On his first day in office, President Biden signed an executive order acknowledging the toll of structural racism in the United States and initiating a data-driven approach to advancing racial equity. As part of this approach, a high-level Equitable Data Working Group will work to improve federal data available “to measure equity and capture the diversity of the American people.”
Critical to the membership and success of this working group is the US Treasury Department’s assistant secretary for tax policy. Although the Internal Revenue Service (IRS) does not ask about a tax filer’s race and ethnicity, tax laws still create different outcomes depending on a person’s race and ethnicity because of existing inequities in housing, wealth, education, and employment.
Like the rest of our Urban Institute colleagues, the Urban-Brookings Tax Policy Center (TPC) is focused on using rigorous data tools and analysis strategies to dismantle structural racism. TPC is refining its tools and methods to equip lawmakers with a better understanding of how race and ethnicity interact with the federal tax code so they can create more equitable policies in the future.
Here are some of the challenges we’re working to overcome and considerations for our updated model.
TPC’s microsimulation model is the preeminent tax model outside of government. It produces rigorous, quick-turnaround analyses of how current and proposed tax policies affect federal revenues and tax burdens across the income distribution and for various age groups and household types (such as single, married, or households with children).
But, like the microsimulation models used in federal government, TPC’s model cannot estimate effects of tax policies by race and ethnicity. This is because, as noted above, the IRS does not collect information on race and ethnicity.
Sometimes we have worked around this limitation by combining the model’s findings with other sources, such as the Transfer Income Model, to uncover racial disparities in the child tax credit. Such ad hoc approaches, however, cannot be applied on an ongoing basis nor to all policy questions consistently.
What would it take to add race and ethnicity?
Our model starts with a large, publicly available dataset of information from privacy-protected individual tax returns produced by the IRS’s Statistics of Income division. We add to this dataset people who do not file tax returns but who would be affected by proposals to expand the tax rolls or to use the federal tax system to distribute social benefits, such as a monthly child allowance.
We augment these data with survey data on education, consumption, health, and wealth, using matching procedures and projections refined over many years. To each observation in the resulting database, we apply the details of federal tax law—credits, deductions, exemptions, and tax rates—just like many commercial tax preparation software packages. We further incorporate estimates from the economics literature on how individuals respond to tax policy changes.
The US Census Bureau’s Current Population Survey (CPS), which we already match with our tax model, includes information on each respondent’s race and ethnicity, and we could simply carry over this information from the match. This would allow us to assign information on race and ethnicity to each matched tax filer in our model. But an outstanding question is whether doing so would produce the right correlations between race and other tax data.
To take one example: though the CPS is an excellent source of information generally on demographics and wage income, it excludes large and important determinants of taxable income, such as capital gains, passive business income, and mortgage interest. So, TPC’s current statistical match with the CPS will only capture variation by race and ethnicity for those tax items to the extent they correlate with other income and demographic information. For instance, it wouldn’t capture how much mortgage interest differs between Black taxpayers and other groups for reasons beyond differences in income, family structure, and age.
Similarly, charitable contributions, health care spending, and higher education payments likely vary systemically by race and ethnicity for reasons beyond income, family structure, and age. TPC may be able to use other data sources or integrate our model’s findings with Urban’s other microsimulation models to capture such differences. But understanding how people facing different constraints might spend, save, and respond to tax policy changes differently will require many technical judgments, in addition to consultations with a limited knowledge base.
Over the next several months, we plan to dig into these questions. We will be transparent about our process and the inherent limitations of representing people’s diverse experiences in a statistical model. We will also take proactive measures to ensure we do not include oversimplified behavioral assumptions or present data in a way that could lead to misleading conclusions about racial disparities.
There is much we don’t know about how progressive federal income tax rates interact with various credits, deductions, and exemptions that distribute more than a trillion dollars in tax benefits annually. Though credits geared toward families with low to moderate incomes, such as the earned income tax credit, likely reduce racial disparities, other provisions, such as the home mortgage interest deduction, may increase them because of housing discrimination and unequal access to credit.
We are eager to undertake this project and are optimistic that better data will help policymakers design more policies that advance racial equity.