Bias possible in IRS program for auditing refundable credit returns

The automated system that the IRS uses to choose which returns claiming refundable tax credits will be audited could have bias that should be addressed, the Government Accountability Office (GAO) said in a report (GAO Rep’t No. GAO-24-106126 (April 2024)).

The IRS agreed, and the report notes that the agency has already begun “to examine potential systemic biases in audit enforcement across several demographic characteristics, including age, gender, geography, race, and ethnicity.”

Review of IRS system

The IRS uses an automated system called the Dependent Database (DDB) to choose which returns claiming refundable credits — mainly the earned income tax credit (EITC) — it will examine. And although the IRS meets annually to review the DDB’s operation, that process “does not consider all data inputs, assumptions, and other model components that could inform IRS about the demographic equity of the audit selection process,” the report said.

“For example, some risk scores contained in the DDB program vary by sex and have not been updated since the implementation of the DDB system in 2001,” the report said.

In a letter to Sen. Ron Wyden, D-Ore., chair of the Senate Finance Committee, the GAO explained why its review is important.

“Audit selection criteria and methods, including those established in automated processes, play an important role in promoting fairness,” the GAO wrote. “It is critical to understand the different implications these selection methods could have for taxpayers depending on their race, ethnicity, income level, or other demographic and economic characteristics.”

GAO recommendations

The GAO recommended six changes, including that the IRS develop guidance for considering audit equity research in developing the audit workplan. The GAO also recommended that the Service calculate multiple “no-change” rates (discussed below); that it establish a systematic process to ensure its audit selection algorithms “consider all data inputs, assumptions, and other model components”; that it develop clear guidance for assessing algorithmic bias in the DDB system; that it assess how effective its automated audit selection system is in predicting taxpayer noncompliance risk; and that it assess the reliability of the databases it uses.

The IRS agreed with all six recommendations.

“The IRS is committed to enforcing the tax laws in a manner that is fair and impartial,” IRS Deputy Commissioner Douglas O’Donnell wrote in response to the report. “It is important to reiterate that the IRS does not and will not consider race in our case selection and audit processes.”

That the IRS agreed with all the GAO recommendations is noteworthy in and of itself, said Jean Wells, CPA, a tax attorney and associate professor in the Howard University School of Business.

“For previous GAO reports, they might not have agreed with the recommendations, but for them to have agreed with all six and that they are already working on some, I think that’s a pretty good sign,” Wells said in an interview with the JofA. “And I think that the establishment of the exam disparity and trustworthy analytics teams demonstrates the IRS’s commitment to change. Hopefully, they will come up with some recommendations that will be implemented.”

IRS changes

Wells was referring to two internal IRS teams that O’Donnell mentioned in his response letter to the GAO. The teams were created “to facilitate research and evaluation of fairness-related questions,” he wrote, listing five team accomplishments to date.

One of the IRS’s goals with the audits of returns claiming refundable tax credits is to have a low “no-change” rate because “that indicates that the audit processes effectively found noncompliant taxpayers and, therefore, are effective at meeting one measure of fairness,” the report said.

But that no-change rate includes audits that are closed as a “change” because the taxpayer did not respond or did not provide sufficient responses to notices. And recent research found that Black taxpayers are less likely to respond to IRS notices than those of other races, the IRS said.

In addition, these audits, known as default audits, may be more common among low-income and EITC taxpayers because it is more difficult for the IRS to communicate with them because, for example, they may be transitory or lack bank accounts, the GAO said.

The GAO recommends that the IRS calculate several no-change rates: the overall no-change rate; the no-change rate excluding default audits; and the no-change rate excluding taxpayer nonresponse default audits. This would “provide insights into potential equity disparities,” it said.

A study led by Stanford University and released in January 2023 found that Black taxpayers were 2.9 to 4.7 times more likely to be audited than non-Black taxpayers. When researchers turned to the EITC, they found that Black taxpayers were the focus of 43% of EITC audits but accounted for just 21% of EITC claims.

IRS Commissioner Danny Werfel said in a recent call with reporters about the IRS’s strategic operating plan that the agency has changed the selection criteria for refundable credit audits. Now, the IRS is monitoring and studying whether those changes have eliminated the disparity, he said.

In addition, the IRS must “make sure that we are continuously monitoring for such disparities not just among Black taxpayers but other minorities and any other inequities that would degrade trust in our tax system,” he said.

Data challenges

The IRS does not collect information about race and ethnicity on tax returns, so any statistics about those are imputed by other means. For example, the Stanford researchers used voter registration records from North Carolina, which, until recently, required citizens to check a box for race and ethnicity when they registered to vote.

However, Wells noted that models can be created to simulate tax outcomes with race and ethnicity. That is what the GAO did for a report published in May 2022 when it used 2017 census data.

That report suggested that “[i]f tax data could be linked to households’ demographic data in a way that still protects the privacy and security of those data, policymakers and researchers would have better tools for consistently and systematically analyzing the relationship between tax policies and household demographics.”

The GAO report recommended that Congress revise relevant laws to facilitate interagency data sharing to help with the data collection and analysis. As of March 2024, Congress had not acted on this recommendation, the GAO said.

— To comment on this article or to suggest an idea for another article, contact Martha Waggoner at Martha.Waggoner@aicpa-cima.com.