Enlarge image
Photo:
Klaus Vedfelt/Getty Images
Corporate scandals often follow a certain pattern: Theranos and its fraudulent blood testing technology, Wells Fargo and its fake financial accounts, Volkswagen and its manipulated emissions data - at some point a whistleblower will come forward.
"The problems that are then uncovered have usually existed for many years," says Dennis Campbell, professor of business administration at Harvard Business School.
To find out if difficulties could be detected earlier, he started an experiment together with Ruidi Shang from the Dutch University of Tilburg.
To spot signs of trouble, the two compiled a dataset of employee reviews published on the Glassdoor.com platform.
It included assessments of around 3,500 listed companies that were written between 2008 and 2016.
The researchers then extracted a vocabulary of 11,772 words from the ratings.
To compare them, they used a database compiled by the nonprofit organization Good Jobs First, containing nearly 27,000 violations by the same companies between 2008 and 2017.
In the next step, they trained an algorithm and let it decide which words were most strongly associated with potential violations.
They ranged from descriptive terms such as "pay" and "promotion" to overtly negative terms such as "discrimination," "anger," "favouritism," and "unethical."
The results enabled the researchers to use the weighted proportion of words to clearly distinguish between companies with a high and a low number of violations - regardless of previous violations or financial difficulties.
Glassdoor isn't the only site that can detect wrongdoing, Campbell says.
The analysis can be applied to any platform on which employees discuss the dynamics of the company.
With their help, supervisory authorities could make better decisions about which companies to investigate.
Investors also benefited when they wanted to assess risks.
However, her method can also be used by companies themselves, says Campbell – for example, to uncover potential problems through external or internal employee communication.
In this way, scandals could not only be recognized but also avoided.
Source:
Dennis Campbell, Ruidi Shang, "Tone at the Bottom: Measuring Corporate Misconduct Risk from the Text of Employee Reviews", Management Science, in press