Research finds algorithms, unscrupulous preparers behind audit bias

Bias in algorithms and unscrupulous tax preparers are among the drivers of bias in the IRS's auditing Black taxpayers at higher rates than non-Black ones, an IRS researcher said as he provided follow-up research to a university study on the bias.

Tom Hertz of the IRS Research, Applied Analytics and Statistics division spoke at a joint IRS–Urban-Brookings Tax Policy Center research conference, providing highlights of follow-up research to a Stanford University-led study released in January. The study, which Hertz co-authored, concluded that the IRS, which does not collect data on race, was up to 4.7 times more likely to audit Black taxpayers than non-Black ones in tax year 2014.

The subsequent research not only replicated that gap in audit rates but also "confirmed that it's relatively stable over time," Hertz said, emphasizing that the IRS does not collect data on race, so researchers used other methods to impute race.

The study attributed much of audit rate gap to differences in audit rates by race among claimants of the earned income tax credit (EITC). Hertz noted that the Service's focus on preventing EITC overclaims is reflected in both the number of EITC audits conducted and in the way EITC claimants are selected for audit. Both result in elevated audit rates for Black taxpayers.

Unscrupulous preparers

On the issue of unscrupulous tax preparers, Hertz said that in 2019, 17 million tax returns — of which about 6 million claimed the EITC — came from tax preparers that the IRS had identified as unscrupulous as part of a tracking program that began in 2005. Those numbers were from a total of 160 million returns filed in 2019, including 26 million total that claimed the EITC.

"What we have shown is that [the identified uscrupulous preparers] do in fact draw clients disproportionately from minority communities," Hertz said. Once you exclude the 17 million or 6 million, the gap between Black and non-Black audit rates decreases by about 21%, Hertz said. "It's a big chunk of returns coming from preparers who we don't think are exercising due diligence on behalf of their clients," he said.

The issue is one that several Black CPAs mentioned in interviews with the JofA about the Stanford study. In addition, IRS Commissioner Danny Werfel said in May in a letter to U.S. senators that early research also pointed to unscrupulous preparers.

Algorithm bias

On bias in algorithms, Hertz opened by saying: "We have not found a magic bullet. There is no magic bullet that we know of."

He described three mechanisms that point to algorithmic bias in the IRS's EITC audit selection process:

• Residency and relationship status of dependents are a crucial part of EITC eligibility criteria, but the IRS must impute that information from incomplete data, Hertz said. "Moreover, it looks like imputation errors raise the audit risk for Black EITC claimants relative to others. Hopefully, better models and better data supplementing existing data sources might be able to mitigate that problem."
• An aging dependent database model. "Updating the existing EITC scoring model could potentially increase exam revenue and reduce racial bias," he said. Getting rid of algorithmic bias creates a win-win because "you make more revenue and you have less disparity," Hertz said. "That's what we're hoping might be possible with [devoting] better machine learning applications to the problem."
• ­­Weekly audit selection targets that led to over-selection in some weeks — going too far down the scoring ranking — which decreased audit quality and "appears to also have generated some racial bias," a problem that has been largely resolved by better scheduling.

Besides the effects of bias in algorithms and unscrupulous tax preparers, audit objectives also play a role, Hertz said. "Our ongoing research is evaluating the hypothesis that a change in audit objectives to focus on top-dollar tax understatements is feasible in a pre-refund, correspondence audit environment," he said.

This change would include estimating audit outcomes "in terms of differences in audit rates by race, by burden on the applying taxpayer, and enforcement revenue."

It is important that the IRS's main EITC audit selection objective remain lessening the chances of auditing compliant taxpayers, he said.

"Avoiding selecting compliant taxpayers is really important for pre-refund audits of low-income tax payers because holding their refund check for to six to nine months — whatever it takes for them to resolve their correct claim and incorrect audit status — is a big deal," Hertz said.

In his letter to senators, Werfel said the IRS would review its processes to make sure the Service does not audit Black taxpayers at higher rates than others and will make any changes before the next filing season.

Source: www.journalofaccountancy.com