Abstract
Minimizing error in self-reported sexual behavior could reduce investigators' likelihood of rejecting truly successful interventions to decrease HIV and STI transmission risk. Sexual behavior assessments can elicit problematic data. This may manifest in the form of elevated levels of non-response, inaccurate point estimates, or misclassification errors resulting in inappropriately answering or, perhaps more importantly, skipping questions. We programed conversational interviewing elements into 20 sexual behavior questions in an exit survey of gay bathhouse patrons (N = 459) administered using ACASI. Those elements, called alternate pathways, included follow-up questions to responses to confirm that operational definitions were applied in the answer (with return to the initial question if confirmation failed), and assurances of confidentiality and requests for best guesses in reaction to non-response (including "don't know"). These elements were invoked in nearly 10% of participants, and approximately 74% of all invocations resulted in a usable numeric response, or 87% if the data need only estimate prevalence. Almost two-thirds of the problematic data issues occurred in answers to sexual contact questions, with others related to follow-up questions about specific sexual behavior. It is at this level of important filtering questions where the benefits of the approach are likely to be maximized.