Using discordant question types to collect sexual behavior, attitude, and intention data can alter effect measure estimates

Background: Question framing is important when measuring sexual behaviors, attitudes, and intentions— “If you do not use a condom, how likely is it that you could get HIV from having sex?” yields different responses than “…from having sex with [NAME]?” Objective: To estimate the magnitude of changes in effect measure estimates that results from using partner-specific factors to predict general sexual behavior outcomes and vice-versa. Methods: 481 young adults completed structured interviews in which subsets of items related to sexual behavior, attitudes, and intentions were worded in both partner-specific and general formats. We used the percentage agreement index to assess intra-person agreement between partner-specific and general questions. Binary regression was used to control for appropriate confounders while estimating a series of risk ratios (RR); we examined change in RR related to use of partner-specific versus general wording. Results: Intra-person agreement on partner-specific vs. general items ranged from 25.2% to 77.8%. Effect measure estimates differed depending on whether the predictor and outcome were the same question type; in some cases the direction of the effect was reversed. For example, the RR for “importance of avoiding pregnancy” as a predictor of partner-specific condom use changed from 0.77 [95% CI: 0.30, 1.99] for the partner-specific predictor to 3.10 [0.95, 10.14] for the generically-worded predictor. Additional results will be presented on other predictor/outcome pairs. Conclusions: The choice to use partner-specific or general questions depends on the research aims. Public health researchers and practitioners should be aware that partner-specific and general questions yield different results.