The consequences of self‐reporting biases: Evidence from the crash preventability program
研究企业自我报告数据中的内生选择偏差，以美国交通部政策变化为例，发现大型运输公司报告事故无责的概率是独立司机的40倍以上，导致安全评估失真。
Abstract Relying on firms to self‐report information is an information‐gathering mechanism that often results in biased measures due to the incentives of the reporting firms. What is less commonly understood is that using self‐reported information for decision‐making results in endogenous selection bias, which creates spurious associations between the measure being reported and factors that influence reporting. Thus, conditioning on self‐reported information can lead to inaccurate evaluations of firms and bias predictions of future performance, even when the self‐reported measure is not intentionally misrepresented. We examine endogenous selection bias in self‐reporting regimes using directed acyclic graphs (DAGs). We illustrate the problem using data from a policy change by the U.S. Department of Transportation that allowed firms to report not‐at‐fault for accidents. We find that large for‐hire firms are much more likely to report not‐at‐fault for accidents—over 40 times more likely than independent drivers—even after controlling for time, location, and weather. When comparing independent drivers with large firms, the reporting disparities make large firms appear 25% safer when using at‐fault accidents versus all accidents while providing no improvement in predicting future accidents. This study highlights the consequences of poorly designed information‐gathering mechanisms and the usefulness of DAGs for understanding causality in supply chain research.