My continually interrupted consideration of the new Jacob Hacker/Paul Pierson book Winner-Take-All Politics has led me to go back to the original inequality estimates of Emmanuel Saez and Thomas Piketty to understand better what they did. And what they did involves making a lot of assumptions about the size of their baseline populations (from which the top 1 percent is identified) and about the size of total income in each year (to which the top 1 percent's income is compared in order to get the top 1 percent's share of income).
I get really nervous about these kinds of assumptions, so I went looking for some alternative data sources. I remembered that the Federal Reserve Board's Survey of Consumer Finances explicitly samples very rich families separately from its primary survey efforts, drawing names of rich folk from IRS data and then tracking them down to interview them. This data is not perfect either--the response rates among the very rich are very low, and comparable data only go back to 1988--but the SCF has a number of features that are preferable to the Piketty-Saez dataset. For one, it provides direct estimates of the baseline population's size and of the population's total income. No need for worrisome assumptions to compute the top-share estimates. Second, a more comprehensive income measure may be used, including public transfer income, such as unemployment insurance benefits and worker's compensation, and retirement income from Social Security and drawn down from private pensions.
The Fed's main SCF analyst, Arthur Kennickell, put out a paper last year that provides top income shares for 1988 to 2006 (see Table 4). I plotted those against part of the Piketty-Saez series that includes realized capital gains, and here's what I found:
I get really nervous about these kinds of assumptions, so I went looking for some alternative data sources. I remembered that the Federal Reserve Board's Survey of Consumer Finances explicitly samples very rich families separately from its primary survey efforts, drawing names of rich folk from IRS data and then tracking them down to interview them. This data is not perfect either--the response rates among the very rich are very low, and comparable data only go back to 1988--but the SCF has a number of features that are preferable to the Piketty-Saez dataset. For one, it provides direct estimates of the baseline population's size and of the population's total income. No need for worrisome assumptions to compute the top-share estimates. Second, a more comprehensive income measure may be used, including public transfer income, such as unemployment insurance benefits and worker's compensation, and retirement income from Social Security and drawn down from private pensions.
The Fed's main SCF analyst, Arthur Kennickell, put out a paper last year that provides top income shares for 1988 to 2006 (see Table 4). I plotted those against part of the Piketty-Saez series that includes realized capital gains, and here's what I found:
Awfully consistent....What this tells me is that the concern I expressed in my Hacker-mania post that the Piketty-Saez series inflates the top income groups is probably not that big a deal. On the other hand, since the SCF data only go back to 1988, changes in the extent to which the P-S series inflate the top income groups may still create the exaggeration of the increase in inequality since the 1970s. But certainly the SCF estimates should be viewed as making that less likely.
So do I now think the Piketty-Saez estimates are correct? No--there is still the potentially very important issue of how people receive compensation and report their incomes on tax returns in response to changes in tax law. And that issue still affects the SCF data too because it affects what gets reported to SCF surveyors when they ask about specific types of personal income.
Just as importantly (maybe more), is the fact that neither data series accounts for employer and employee contributions to benefits like health insurance, and neither accounts for unrealized capital gains that investors accrue or the returns accruing to pension benefits pre-retirement. But I feel more confident that the assumptions Piketty and Saez make in measuring the quantity they claim to measure are solid (save that hugely important assumption that tax law changes don't meaningfully affect their series).
So do I now think the Piketty-Saez estimates are correct? No--there is still the potentially very important issue of how people receive compensation and report their incomes on tax returns in response to changes in tax law. And that issue still affects the SCF data too because it affects what gets reported to SCF surveyors when they ask about specific types of personal income.
Just as importantly (maybe more), is the fact that neither data series accounts for employer and employee contributions to benefits like health insurance, and neither accounts for unrealized capital gains that investors accrue or the returns accruing to pension benefits pre-retirement. But I feel more confident that the assumptions Piketty and Saez make in measuring the quantity they claim to measure are solid (save that hugely important assumption that tax law changes don't meaningfully affect their series).
RSS Feed