In recent years, marketing researchers have become increasingly interested in under- and overreporting. However, there are few suitable approaches to operationalize deviations from the truth, particularly in behavioral domains in which self-reports are usually the only viable method of choice to measure behavior or attitudes. An especially difficult situation arises if some people underreport while others overreport. This article proposes a Bayesian item response theory model to quantify under- and overreporting in surveys. The method utilizes within-person differences between answers obtained under direct questioning (no privacy protection) and randomized-response questioning (which ensures item-level privacy protection). This method has the important features of incorporating behavioral response-mode effects (e.g., privacy loss when switching from direct to randomized-response questioning, response-mode inertia effects) and allowing the direction of bias to differ across respondents. The authors provide an empirical application for excessive alcohol consumption involving 1,408 respondents from a commercial web panel. The results show that respondents are averse to decreases in privacy and that randomized response is less effective if respondents provide biased responses to earlier direct questions.
de Jong, M. G., Fox, G. J. A., & Steenkamp, J-B. E. M. (2015). Quantifying Under- and Overreporting in Surveys Through a Dual-Questioning-Technique Design. Journal of marketing research, 52(6), 737-753. https://doi.org/10.1509/jmr.12.0336