Anonymous company reviews that employees leave online could be used to predict corporate misconduct and potentially head it off, according to new research.
A study from researchers at Harvard Business School and the Netherlands’ Tilburg University found that information extracted from employee reviews left on company-review site Glassdoor.com was useful in predicting misconduct beyond other readily observable factors, such as a firm’s performance, press coverage, industry risk and prior violations.
The reviews offer employees’ observations about companies’ control practices, cultures, operations and performance pressures that can contribute to misconduct risk, says Dennis Campbell, a professor of business management at Harvard Business School, who conducted the research with Ruidi Shang, assistant professor at the Tilburg School of Economics and Management. Listening to that “tone at the bottom” offers an early warning of potential misconduct, he says.
“Our theory is that what leads people to commit misconduct is actually the environment they are in,” says Dr. Shang.
Anonymous reviews
For their study, the researchers extracted information from anonymous reviews of publicly traded U.S. companies on the employee-review site Glassdoor.com from June 2008 to December 2016. They excluded firms that received fewer than 10 reviews during the period.
They then obtained data on the companies, such as their size, capital structure and profitability, and press-coverage data, such as the number of media articles related to each company, from 2008 through 2017. They merged all the data, dropping reviews of firms for which they didn’t have the necessary variables or data—such as those that went out of business or were acquired. Their final sample consisted of 13,363 observations about 1,478 companies.
Finally, they extracted all 26,934 corporate misconduct cases committed by public U.S. firms from 2008 through 2017 from Violation Tracker, a search engine that covers civil and criminal cases brought against firms. That allowed them to determine which words came up disproportionately in reviews of firms that were found guilty of misconduct.
Using machine-learning techniques, they created a risk measure that can predict future misconduct violations by capturing the extent to which a firm’s reviews included those “misconduct words,” such as bureaucracy, compliance, discouraging, favoritism, harass, hostile, meritocracy and strict, the researchers say.
Value and limitations
Hui Chen, a former compliance counsel expert at the Justice Department, says that there is value in this type of analysis, but that it is important to note the study’s limitations. Because the researchers measured misconduct based on what the government penalized, any prediction based on their methodology may miss a good deal of “hidden misconduct”—acts that aren’t known to or pursued by the government for various reasons, she says.
Dr. Campbell says that although the risk index was developed and validated with observed misconduct cases, he believes it could be used to identify possible “hidden” misconduct cases.
Ms. Maxey is a writer in Union City, N.J. She can be reached at [email protected].
Copyright ©2021 Dow Jones & Company, Inc. All Rights Reserved. 87990cbe856818d5eddac44c7b1cdeb8
