ITEP has maintained a microsimulation model of the tax systems of the U.S. government and of all 50 states and the District of Columbia since 1996. The ITEP model is a tool for calculating revenue yield and incidence, by income group, of federal, state, and local taxes. It calculates revenue yield for current tax law and proposed amendments to current law. Separate incidence analyses can be done for categories of taxpayers specified by marital status, the presence of children, and age. More recently, ITEP has added the ability to disaggregate its model results by race and ethnicity, as explained below. To forecast revenues and incidence, the model relies on government or other widely respected economic projections.
The ITEP model’s federal tax calculations are very similar to those produced by the congressional Joint Committee on Taxation, the U.S. Treasury Department and the Congressional Budget Office (although each of these four models differs in varying degrees as to how the results are presented). The ITEP model, however, adds state-by-state estimating capabilities not found in those government models.
RACE AND ETHNICITY ANALYSIS IN GENERAL
To make disaggregation of our tax results by race and ethnicity possible, we assign each observation in the ITEP model a probability of being a certain race or ethnicity based on seven characteristics for which we were able to crosswalk between the ITEP model and 5-year estimates from the American Community Survey (ACS). Those characteristics are state of residence, income level, marital status, homeownership, age (a binary for over/under 65 years old), number of children in the household under age 6, and number of children in the household between the ages of 6 and 17.
Racial and ethnic outcomes were produced by matching ACS individual and household data, followed by producing frequencies of demographic sub-tax units within all tax units. We translate demographic information on ACS households to characteristics of tax units defined in the ITEP model using an algorithm similar to those described in Cilke (1994) and Rohaly et. al. (2005). Under this methodology, where relevant we break out tax units from households and assign dependents to adults based on several criteria related to age, income, and education status. ITEP’s resulting estimates of tax units’ racial and ethnic composition are validated by checking against published Census data on households and persons, as well as demographic data published by states.
Based on those seven characteristics we calculated “race weights” for each observation that allow it to represent tax units of up to eight different racial or ethnic groups. A middle-income, non-elderly married observation with two children, for instance, might represent 10 tax units. This method then imposes on that weight assumptions about the likelihood that tax units of the given profile are led by individuals of one of the eight Census-derived racial groups (explained below). Functionally, then a single observation might simultaneously represent five white tax units, three Black tax units and two Hispanic tax units. Further up the income scale most observations will tend to represent more white and Asian tax units and fewer tax units of other races. The eight possible racial and ethnic groups included in our model are:
- White (non-Hispanic)
- Black or African American (non-Hispanic)
- Asian (non-Hispanic)
- American Indian, Alaska Native, or American Indian and Alaska Native (non-Hispanic) living away from federally recognized reservations
- Native Hawaiian, Other Pacific Islander (non-Hispanic)
- Multiple races (non-Hispanic)
- Other single race (typically non-Hispanic, but can include Hispanic tax units as well if data in a small number of states where data limitations do not allow for separate estimation of the Hispanic population)
In approaching each state, we first evaluated the sample of tax units estimated via the 5-year ACS (functionally households broken down into tax unit estimates using the methods described above), as well as our own internal samples of tax units, along the seven dimensions described above for the first seven racial and ethnic groups. We selected minimum sample sizes that balanced interests in illuminating subgroup incidence with interests in reducing noise over time and ensuring the validity of subgroup estimates. Where the sample was deemed sufficient both within the ACS and within the ITEP dataset, we moved forward with assigning race weights for that subgroup in our model. White tax units are the only discrete group for which we were able to calculate reliable estimates for the non-elderly population in all 50 states and the District of Columbia. Where sample size was lacking either in the ACS data or within ITEP data, we subsumed that group into our eighth racial group (“Other single race”). This method has both strengths and weaknesses. We thought it important that revenues are able to sum across racial and ethnic groups within each state, and this method allows for that. However, we are also allowing the definition of the “Other single race” group to vary across states. This means that the composition of group eight (other single race) is both highly diverse and highly variable across states, and we, therefore, caution against drawing conclusions from this group’s tax payments or demographic distribution.
For state modeling, the fourth group identified above (American Indian and/or Alaska Native) only includes those tax units living away from federally recognized reservations, a group sometimes referred to as urban Indians. American Indians living on reservations are excluded from our state estimates because, as sovereign nations, many tribal governments have their own systems of taxation and those tax systems are outside the scope of our analyses. In assigning race weights for the urban Indian population, we began with estimates based on the ACS microdata for all American Indian and Alaska Native households and then backed out a subset of those households that resemble those living on federally recognized reservations. This was accomplished by comparing ACS tables showing the number of American Indian and/or Alaska Native households in specific income bands that live in a state to tables showing the number living on federally recognized reservations. In many instances we found that lower-income American Indian and/or Alaska Native households were disproportionately likely to reside on reservations, though this was not found to be the case in every state.
In our examination of the ACS data, any household where the respondent identified as being of Hispanic ethnicity is classified in group seven regardless of their racial identity. For non-Hispanic households, if the respondent identified as being of more than one race they are categorized in group six (“multiple races”). The remaining groups contain non-Hispanic households where the respondent identified as being of only one race. This construction is a helpful simplification of a complex topic that allows us to identify discrete racial and ethnic groups that can be analyzed separately, but it does have drawbacks.
For instance, because some groups – such as American Indians – are more likely than others to be of multiple races we risk undercounting those groups by reporting them in the “multiple races” category. Moreover, because Hispanic versus non-Hispanic ethnicity comes first in our assignment of racial and ethnic identity, we classify some households as being Hispanic even if, in fact, the respondent may identify more closely with their race than with their ethnicity. Finally, while the eight-group lens described above allows us to produce estimates for the largest groups in most states, there are some states – especially Hawaii – with unique racial and ethnic compositions where the lens is a poor fit. All these areas present opportunities for future study.
REPORTING ACROSS INCOME GROUPS
When analyzing tax laws and proposals strictly by income level, we typically present estimates for seven income groups: the bottom four quintiles as well as the “next 15 percent,” “next 4 percent,” and “top 1 percent” of earners. Disaggregating the data further by race and ethnicity creates additional demands on the data that, in many instances, require us to report results for larger income bands.
Our data on residential property taxes, for instance, are currently aggregated into just three income groupings: the bottom 80 percent of tax units, top 20 percent, and all tax units regardless of income level. We often report consumption tax results using the same groupings. This division between the bottom 80 and top 20 can be thought of as roughly splitting most states’ income in half: in many states we see that 55-60 percent of income flows to its top 20 percent of earners compared to 40-45 percent flowing to the bottom 80 percent.
Income tax data tend to be more detailed than those needed to model consumption or property taxes, and this allows us to report income tax effects for a larger number of income groups. Typically, we can disaggregate federal income tax results by race for all seven income groups while state income tax estimates are available for five quintiles (that is, without a separate breakout for the next 15, next 4, and top 1 percent of earners).
One high priority area of future study is the state-by-state racial and ethnic makeup of the very top of the income distribution. Achieving a sufficiently large sample size at the top of the income distribution to model state-level taxes by race and ethnicity on top earners has proven challenging. Top earners are overwhelmingly white, leading to small sample sizes for other races and ethnicities at high income levels.
In our underlying model data, our top 1 percent of tax units tends to be somewhat more white, and somewhat less Black and Hispanic, than our broader top 20 percent grouping. These findings are qualitatively consistent with what we see in the national ACS data, though we have reason to suspect that our top 1 percent may still have too few white households and too many households of color. In future research we hope to delve more deeply into the racial and ethnic profile of each state’s top earners.
The federal government, 41 states, the District of Columbia, and many other local governments levy broad taxes on personal income. Most states and the federal government also levy entity-level taxes on corporate profits. Our approach to modeling these taxes by race and ethnicity is described below.
Personal Income Taxes
In state-level analyses, personal income tax liability by race is affected by all seven of the characteristics we targeted: state of residence, income level, marital status, homeownership, age (a binary for over/under 65 years old), number of children in the household under age 6, and number of children in the household between the ages of 6 and 17. Of these factors, income and family size are the two most important. In national analyses we assign race weights without regard to a tax record’s assigned state of residence.
Race and ethnic groups with higher average incomes tend to pay higher effective tax rates under progressive personal income taxes, as these taxes are explicitly designed to ask more of taxpayers with a greater ability to pay.
The federal income tax and nearly every state personal income tax includes an adjustment for family size that results in lower tax bills on larger families that are judged to have a lower ability to pay. A single taxpayer earning $50,000 per year, for instance, likely has much more disposable income than a family of four trying to get by on this same level of income. Family size adjustments can take the form of expanded tax brackets or larger standard deductions for married couples, or dependent exemptions, Child Tax Credits, or Child and Dependent Care Tax Credits for families with children. The Earned Income Tax Credit (EITC) is also much more generous for families with children.
Average family size (impacted by both marital status and number of children) varies by race and ethnicity, and in some cases leads to meaningful differences in the effective tax rates faced at any given income level. This effect tends to be most noticeable for Hispanic tax units, who have larger average family sizes, though it varies by state.
Corporate Income Taxes
The federal government and 44 states levy corporate income taxes. For these profits-based taxes, three-fourths of the tax is assumed to fall on capital income, with the remaining one-fourth falling on labor income.
Our model data directly reflect differences in total income and several other characteristics by race and ethnicity and we therefore indirectly capture many differences in specific categories of income, such as the capital income and labor income amounts that are used to distribute the corporate income tax across income levels, races, and ethnicities. For instance, because a comparatively high share of elderly taxpayers are white, our model shows a high share of retirement income flowing to white taxpayers. Similarly, because a disproportionate number of high-income earners are white or Asian, a disproportionate share of capital gains and dividend income in our model flows to these groups.
Labor income and, to an even greater extent, capital income are both concentrated at the higher end of the income distribution where a disproportionate number of tax units are white or Asian. Because of this, white and Asian tax units face higher effective corporate tax rates than most other groups. In future research we hope to assess more deeply the extent to which differences in wealth holdings across race and ethnicity affect the distribution of capital income.
Consumption taxes – primarily sales and excise taxes – tend to be regressive, meaning that they result in higher effective tax rates on lower- and moderate-income families than on upper-income families. Because households of color are more likely than white households to be in the lower income groups, households of color overall tend to face higher consumption tax rates, on average, than white households.
While income level is the most important determinant of estimated consumption tax liability in our modeling of state and local taxes by race, we do make some adjustments for differences in motor fuel and tobacco spending that vary by race even within income groups.
General Sales Taxes
Forty-five states and the District of Columbia levy broad general sales taxes applying to tangible goods and, to a lesser extent, services as well. At least some local governments in thirty-eight states also levy these types of taxes.
For all race and ethnic groups, our analyses assume that income level can be used to predict both direct sales tax liability and indirect sales tax payments made when businesses pass along their own sales tax costs to final consumers through higher prices. In other words, effective tax rates for the general sales tax are held constant across races and ethnicities within each of the seven income groups underlying our data. When reporting our data, we combine those seven groups into three larger groups – the bottom 80 percent of earners, the top 20 percent, and all tax units. Under this method, differences in the distribution of tax units by race and ethnicity across the income scale lead to differences in the effective sales tax rate shown for each of the three larger groups. For example, because Black tax units tend to be overrepresented in the bottom half of the income distribution where effective sales tax rates are highest, the average effective sales tax rate for Black households within the bottom 80 percent grouping, and for Black households overall, tends to be higher than for most other races and ethnicities in these groups.
In arriving at this approach, we first analyzed variations in household spending patterns in publicly available microdata from the Consumer Expenditure Survey (CEX). Our analysis of the microdata focused on identifying potential differences in the spending patterns of white, Black, Asian, and Hispanic consumer units across the income distribution. Sample sizes were too small for us to reliably detect variation among other race and ethnic groups reported in the CEX.
We found meaningful differences by race and ethnicity in some specific categories of spending. For example, after controlling for differences in income we found that Black households spend more than average on utilities, white households spend more than average on entertainment, Asian households spend above-average amounts on restaurant meals, and Hispanic households tend to spend more on groceries. But when these specific categories of spending were aggregated into broad baskets of purchases that might be directly taxed under a general sales tax or that might be indirectly affected by business-paid sales taxes, most meaningful differences across race and ethnicity fell away and income usually became a reliable predictor of spending.
The most significant exception we found to this pattern of uniformity in spending in the broad baskets across race and ethnicity was for Asian households in the taxable spending basket. The national data indicate that Asian households spend less than average on the often-taxable categories of vehicle purchases, entertainment, and utilities and more than average in the exempt categories of education and shelter. Most of these findings are in line with those previously made by Fan (1997) and suggest that we may be slightly overstating the direct general sales tax payments made by Asian households in our analyses. We considered making an adjustment to our sales tax calculations to account for potential lower spending on taxable purchases by Asian tax units but ultimately opted against doing so because sample size limitations in the CEX prevented us from examining data for Asian households on a state-by-state basis and we suspect that some of the differences in spending patterns seen in the national data are either less pronounced at the state level or are absent entirely.
For instance, 50-state data on car access made available by the National Equity Atlas, produced by PolicyLink and the USC Program for Environmental and Regional Equity, suggest that Asian households overall are less likely than average to have access to a vehicle – a finding in line with the lower vehicle spending we observed in the CEX data. But in the PolicyLink analysis this finding only holds in 25 of the 46 states with available data. Asian households have below-average car access in California, Massachusetts, and New York, for instance, but above-average access in Alabama and Florida. Therefore, adjusting vehicle spending downward for Asian households across the board would likely improve our accuracy in some states at the cost of lessening it in others.
Similarly, lower spending on utilities by Asian households may be caused partly by the unique geographic distribution of those households – a disproportionate number of whom reside in milder west coast climates. Differences in utility spending are likely less pronounced in any given state than at the national level, but we currently lack the ability to make race-specific adjustments in this category of spending by state.
Our work thus far points to income being a reliable predictor of spending on broad baskets of purchases in most instances, though not necessarily for narrower categories of spending. Moreover, differences in income level are more important than differences in race or ethnicity when determining sales tax liability, and our decision to aggregate our standard seven income groupings into just three groups (bottom 80, top 20, and all) works to counteract divergences in spending patterns by race. In other words, our estimates of sales taxes paid by Hispanic tax units in the bottom 80 percent should be more reliable than our estimates of sales taxes paid by Hispanic tax units in the bottom 20 percent. We therefore conclude that using income level to predict sales tax liability is a reasonable approximation, but our analysis of the impact of general sales taxes by race and ethnicity is an ongoing endeavor and an area in which we hope to continue making advancements in the years ahead.
Another area worthy of future study is the potential correlation between local-level sales tax rates and the racial and ethnic composition of a given area. Our analysis suggests that it is reasonable to assume similar levels of spending across races and ethnicities for households in the same income cohort, but also that the rates at which those sales are taxed might differ to the extent that people of different races or ethnicities tend to shop in jurisdictions with divergent local sales tax rates.
For instance, throughout most of Pennsylvania consumers pay a 6 percent general sales tax but the rate in Allegheny County (including Pittsburgh) stands at 7 percent and the rate in Philadelphia is 8 percent. Because an above-average share of households in both these jurisdictions are Black, it is possible that Black households in Pennsylvania face a higher average effective sales tax rate than is reflected in this study. Modeling this type of impact for all races in all states would require detailed data on local sales tax rates and bases as well as differences in consumption habits by race across urban, rural, and suburban areas, as well as assumptions about the jurisdictions in which people do their shopping, which are sometimes different from the jurisdictions in which they live. In the absence of these data we generally assume that tax units of different races pay the same average local sales tax rate regardless of where they live.
Motor Fuel Taxes
Our motor fuel tax estimates are based primarily on the relationship found in the CEX between motor fuel spending and income level, but are adjusted to reflect differences in car access, by race and ethnicity, as reported on a 50-state basis in the National Equity Atlas, produced by PolicyLink and the USC Program for Environmental and Regional Equity. Households of color are more likely than white households to lack access to a car, even after controlling for differences in income. While the impact of this adjustment varies by state, its most significant effect is to reduce effective motor fuel tax rates for Black households, who tend to have the lowest rates of car access. It bears noting that many households lacking car access are contributing to funding state and local transportation networks through paying transit fares in lieu of gas taxes, but that these fare payments are not included in our analyses of state and local tax policy.
Our tobacco tax estimates are based primarily on the relationship between tobacco spending and income found in the CEX, adjusted to reflect differences in smoking rates by race and ethnicity that cannot be explained by differences in income alone. For this adjustment, we use 50-state data from the Kaiser Family Foundation’s analysis of the Centers for Disease Control and Prevention (CDC)’s Behavioral Risk Factor Surveillance System (BRFSS) 2013-2017 Survey Results. As with our motor fuel tax adjustments, the impact of this adjustment varies by state. But, generally speaking, the most pronounced effect is to adjust our estimates of effective tax rates for tobacco taxes downward for Hispanic households. While lower incomes are associated with higher smoking rates in the U.S. population as a whole, Hispanic households tend to have lower smoking rates than most other racial and ethnic groups while also having below-average incomes. Our adjustment captures this nuance, among other differences in tobacco usage across race, ethnicity, and state.
Most local governments and some states levy taxes on real property (e.g., homes and businesses) and, in some states, on personal property such as motor vehicles or business inventory. The net effect of most states’ overall property tax regime tends to be regressive, though the high concentration of property ownership among white households can lead to higher effective property tax rates, relative to income, on white households than on households of color.
Homeowner Property Taxes
We calculate homeowner property taxes by income level, based on IRS microdata for federal itemizers and on ACS data for non-itemizers. Because property tax rates vary widely within states and the ITEP model does not assign geographic identifiers within states, we use a single statewide legal tax rate (representing the weighted average rate used based on ACS data) to calculate the relationship between home value and tax. The model also uses statewide averages for assessment ratios, based on ratio studies conducted by state revenue departments. The final incidence of homeowner property taxes is assumed to fall on the owners of those homes.
For analyses disaggregated by race and ethnicity, homeowner property tax amounts are apportioned across the eight race and ethnic groups we examine primarily using self-reported property taxes paid (TAXP variable) found in the ACS microdata. Top-coding of this variable was addressed by allowing individual top-coded records’ property tax values to be determined using a combination of the record’s self-reported property value (variable VALP) and state-level disaggregated estimates of homeowner property tax revenue derived from Census Survey of State and Local Government Finances for each state.
In many states we find that white and Asian households tend to pay higher homeowner property tax amounts than Black, Hispanic, and American Indian households. This is partly due to white and Asian households having much higher rates of homeownership both within specific income bands and for these populations overall. Moreover, homes owned by white and Asian households tend to have higher market values, which also pushes up those groups’ average property tax liability. ACS data on household-level homeownership rates and average home value among homeowners, by state and race, are reported in the following table to help illustrate these findings.
But while white and Asian households tend to pay higher property tax bills overall, there are at least two effects that tend to push in the other direction, raising homeowner property tax bills for some households of color and often leading to higher property tax bills, as a share of homes’ market values, for those households of color.
First, recent research indicates that there is a large gap in assessment ratios between white, Black, Hispanic, and other non-white homeowners. Avenancio-León and Howard (2020) estimate that Black homeowners’ homes, for instance are over-assessed by an average of 12.7 percent relative to white homeowners’ homes. Holding all else equal, this gap increases the property tax amount paid by homeowners of color relative to white homeowners.
Second, statutory property tax rates (known as millage rates, or the amount of tax owed per thousand dollars of assessed value) may vary by jurisdiction within each state in ways that correlate with race and ethnicity. Martin and Beck (2017), for instance, conclude that Black homeowners face higher effective property tax rates than white homeowners, in part because of “the segregation of black homeowners in low-property-value jurisdictions that levy high property tax rates.” Because people of color tend to live in areas with lower home values, millage rates need to be higher in those areas to generate even a minimally acceptable level of property tax revenue.
By looking directly to the ACS’s reported values of homeowner property tax amounts, we capture all these differences in homeownership rates, home values, assessment ratios, and millage rates that impact the distribution of homeowner property taxes across race and ethnicity.
Residential Rental Housing Property Taxes
Our calculations for residential rental housing property taxes begin with household-level rent payments reported by state, race, and income level under the variable RNTP in the ACS microdata. The market value of rental units is then estimated using a combination of data from Zillow on each state’s median price-to-rent ratio and estimates from the Census Bureau’s Rental Housing Finance Survey of how that ratio varies by rental unit value. Finally, a tax amount associated with each housing unit is calculated using data from the Lincoln Institute of Land Policy and the Minnesota Center for Fiscal Excellence showing effective property tax rate applied to apartment buildings valued at $600,000 in each state’s largest city, some additional large cities, and a representative rural jurisdiction in each state. We blend those tax rates together using Census population data, including data on the share of each state’s residents living in rural and urban areas, to calculate a statewide average tax rate that is applied to all residential rental housing. Next, we apply that tax rate to our estimates of rental unit value to determine the property tax bill associated with each rental property. We then compare those property tax bills, relative to rent, against state-specific data on rental housing property taxes made available by tax departments in Idaho and Minnesota as a way of confirming that our calculations are producing reasonable property tax amounts.
The final part of this calculation requires an assumption as to the share of property tax on residential rental housing that the landlord is able to pass on to their tenant in the form of higher rents. We assume in each state that half of the tax is passed forward to the renter while the other half falls on the landlord. We are aware of studies finding pass-through percentages both higher and lower than this amount, but have concluded that this is roughly the midpoint estimate of the best available literature and in particular it is close in line with the estimates produced by Orr (1970), Hyman and Pasour (1973), and Black (1974).
Motor Vehicle Taxes
Motor vehicle property taxes and registration fees are based primarily on our estimates of the number of vehicles owned by each tax unit and, if applicable, those vehicles’ value. Information on motor vehicle ownership comes from Federal Highway Administration data. As with our motor fuel tax estimates, described above, these payments are adjusted by race based on data from the National Equity Atlas, with the most pronounced effect of that adjustment being to reduce motor vehicle taxes paid by Black tax units.
Other Property Taxes
The incidence of other taxes on property such as commercial real estate taxes, business inventory taxes, and estate and inheritance taxes is generally approximated based on the share of capital income flowing to each income group, though in some instances a portion of these taxes is assumed to fall on labor and/or consumption. We use race and ethnicity specific estimates of these variables (capital income, labor income, and consumption) to apportion these other types of property taxes.
ITEP MODEL DATA SOURCES
The ITEP model is a “microsimulation model.” That is, it works on a very large stratified sample of tax returns and other data, aged to the year being analyzed. This is the same kind of tax model used by the U.S. Treasury Department, the congressional Joint Committee on Taxation and the Congressional Budget Office. The ITEP model uses the following microdata sets and aggregate data:
IRS 1988 Individual Public Use Tax File, Level III Sample; IRS Individual Public Use Tax Files; Current Population Survey; Consumer Expenditure Survey; U.S. Census; American Community Survey.
Partial List of Aggregated Data Sources
Miscellaneous IRS data; Congressional Budget Office and Joint Committee on Taxation forecasts; other economic data (Commerce Department, WEFA, etc.); state tax department data; data on overall levels of consumption for specific goods (Commerce Department, Census of Services, etc.); state specific consumption and consumption tax data (Census data, Government Finances, etc.); state-specific property tax data (Govt. Finances, etc.); American Housing Survey; Census of Population Housing; Energy Information Administration; Federal Highway Administration; BDS Analytics; Zillow Group, Inc.; Centers for Disease Control and Prevention; Kaiser Family Foundation; National Equity Atlas (produced by PolicyLink and the USC Program for Environmental and Regional Equity); Lincoln Institute of Land Policy and Minnesota Center for Fiscal Excellence; Rental Housing Finance Survey.
Avenancio-León, Carlos, and Howard Troup, 2020. “The Assessment Gap: Racial Inequalities in Property Taxation.” Available at SSRN.
Black, David E., 1974. “The Incidence of Differential Property Taxes on Urban Housing: Some Further Evidence.” National Tax Journal 27 (2), 367-369.
Cilke, James, 1994. “The Treasury Individual Income Tax Simulation Model,” Department of the Treasury, Office of Tax Analysis.
Fan, Jessie X., 1997. “Expenditure Patterns of Asian Americans: Evidence From the U.S. Consumer Expenditure Survey, 1980-1992.” Family and Consumer Sciences Research Journal, Vol. 25 (4), 1997, 339-368.
Hyman, David N., and Ernest C. Pasour, Jr., 1973. “Property Tax Differentials and Residential Rents in North Carolina.” National Tax Journal 26 (2), 303-307.
Martin, Isaac W., and Kevin Beck, 2017. “Property Tax Limitation and Racial Inequality in Effective Rates.” Critical Sociology 43 (2), 221-236.
Orr, Larry L., 1970. “The Incidence of Differential Property Taxes: A Response.” National Tax Journal 23 (1), 99-101.
Rohaly, Jeffrey, Adam Carasso, and Mohammed Adeel Saleem, 2005. “The Urban-Brookings Tax Policy Center Microsimulation Model: Documentation and Methodology for Version 0304.” Urban-Brookings Tax Policy Center.