Everybody knows accounting fraud exists, but outside a court of law or a federal investigation, it’s difficult to get anyone to admit to it. You can’t just take a poll of random CFOs and ask, “Have you ever committed fraud?”

Or can you?

As far as Gies College of Business professor Alex Vandenberg knew, no one had actually attempted to ask accounting executives whether they have manipulated financial reports. So he and two colleagues, Nicole L. Cade and Joshua L. Gunn, both of the University of Pittsburgh, decided to give it a try, for science.

“I believe researchers have avoided asking executives about their involvement with earnings manipulations because executives could simply lie, so why bother?” said Vandenberg, an assistant professor of accountancy. “Because executives are often the only people who know that earnings have been manipulated, we thought it would be worthwhile to at least try asking executives directly about whether they have manipulated earnings.”

But the researchers designed a second, simultaneous test to get answers just in case direct questioning didn’t work – a novel method known as the list experiment. Gunn had learned about it while browsing Twitter (now known as X); it had been used in a study to measure the prevalence of atheists, but the three accountants immediately saw its potential for a fraud study.

In a list experiment, the researchers give participants a series of true-false statements. For the control group, all the statements are innocuous things like, “I have children,” or “I take the bus to work.” The experimental treatment group gets those same statements, too, and then one more like, “I have committed accounting fraud.” By comparing the average number of statements rated as “true” between the two groups, the researchers are able to infer the proportion of executives who responded “true” to the fraud question.

One difficulty with this method is that it requires a lot of responses for researchers to be confident that the difference in the average number of statements rated as “true” between the control and treatment group is due to executives admitting to fraud. Collecting sufficient data for the experiment to be worthwhile takes a lot of time and money.

Fortunately, Vandenberg, Cade, and Gunn could justify these upfront costs because prior to collecting responses, their study received in-principle acceptance via the Journal of Accounting Research’s registered-report editorial process. So long as the researchers adhered to their planned procedures, their study was set to be published in the Journal of Accounting Research. The completed study titled “Measuring the Prevalence of Earnings Manipulations: A Novel Approach” is now published in the journal. 

Much to their surprise, of the executives who were asked directly about different forms of earnings manipulation, approximately 27% report that their company manipulated earnings (excluding fraud) in some way in the last five years. “It was a bit shocking to us to find that so many executives were willing to outright admit to some of these manipulations,” Vandenberg said

While no executive directly admitted to fraud (formally defined, Vandenberg said, as “material misrepresentation with the intent to mislead”), results of the list experiment suggest that 12.4% of Russell 3000 companies may have committed financial statement fraud at least once in the last five years. 

To form these estimates, they had to solve the problem of getting busy executives to respond. At first they tried reaching out to executives through email or LinkedIn, but that approach didn’t yield many responses; email and DMs are too easy to ignore. So instead they turned to a more old-fashioned method: the mail.

The researchers sent out more than 7,500 research requests in the form of padded mailers — hand-stuffed, with the aid of research assistants — to top-level executives from the Russell Index, a list of the 3000 largest companies in the US stock market. The executives were randomly selected to participate either in the survey or the control or experimental group of the list experiment. They could respond via mail or electronically with a QR code. They were promised complete confidentiality in the form of a signed one-way NDA; in addition, the researchers promised to make a donation to the American Cancer Society for each response. 

One of the more delicate parts of the operation was wording the specific descriptions of each form of earnings manipulation so that there could be confusion about what the researchers were asking about. “If you precisely define each form of earnings manipulation,” Vandenberg explained, “it helps ensure that executives understand the question and importantly, it limits executives’ ability to rationalize their way out of admitting to something by exploiting ambiguity in the wording of the question.”

In addition to accounting fraud, the executives were also asked about real earnings management, disclosure obfuscation, accrual-based earnings management, and material omission, all of which occupy an ethically gray area that is not necessarily strictly illegal. It took several months to collect all the responses. The researchers’ recruitment efforts yielded a response rate of 12.9%, which is higher than most comparable accounting studies. 

Vandenberg and his colleagues believe their findings further our understanding of the prevalence of earnings manipulations and demonstrate some potentially useful research approaches for future researchers.