Effects of question format and collection mode on the accuracy of retrospective surveys of health risk behavior: a comparison with daily sexual activity diaries.

OBJECTIVE: To use prospective diaries to assess the effects of alternative survey question formats and data collection modes on the accuracy of retrospective self-reports of sexual behavior for the same time period.
DESIGN: Over a 3-month period, 493 adults completed and returned by mail daily diaries of their sexual activities. Participants then returned to complete a retrospective survey of their sexual activities during the same 3-month time period. Participants were randomized to 1 of 6 retrospective experimental survey conditions that represented the combinations of 2 question formats (in aggregate and by partner) and 3 collection modes (paper questionnaire, computer-assisted self-interview [CASI], and audio-enhanced CASI). MAIN OUTCOME MEASURES: Frequencies of sexual activities and condom use reported in surveys were compared with what participants had recorded in their prospective daily diaries.
RESULTS: Relative to activity recorded in daily diaries, participants more often underreported than overreported (61% vs. 28%, p < .05) sexual behavior in retrospective surveys. Greater survey-to-diary consistency in reported behavior was achieved with by-partner question format (p < .032) and CASI collection modes (p < .05) used in surveys. However, median rates of error in survey-to-diary comparisons of reported behavior were as high as 40%.
CONCLUSION: Even after refinements in survey question format and use of CASI techniques, there remained substantial error in participants' retrospective reports of sexual practices. These error rates may be comparable to, or even larger than, the expected effect size of some intervention studies. Further research is needed to improve the accuracy of sexual behavior reports.