Tone at the Bottom: Measuring Corporate Misconduct Risk from the Text of Employee Reviews
研究从Glassdoor员工评价中提取文本信息，构建衡量企业不当行为风险的指标，发现这些指标能有效区分高风险与低风险公司，并提升对实际不当行为的预测能力。
This paper examines whether information extracted via text-based statistical methods applied to employee reviews left on the website Glassdoor.com can be used to develop indicators of corporate misconduct risk. We argue that inside information on the incidence of misconduct as well as the control environments and broader organizational cultures that contribute to its occurrence are likely to be widespread among employees and to be reflected in the text of these reviews. Our results show that information extracted from such text can be used to develop measures with useful properties for measuring misconduct risk. Specifically, the measures we develop clearly discriminate between high- and low-misconduct-risk firms and improve out-of-sample predictions of realized misconduct risk above and beyond other readily observable characteristics, such as Glassdoor firm ratings, firm size, performance, industry risk, violation history, and press coverage. We provide further evidence on the efficacy of our text-based measures of misconduct risk by showing that they are associated with future employee whistleblower complaints even after controlling for these same observable characteristics. This paper was accepted by Brian Bushee, accounting.