If you're a single Black man with dependents who claims the Earned Income Tax Credit (EITC), you have a 7.73% chance of being audited by the IRS in any given year. For Americans as a whole, the equivalent figure is just 0.54%.
- Why it matters: Single men with dependents claiming the EITC are also audited at a very high rate if they're not Black — 3.46% of them get the dreaded letter in the mail — but that number is still small when compared to their Black counterparts.
The big picture: Black Americans at all levels of the income spectrum get audited at significantly higher rates, according to an extremely important new study conducted by Stanford researchers with the cooperation of the IRS.
- The study used comprehensive data from 2014 — "the most recent year for which the vast majority of audits were complete and available to us," the researchers write.
What they found: The biggest disparity, by far, comes among Americans claiming the EITC. When looking at EITC claimants, Black filers are much more likely than their non-Black counterparts to get audited.
- "The racial audit disparity within EITC returns contributes 78% of the overall disparity," conclude the researchers.
Between the lines: As Axios' Emily Peck reported last year, EITC audits are much easier to conduct than audits of high earners, even though the rich tend to evade much more in the way of taxes.
- Even within EITC audits, it's easier for the IRS to deal with filers who have no business income. Black filers constitute 21% of non-business EITC returns, but only 11% of business EITC returns.
- The IRS seems to be effectively ignoring the business returns: Non-business audits constitute 93% of EITC audits, despite the fact that business EITC returns have much more underreporting, in dollar terms.
How it works: The IRS does not observe race directly, and EITC audit selection is done by algorithm. But the algorithm design seems to be having a disparate impact on Black Americans.
- Specifically, the algorithm seems much more likely to select tax returns where the chance of some kind of underpayment, even if small, is extremely high, rather than returns where the underpayment is likely to be largest.
- The latter returns tend to be filed by high-income taxpayers, and require much more work — and more experienced auditors — than the relatively automated audits of non-business EITC claimants.
The bottom line: It looks very much as though the problem of disparate algorithmic impact could be solved pretty easily — by changing the IRS auditing algorithm from being based on a binary "is this tax return likely to be false" question, to being based on the expected dollar amount of underpayment.
- That, however, would mean an increase in the number of expensive auditors needed at the IRS. And Congress hates to increase the IRS budget, even when doing so pays for itself many times over.
My thought bubble: The even better solution would be to stop making everybody file a tax return in the first place.