Unanswered Questions Remain on Causes of Racial Disparities in Audit Selection and the IRS’s Plans for Further Evaluation

By Kathleen Bryant

In a recent letter, the IRS shared initial findings that confirm the existence of racial disparities in audit selection between Black and non-Black tax filers. However, many critical questions about the root causes of these disparities remain unanswered. These questions must be answered to enable swift action to address the root causes of racial audit disparities. Additional detail is also needed on the IRS’s future plans to evaluate audit disparities on a recurring basis and to conduct research on potential audit disparities along other lines of difference, including by age, gender, geography, and other racial and ethnic categories.

Last week, the IRS published a letter in response to an inquiry from Senate Finance Committee Chairman Ron Wyden with its initial findings that Black tax filers are more likely to be audited than non-Black tax filers. These findings corroborate a recent working paper that was written by academic and Treasury Department researchers and published by Stanford University’s RegLab, which found that Black tax filers are 2.9 to 4.7 times as likely to be audited by the IRS as non-Black tax filers.

The IRS indicated that evaluating its “EITC audit selection algorithms” is the agency’s “topmost priority” for “improving fairness in tax administration.” Focusing on EITC audit algorithms is consistent with the Stanford working paper’s conclusion that nearly four-fifths of the overall audit disparity between Black and non-Black tax filers can be explained by the distribution of audits within the population of EITC claimants. Furthermore, the IRS stated that it will “identify and implement” changes to the IRS’s “exam priorities and automated processes” that will address racial audit disparities “prior to the next tax filing season.”

The IRS has taken important steps forward in confirming the existence of audit disparities between Black and non-Black tax filers (particularly among EITC claimants) and setting a concrete deadline for implementing changes to audit selection algorithms and processes.

However, many critical questions about the root causes of these disparities remain unanswered. The IRS’s letter does not specify which design elements of the IRS’s audit selection algorithms, or other parts of the agency’s audit selection processes, are driving racial audit disparities. The IRS needs to answer these questions so that it can determine which changes to its audit algorithms and procedures need to be implemented.

In a previous report, we outlined several questions that the IRS should address about the extent to which certain design elements of audit selection algorithms contribute to racial audit disparities. These specific design elements include:

• The measure of tax under-reporting being “targeted” in IRS algorithms. As the Stanford working paper explains, the choice of “target” for audit selection models can significantly affect racial disparities in audit selection. Because the IRS did not grant researchers full access to the agency’s actual audit selection algorithms, researchers had to create their own sample models to illustrate how different design choices can affect racial audit disparities. They found that in general, algorithms targeting the presence of tax under-reporting disproportionately audit Black filers while algorithms targeting the magnitude of tax under-reporting audit Black filers at lower rates. Similarly, algorithms that specifically target over-claims of refundable tax credits, rather than tax under-reporting in general, produce significant racial audit disparities.
• The characteristics of tax filers used to predict the likelihood of tax under-reporting. Once an audit algorithm is given a “target,” it will select tax returns for audits based on the characteristics of tax filers that are most closely associated with that target. While race and ethnicity are not reported on tax returns and therefore cannot be directly used to select tax filers for audit, certain characteristics used to predict tax under-reporting may be highly correlated with race and drive racial audit disparities.
• Constraints imposed on audit selection algorithms due to funding or operational limitations. Funding and operational limitations may require the IRS to constrain their audit selection algorithms and prioritize certain types of audits that are faster and cheaper to conduct. The Stanford working paper’s simulations of audit selection models suggest that such constraints can exacerbate racial disparities.
• The use of historical audit data to “train” algorithms, and whether this may create “doom loops” in audit selection. As we have previously explained, the IRS’s process for auditing low-income tax filers can be very difficult to navigate. Consequently, many filers do not participate in the process and have claims for the EITC and other tax benefits denied by default. If the administrative burdens of the IRS’s audit process are inequitably distributed by race, algorithms that are “trained” using historical audit outcomes may “learn” over time to select tax filers with the greatest barriers to navigating audits and create a “doom loop” that entrenches racial audit disparities.

The IRS should explain which elements of its audit selection algorithms and processes are driving racial disparities, and describe which changes will be put into effect before the next filing season.

Additional detail is also needed on the IRS’s future plans to evaluate its audit selection practices for bias on a recurring basis and to conduct research that would identify whether audit disparities exist for any other marginalized communities.

Plans for further evaluation and research are present in both the IRS’s letter to Senator Wyden and the agency’s Strategic Operating Plan (SOP) for implementation of the Inflation Reduction Act. For example, in the SOP, the IRS committed to “regularly assess[ing] whether IRS enforcement actions, their application and enforcement-related services are disproportionately burdening or advantaging specific demographic populations, geographies or customer categories” and to making “real-time and regular adjustments to [its] approach.” Furthermore, the IRS’s letter and the SOP both indicate that the agency will conduct further research on potential audit disparities along other lines of difference, including age, gender, geography, and other racial and ethnic categories.

The IRS committed to providing “regular updates” to Senator Wyden and the Senate Finance Committee more broadly on its work to understand and address racial disparities in audit selection. Those updates should be made public either by the Committee or by the IRS, and should address the outstanding issues that we have described above.

Please subscribe to our mailing list to receive our posts and general updates via email.