On January 12, 2012, Alan Krueger, then the chair of President Obama’s Council of Economic Advisors gave a speech at the Center for American Progress entitled, “The Rise and Consequences of Inequality in the United States.” The administration had attempted to show in December that rising inequality hurt upward mobility, but the rough modeling it conducted at the time to make its case failed to hold up. Krueger’s speech took a new tack—one that referenced the U.S.’s lower mobility as compared with other countries.
Modifying a chart that economist Miles Corak had produced earlier, Krueger plotted a number of countries as dots, arraying them along the dimensions of income inequality and intergenerational mobility. Following Corak, he displayed the straight line that ran most closely through the dots. This upward-sloping line—which Krueger dubbed the “Great Gatsby Curve”—indicated a positive relationship between inequality and immobility across countries; countries with more income inequality had more immobility (less mobility). Effectively arguing that this relationship was a causal one, he then plotted the current level of inequality in America on this chart and used the Great Gatsby Curve to predict a large increase in immobility in the United States arising because of increasing inequality.
In response to this argument, I issued a variety of criticisms. But it turns out that the biggest problem with the chart was one I neglected at the time, even though my earlier National Review essay had anticipated it: the Great Gatsby Curve was based on a comparison of income inequality levels to intergenerational elasticities (IGEs). As I discussed in Part 1 of this series, the IGE is a problematic measure of mobility. It indicates lower “mobility” when income inequality grows at a faster rate between generations. Countries with the same relative mobility (the same pattern of movement from bottom to middle, middle to top, and top to bottom) have different IGEs if they experience different inequality growth. Since inequality has grown more in the U.S. than in most other nations, its IGE shows worse “mobility,” and the same is true generally for countries with higher inequality growth.*
What no one realized at the time was that Corak’s own on-going research was demonstrating this problematic feature of the IGE. In a working paper with Bhashkar Mazumder and Matthew Lindquist, a version of which was publicly available** just three months after Corak and Krueger rolled out their Great Gatsby Curves, Corak found that when measures of relative mobility are used rather than the IGE, intergenerational mobility may be no lower in the United States than in Sweden, and only somewhat lower than in Canada. The higher rates of Canadian mobility reflect more downward mobility from the top rather than more upward mobility from the bottom.
The paper was published in the journal, Labour Economics, in October of last year. Its first key finding was that comparing IGEs across countries is potentially misleading unless data sources are of comparable reliability and the methods used are comparable. Corak, Lindquist, and Mazumder took great care to make the estimates from the three countries in their paper as comparable as they could. They found IGEs of 0.26 in Canada, 0.25 in Sweden, and 0.40 in the United States. In other words, the U.S. has the lowest mobility and Sweden the highest (barely beating out Canada). In Corak’s version of the Great Gatsby Curve, the estimates were 0.19 for Canada, 0.27 for Sweden, and 0.47 for the United States.
More important, however, is what the paper revealed about comparisons of relative mobility. In Part 1 of this series I described the “rank-rank slope,” a measure of relative mobility that gives the typical percentile difference in adulthood income between the poorest and the richest children. An equivalent measure used by Corak, Lindquist, and Mazumder is the Spearman rank correlation. This measure was 0.24 in Canada, 0.30 in Sweden, and 0.30 in the U.S.—very small differences given that this measure can range from -1.00 to 1.00. The authors indicate that part of the reason that the U.S. and Sweden have the same rank correlation is that the U.S. data include fewer years of earnings averaged together for fathers and sons. They estimate that if the Swedish data were comparable to the American data, its rank correlation would be 0.26. Even this is a small difference, though, and not substantively meaningful.
Their paper examined additional measures of relative mobility that describe specific kinds of upward and downward mobility. For example, the share of sons starting out in the bottom fifth of father earnings that remains in the bottom fifth in adulthood is 31 percent in Canada, 32 percent in Sweden, and 32 percent in the United States. That is to say, upward mobility from the bottom is no worse in the U.S. than in Sweden or Canada. The share of sons starting out in the top fifth of father earnings that stays in the top fifth in adulthood is 33 percent in Canada, 40 percent in Sweden, and 38 percent in the U.S. Sweden looks a bit better than the U.S. when the sample is adjusted to correspond better to the U.S. data, but not by enough to alter the conclusion that there are minimal substantive differences between the two countries.
Remarkably, the findings of this paper have been largely ignored. The only references to the working paper’s findings of which I am aware were in a comprehensive review paper by Markus Jantti and Stephen Jenkins in November 2013 and in an essay I wrote with Donald Schneider in early 2014. Google Scholar indicates the published article has been cited twice. Corak, to my knowledge, has never cited the findings. In contrast, a Google web search indicates over 250 hits for “Great Gatsby Curve” in the past twelve months, 35 of them from Corak’s blog.
Corak has sought to distance himself from the emphasis I have placed on his paper and my interpretation that it is devastating for the argument that the Great Gatsby Curve tells about inequality hurting mobility. On his blog, in a post that suggested I may have the facts wrong and that I was letting my ideological priors affect my interpretation of evidence, he wrote,
I also want to go on the record and note that when he says my ‘most recent paper highlights serious problems [with the Great Gatsby Curve and my] previous research,’ it should be clear that this is Mr. Winship’s interpretation of my research, and not my understanding of my own research.
He also suggested I should reach out to him to get his views of his results. I have since done so, but after a brief (cordial) email exchange, we failed to connect. But honestly, when it comes to interpreting evidence, it does not really matter what the author of a paper believes. If he does not believe his results, he should not attempt to publish the paper. Otherwise, it is the scholar’s job to lay out all of the relevant evidence in such a way that misinterpretations of it are not possible. I have not mischaracterized the findings of the paper. It is true that I generally have not noted the caveats included in the paper by Corak and his coauthors that downplay their findings—and I should emphasize here that I know and admire Mazumder—but that is because I do not find the caveats compelling.
The authors argue that the IGE may be a better indicator of inequality of opportunity than relative mobility measures because it incorporates income inequality. For the reasons I gave in Part 1 of this series, I don’t believe that aligns with the way most Americans think about opportunity, and it begs the question of why income inequality per se is important. Regardless, the IGE is definitively not a good measure of relative mobility.
More substantively, Corak and his coauthors note that their U.S. estimates indicate more mobility than previous research, citing estimates from the 2006 paper by Jantti and his colleagues that I mentioned in Part 1 of this series—and to which I will return—as well as a study by my former colleagues in the Brookings Center on Children and Families for my former colleagues at the Pew Charitable Trusts. They note that both those studies use parental family income rather than father earnings in childhood. They found U.S. results more comparable to these earlier studies when they looked at combined parental earnings instead of fathers’ earnings.
The authors also cite research indicating that the administrative data they use to obtain their U.S. estimate may be worse than survey data at capturing low incomes accurately, implying that their estimate for the U.S. may be inferior to estimates based on survey data.
Let’s put these concerns to rest right now. First, as the authors note, they cannot estimate mobility for Sweden and Canada using combined parental income or earnings, so we have no way of knowing whether Sweden and Canada would have more mobility than the U.S. by this approach. Even if they did, that would not invalidate the paper’s finding that male earnings mobility is not substantively different across the three countries. Without any estimates of family income mobility in other countries, it is simply putting a thumb on the scale for the authors to discount their own findings on this basis.
More to the point, there are estimates of father-son earnings mobility in the United States against which to compare the Corak, Lindquist, and Mazumder estimates against. Using a survey called the Panel Study of Income Dynamics, Pew estimated that 31 percent of men who grew up in the bottom fifth of father earnings remained in the bottom fifth of earnings as adults. Corak and his coauthors estimated it was 32 percent. Their conjecture that the administrative data produce too-high upward mobility because they poorly measure incomes at the bottom is simply wrong.
Even the paper’s finding that Canada has more mobility than the United States—and than Sweden—should be viewed as provisional. One reason to be concerned with the Canadian estimate: in a 1999 paper with Andrew Heisz, Corak reports that about half of sons are excluded from the Canadian data used in the new paper because they did not file taxes, could not be matched to fathers who were tax filers, or both. This group includes many sons of immigrants and sons of single mothers, who are generally excluded from all analyses of father-son mobility. But even accounting for these men, that leaves a sizable fraction of sons out of the Canadian sample. To be included, sons had to file tax returns while living at home as adolescents. As Corak notes in the paper with Heisz, there are good reasons to think that the omitted sons are poorer than the sons in the sample. He conducted a crude test in the earlier paper of whether the omission biased his results and found that it did not, but I think most economists would characterize the test as not especially informative.
But nothing in the Corak, Linduist, and Mazumder paper suggests that U.S. and Swedish levels of mobility differ meaningfully from each other. That still leaves the 2006 paper by Jantti and his coauthors, which found that the U.S. had lower relative mobility—at least for sons starting out at the bottom—than Denmark, Norway, Sweden, Finland, and the U.K. I’ll explain why this paper’s conclusion is also incorrect and explore some additional research comparing the U.S. to other countries in my final installment.
*In fact, when combined with another feature of Corak’s and Krueger’s versions of the Great Gatsby Curve, a relationship between inequality and “mobility” was almost baked into the cake. That’s because the measure of inequality they used indicated inequality levels in adulthood rather than in childhood. If people in countries with higher inequality in adulthood tend to have experienced stronger inequality growth between childhood and adulthood, then a positive relationship between income inequality in adulthood and immobility is mathematically inevitable.
** The link is now dead and unavailable on archive.org, but was http://www.eale.nl/Conference2013/program/Parallel%20session%20A/add215310_konuoeQdIq.pdf. A later draft was also posted online but is available now only at archive.org: https://web.archive.org/web/20150401075918/http://www2.sofi.su.se/~mjl/docs/Cross_country_mobility_March_21_2014.pdf. A copy of the earlier draft is in my possession.
Scott Winship is the Walter B. Wriston Fellow at the Manhattan Institute for Policy Research. You can follow him on Twitter here.
Interested in real economic insights? Want to stay ahead of the competition? Each weekday morning, e21 delivers a short email that includes e21 exclusive commentaries and the latest market news and updates from Washington. Sign up for the e21 Morning eBrief.