(iPoster) Correcting Nonresponse and Social Desirability Bias in Online Surveys

• In Event: Poster Sessions: Political Methodology & Foreign Policy I – Great Power Competition and Strategic Rivalry
  In Poster Session: Poster Session: Political Methodology

Abstract

Survey-based estimates of politically sensitive behaviors, such as voter turnout, are frequently biased upward due to a combination of nonresponse bias and social desirability bias (SDB). A central challenge is that survey nonresponse is often missing not at random (MNAR), rendering standard weighting adjustments insufficient.

This study proposes a unified estimation framework that simultaneously corrects for MNAR nonresponse bias and SDB in online survey data. Building on recent methodological advances and evidence from our pilot studies, we integrate Bailey’s (2025) Randomized Response Instrument (RRI)–based inverse probability weighting (MNAR-IPW) estimator with list experiments. RRIs, implemented here through small participation bonuses, affect respondents’ propensity to participate in a survey while remaining orthogonal to the sensitive outcome of interest, thereby enabling identification of nonresponse bias under MNAR. List experiments, in turn, mitigate SDB by allowing respondents to report sensitive behaviors indirectly. Specifically, we construct a list-experiment–based proxy for the sensitive outcome and incorporate it into the RRI-based MNAR-IPW estimator, yielding a unified estimator that corrects for nonignorable nonresponse at the participation stage and social desirability bias at the measurement stage. We field a large-scale online survey experiment in Japan to evaluate the performance of our approach by comparing corrected estimates to population benchmarks derived from official statistics.

Overall, this study offers a practical and theoretically grounded approach to addressing two pervasive sources of bias in survey research. The proposed framework is broadly applicable to the measurement of sensitive behaviors in contexts where both nonresponse and social desirability concerns are present, thereby contributing to more accurate population-level inference.

Authors

• Masataka Harada, Tohoku University

• Tetsuya Matsubayashi, University of Osaka

• Ikuma Ogura, Hitotsubashi University