When we ask the question in the title, we normally go back to the annual surveys of income distribution conducted by the Census Bureau under the name of Current Population Surveys (CPS). I am using here, to derive the figure below, the “lissified” (internationally harmonized) CPS data for 2010 and 2013. The data very clearly show that the average tax rate (federal and state) increases in disposable income. For the top percentile, it reaches 34 percent. This is exactly what we would expect from a progressive income tax system: richer people pay higher share of their income in taxes.
There are however two problems with these data. First, survey data underestimate income at the top (most of all, income from capital). For example, interest, dividends etc. represent only 7% of market income in survey data, but about 12% of income in IRS fiscal data. And most of capital income is received by the very rich. (And to make things simpler I will speak throughout of capital income like interest and dividends, not of capital gains or losses.) So if top incomes are underestimated then the calculated average tax rate (taxes/income) shown in Figure 1 is overestimated. The “true” tax rate is less. I will call the underestimation of top incomes by surveys, “top income survey bias”.
But there is an even bigger problem: tax data shown here are only estimates, calculated by CPS from the simulations based on what people with such reported incomes should pay in taxes. CPS does not ask for taxes paid; it asks only for all types of incomes and then applies the tax rules in the simulations it runs to come up with the estimates shown in Figure 1. The bottom line is this: we really do not know how much is being paid in taxes.
Let’s move to the second source: fiscal data provided by the IRS. These are the data used, among others by Saez and Piketty to show changes in US inequality over the long-run. But while IRS provides for the recent years, a rather large sample of income data (for up to several hundred thousands of respondents) it does not provide data on taxes paid. Moreover, the income concept used by the IRS, the so-called Adjusted Gross Income, is what should be best described as “political income”. It is whatever income the Congress decides is income. It excludes such obvious economic incomes like interest on state and local bonds. It also excludes all non-taxable transfers like social security benefits or cash value of various in-kind socials transfers, or indeed imputed income from own housing. (Obviously, if something is not subject to taxation, you are not going to report it on your tax form.) At the very top, the data are “blurred” for confidentiality reasons: some income fields are represented as means of several households. The same approach (“censoring” or “top-coding”) is used by household surveys too. Finally, IRS data do not cover the entire population because some 5-7% of people do not file tax returns.
Yet its most serious defect for our purposes is that income reported by the rich is a “castrated” income, compared to a true economic income, because it deducts all kinds of “expenses” that no economist would consider as such but which the fiscal system allows. Vacations, travels, lunches, are treated as “business expenses” with which to offset income, so that the ultimately reported fiscal or political income may be significantly lower than a real income as defined by economists. Many of us have heard of business trips to France in July, entirely written off (thus reducing the fiscal or “political” income). Such expenses should normally be classified as any other consumption item and not as an income deduction. Similar examples abound. The most important thing is that such loopholes (as they are I think mistakenly called) are really designed and used only by the rich. Majority of people whose income is in wages do not have incentive or wherewithal or rationale to create false companies or consultancies whose main objective is to reduce their fiscal income. So, for the top income group, we are faced here too with a significant underestimate of income. Let us call it “top income fiscal bias”.
Thus we come to the third source of data: Congressional Budget Office (CBO) that gets the data on fiscal (political) income from IRS plus some confidential data on taxes actually paid, and tries to match this information with income from household surveys. CBO is quite forthcoming about the serious data inadequacies it faces (see appendix on Data and Methods, p. 31).
For 2011, CBO comes up with the estimate of the overall federal tax rate (including indirect taxes and social security) by income level as shown in Figure 2. The top 1% is estimated to have paid some 29% of their market income in taxes, a rate higher than that for other groups.
While CBO does probably the best job possible with what it has, note that the results are based on (1) matching of information, not on actual data on taxes and incomes, and, more importantly, on (2) income data which are seriously underestimated at the top. They suffer from both survey and fiscal top income biases, in addition to another bias due to tax evasion, which I have left out of consideration altogether, but which is certainly the greatest among the top.
Or to give a concrete example. In 2011, the average market income of households that were in the top 1% was, according to the CBO, $1.45 million. The calculated federal tax rate was, as we have seen, 29%. But to $1.45 million, we should add income that was wiped out thanks to the political income bias plus all income that was simply not reported whether because of survey inadequacies or because of tax evasion. If these two sources add to say 50% of the officially reported income, then the average tax rate is no longer 29% but 19%. And we cannot be sure that it is the highest average tax rate assessed as we would normally expect if the system were progressive throughout.
In conclusion, I think it is fair to say that, a century after the system of direct personal taxation was introduced in the US, one really does not know what is the tax rate paid by the very top of the income distribution. It could be that Mitt Romney’s average tax rate of 13% is not that uncommon.