Does Help Help? An Empirical Analysis of Social Desirability Bias in Ratings
研究揭示在线评论中的“评中评”功能会引发社会期望偏差，导致约7.4%的评分者调整评分以获取更多有用反馈，并提出了减轻偏差的方法。
Review-in-review (RIR) is a feature that allows viewers to generate positive or negative evaluations for primary quality evaluations of a product (e.g., ratings and reviews). This study reveals that it can cause social desirability bias in primary ratings: Reviewers who desire social recognition are driven to adjust their ratings (about 7.4% likelihood) to elicit more helpful responses and avoid unhelpful ones. This bias can be shown as distorted conformity to the prior rating distribution or extremity, depending on the RIR types. The model identifies how bias magnitude correlates with users’ social characteristics, thereby identifying vulnerable individuals. Platforms can incentivize less vulnerable users and remind susceptible ones to decrease the bias and can supplement rating conditional on the identified vulnerability extent (e.g., the distribution by the “independent” raters) to mitigate the bias’s impact on rating viewers. The simulation analysis compares the bias under different counterfactual RIR system designs, finding a composite RIR system (e.g., helpful and unhelpful RIRs) partially neutralizes the bias, obviating the need to remove all RIR features. The model further adapts to evaluate underexplored RIRs forms and can provide a “de-biased” metric while preserving individual ratings.