Limited Self-Knowledge and Survey Response Behavior

We study response behavior in surveys and propose a method to identify and improve the informativeness of survey evidence. First, we develop a choice model of survey response behavior under the assumption that responses imperfectly reveal respondents' characteristics due to limited self-knowledge, inattention, or lack of engagement. Respondents receive individual-specific signals about their characteristics and choose their responses accordingly. We identify the conditions under which this process leads to biased inference from survey evidence and demonstrate how focusing on respondents with high signal precision mitigates bias. Importantly, we show that a respondent's signal precision can be inferred from observed response patterns. Second, based on these insights, we develop a consistent and unbiased estimator for a respondent's signal precision. Third, we provide experimental and survey evidence concerning the performance of the model and estimator. We experimentally test the model's key predictions in a context where the researcher knows the true characteristics. The data confirm both the model's predictions and the estimator's validity. Using a large survey, we show how our estimator can be used to improve survey evidence. Our estimator significantly increases the explanatory power of self-assessments and their association with behavior, and performs well relative to alternative methods proposed in the literature.