Replication Data for: Response Bias in Survey Measures of Voter Behavior: Implications for Measurement and Inference

This short report exploits a unique opportunity to investigate the implications of response bias in survey questions about voter turnout and vote choice in new democracies. We analyze data from a field experiment in Benin, where we gathered official elections results and panel survey data representative at the village level, allowing us to directly compare average outcomes across both measurement instruments in a large number of units. We show that survey respondents consistently over-report turning out to vote and voting for the incumbent, that the bias is large and worse in contexts where question sensitivity is higher. This has important effects on the inferences we draw about an experimental treatment, indicating that the response bias we identify is correlated with treatment. Although the results using the survey data suggest that the treatment had the hypothesized impact, they are also consistent with social desirability bias. By contrast, the administrative data leads to the conclusion that the treatment had no effect.

本短篇报告依托独特契机，针对新兴民主国家中有关选民投票率与投票选择的调查问题所存在的回应偏差（response bias）展开影响性研究。本研究分析了贝宁开展的一项田野实验数据：我们收集了官方选举结果数据与村级代表性面板调查（panel survey）数据，得以在大量观测单元中直接对比两种测量工具所得的平均结果。研究发现，调查受访者始终存在虚报投票率与支持现任候选人的情况；且在问题敏感性更高的场景中，此类偏差更为显著。这一偏差会对我们基于实验干预（experimental treatment）所得出的推论产生重要影响，表明本研究识别出的回应偏差与实验干预存在相关性。尽管基于调查数据得到的结果显示实验干预达到了预设的影响效果，但该结果也可通过社会期望偏差（social desirability bias）得到解释。与之形成鲜明对比的是，行政数据则得出该实验干预未产生任何影响的结论。