Following the January release of a groundbreaking study by Hadi Elzayn, Robin Fisher, Jacob Goldin, Thomas Hertz, Daniel E. Ho, Arun Ramesh, and Evelyn Smith and the resulting media, Congressional, and IRS attention, it is now well-known that Black taxpayers are audited at rates three to five times the rates of non-Black taxpayers. The audit study is a landmark both for its results (which contradict past IRS statements) and also for its novel methodology, which uses individually-estimated taxpayer race probabilities to obtain informative bounds on the racial audit rate disparity. In addition to illuminating problematic patterns in current IRS audit selection procedures, the study’s methodology offers promise for the future in investigating other race-based patterns in tax enforcement.
In what non-audit enforcement areas might such patterns arise? Here is where the prescient work of Jeremy Bearer-Friend (as cited in the audit study, P. 41) comes in, building on the work of other scholars working at the intersection of race and tax. In “Colorblind Tax Enforcement,” Bearer-Friend refutes on first principles the now-debunked claim that because the IRS does not collect race data, it cannot discriminate by race when enforcing tax laws. He points out that, for IRS agents, making inferences about the race of a taxpayer on the basis of the information provided on the return (names of taxpayer, spouse, and children, address including zip code, family structure, and occupation) is plausible and probable: “[e]ach of these datapoints can lead to inferences of racial identity in the mind of the relevant IRS personnel, with the combination of data points creating a stronger likelihood of inference” (P. 19). Moreover, at many points in the enforcement process there are telephonic or in-person conferences that allow for further racial inferences.
The paper then provides two models of racial bias that can produce racially disparate tax enforcement where IRS personnel do make racial inferences: racial animus (intentional harm on basis of race, P. 17) and implicit racial bias (harm on the basis of race that is unconscious or unintentional, P. 21). However, even if no racial inferences are made or neither of these models apply, Bearer-Friend offers a third model that can produce harmful disparate results by race: transmitted bias. This is where racial animus in another area of life determines taxpayer characteristics that intersect with tax enforcement. For example, having unstable housing as a result of racial discrimination or racially-skewed mass incarceration is associated with unstable mailing addresses, which in turn compromises a taxpayer’s ability to respond to IRS communications that are distributed by mail (of which there are many; P. 25). Transmitted bias requires neither informational inferences nor discretion on the part of an IRS agent: it operates through broader societal forces that affect taxpayers differentially by race.
Bearer-Friend goes on to identify seven tax enforcement settings that are vulnerable to racial bias as operationalized through each of the three models. He emphasizes that the seven settings are “representative” and thus non-exhaustive, but they suggest particularly promising areas of investigation. I summarize them below and offer questions that flow from Bearer-Friend’s observations.
- Summonses: The IRS has broad discretion in determining the scope of information to be summoned (e.g., range of financial documents and tax years). Such information can support racial inferences as well as determine the compliance burden of the summons. The IRS also has discretion concerning who to summon, such as the taxpayer’s business counterparty, which might have ramifications for the counterparty’s trust in or future willingness to do business with the taxpayer, and whether to enforce the summons. Thus, there is scope for racial animus and implicit racial bias. In addition, summons issuance is an area ripe for transmitted bias because a summons relies on a taxpayer’s last known address. Does the summons enforcement rate vary by the race of the taxpayer?
- Civil penalties: Many civil penalties are automated, but IRS personnel have discretion on some and can choose to abate billions of dollars in penalties. Where a taxpayer’s race can be inferred on the basis of information provided on the return, this context is vulnerable to racial animus and implicit bias. Transmitted bias also may be present here: in the case of the penalty for civil fraud, for example, the taxpayer’s use of cash is a factor that weighs in favor of finding fraud. However, the use of cash is racially non-neutral due to a history of racial animus in banking. Do penalty abatement rates vary by the race of the taxpayer?
- Appeals: The IRS Independent Office of Appeals is instructed by the Internal Revenue Manual to use their “experience and judgement” when reviewing the merits of a case for settlement. This discretion, plus the fact that settlement conferences are held telephonically, virtually, in person, or through correspondence, opens the door to disparate racial treatment via racial animus and implicit racial bias. While many difficult-to-observe aspects of cases are likely to affect settlement rates, investigating which cases settle and whether race appears to be a factor might be revealing.
- Offers in compromise: The decision to accept an offer in compromise is discretionary and, regardless of the grounds on which it may be requested, the Internal Revenue Manual instructs collections officers to contact a taxpayer telephonically to collect additional information necessary to consider the offer. Here again, there is scope for racial animus and implicit racial bias. Do rates of acceptance of offers in compromise vary by race?
- Collection due process hearings: Taxpayers who receive a notice of collections (a lien or a levy) from the IRS can request, in writing within 30 days, a Collection Due Process (CDP) hearing. The hearing occurs in person or telephonically with a settlement officer in the IRS Appeals Office, who exercises discretion in proposing or agreeing to a settlement. Thus, racial animus or implicit racial bias may be present. Transmitted bias may also arise: the notice of collections is valid if sent by certified mail to a taxpayer’s last known address, even if the taxpayer has no actual knowledge of the notice. Among taxpayers who are mailed a notice of collections, do rates of requests for CDP hearings vary by race?
- Innocent spouse relief: To request relief from a spouse’s tax liabilities for taxpayers deemed to be innocent spouses, a taxpayer must file Form 8857, which is evaluated by an IRS technician. The form has narrative sections that can “include qualitative information and natural language patterns;” the technician may also review “personal financial information that might also imply racial identity” (P. 43), thus allowing for racial animus or implicit racial bias. In addition, transmitted bias may occur if there are behavioral patterns in who feels empowered to submit a form in which substantial personal disclosure is required. In cases where joint and several liability of a spouse is at issue, do filing rates for Form 8857 vary by race?
- Criminal tax referrals: Although no single CI agent can make a recommendation to DOJ’s Tax Division without review by a supervisor (P. 45), agents have substantial discretion regarding which referrals to pursue and how much investigating to do, a process which often involves the IRS’s broad summons power, which can produce racially-revealing information (discussed above). In addition, to the extent that certain kinds of cases are less likely to be recommended for prosecution, and the defendants in those cases are more likely to be white (i.e., recipients of pass-through income, according to a Tax Policy Center study that Bearer-Friend cites), racially disparate outcomes may occur through transmitted bias. DOJ collects and reports the race and ethnicity of defendants in its prosecutions. How would a similar report by CI look? Are there racial patterns in the referrals that are chosen for investigation?
Bearer-Friend is clear that his review of the scope for racial bias in these tax enforcement settings “is not an assertion that such racial bias is already occurring at the IRS” (P. 47). The problem, rather, is that at the time that Bearer-Friend was writing, current data practices at the IRS inhibited such an inquiry. To address this, Bearer-Friend recommends exactly the kind of imputation techniques that both the audit study and a Treasury Department analysis of tax expenditures have since performed. Analyzing IRS enforcement procedures for possible racial bias is thus feasible (although it is far from costless, which is one more reason that IRS funding increases must be preserved). Bearer-Friend’s article gives us tremendously fruitful suggestions for where such analyses should start.
