mode of administration, place of interview, questionaire context, and question wording, and 5) a third survey experiment to test anonymity vs. confidentiality. These studies should identify the factors that lead to misreports of sensitive behaviors about sexual partners, abortions, and drug use; contribute to the design of an optimal measurement strategy; and provide improved estimates of the true incidence levels of these sensitive behaviors. It is crucial that work of this sort be conducted, particularly in view of questions raised by the AIDS epidemic and the potential extent of its applications.
