Each year on April 15, the Internal Revenue Service expects Americans to file their annual tax returns, and in doing so, Americans expect U.S. tax policy to be administered fairly across demographic groups.
While the IRS and the Department of the Treasury have committed to tax administration equity, the Government Accountability Office
GAO examined IRS’s safeguards for ensuring audits do not target filers based on demographic characteristics and made recommendations to improve its audit selection process and mitigate potential unintended biases. Such changes to the audit selection process could minimize the chances of the IRS inadvertently generating disparities in audits.
IRS does not collect information on the race and ethnicity of taxpayers; however, GAO found that some audit selection criteria and methods could have different implications for taxpayers depending on their race or ethnicity. Recent academic research estimated that Black taxpayers are audited at 3-7 times that of taxpayers of other races. Furthermore, returns claiming the Earned Income Tax Credit appeared to account for more than 70% of that disparity. IRS researchers confirmed there are disparities in the number of audits of Black taxpayers in relation to taxpayers of other races.
IRS’s audit selection process focuses on areas with the highest likelihood of noncompliance. As part of this process, it relies, in part, on a measure called the no-change rate — the percentage of returns that will yield no additional revenue after audit. IRS seeks to achieve a low no-change rate because it indicates noncompliant taxpayers are being audited — those that had a “change” to their tax return.
The IRS’s calculation of the no-change rate includes default audits — audits closed as a “change” because taxpayers did not respond or provided insufficient responses to IRS’s notices — and research shows that Black taxpayers are more likely to not respond to IRS correspondence than taxpayers of other races. Default audits also may be more common among low-income and EITC taxpayers because of challenges that make communicating successfully with the IRS more difficult, such as being transitory or not having bank accounts.
Because IRS is relying on a no-change rate that is likely lowered due to the inclusion of taxpayer non-response audit selections, IRS could be disproportionately selecting returns that are more likely to be default audits. This selection process could also lead to inadvertently selecting taxpayers with certain demographic characteristics. GAO has advised IRS to calculate multiple no-change rates, including one without default audits, to help remove any unintended biases from its audit selection method.
After the IRS has determined the type and amount of refundable credit returns that will be audited, it uses a system of algorithms to select specific returns for audit. The primary system the IRS uses to select specific returns for audit is the Dependent Database (DDB) program, an automated system that flags returns for potential risk of noncompliance. GAO found weaknesses in IRS’s review and use of these algorithms. While the IRS regularly reviews the program, the review process does not comprehensively consider the data inputs and assumptions that could inform IRS about the demographic equity of the audit selection process. For example, GAO found that some risk scores contained in algorithms vary by sex, which could skew selection, and have not been updated since 2001. This lack of review could create the potential for unintended bias in audit selection.
IRS has begun work to investigate these issues, including implementing a new risk-scoring model used for audit selection in the tax filing season of 2024. However, it does not have guidance for how to consider its research on audit disparities while reviewing its automated audit selection processes. For example, IRS research identified outdated models and concerns with the reliability of external data sources as potential contributors to disparities in audits, but it does not have guidance for implementing these findings. IRS has also indicated plans to increase the use of artificial intelligence methods to select returns for audit. In doing so, it is imperative the IRS have a comprehensive and systematic process for reviewing these systems that considers the potential for imbedding unintended biases in audit selection.
GAO made six recommendations to the IRS to address these issues. This included calculating no-change rates with and without default audits, developing guidance for considering audit equity research in developing audit workplans and assessing the DDB system’s potential to contain algorithmic biases, and conducting more comprehensive reviews of its algorithms and the data sources those utilize. IRS concurred with GAO’s recommendations and is assembling a team of experts across the agency to implement these recommendations and address unintentional disparity in audit selection.
Credit: Source link
