As expected, President Obama’s State of the Union address to the nation last week was suffused with the theme of opportunity. Over the past two years, the president has consistently tied the opportunities of poor and middle class Americans to rising income inequality, and Tuesday he stayed true to this narrative:
Today, after four years of economic growth, corporate profits and stock prices have rarely been higher, and those at the top have never done better. But average wages have barely budged. Inequality has deepened. Upward mobility has stalled.
This is a more ambiguous and less direct formulation than past ones, such as when Obama inveighed in December, at an event sponsored by the Center for American Progress, “[G]reater inequality is associated with less mobility between generations. That means it’s not just temporary; the effects last. It creates a vicious cycle.” Further, he insisted that, “[A]longside increased inequality, we’ve seen diminished levels of upward mobility in recent years.” Gone Tuesday were both the assertion that mobility has fallen and the claim that mobility has been directly affected by rising inequality. Nonetheless, the Administration clearly continues to want to link the two issues.
It has become harder over the past month. New research released between the December speech and the State of the Union strongly undermined the Administration. But its case has always been based on indefensibly weak evidence, proffered by a number of leading economists. The saga of the “Great Gatsby Curve,” the chart that has shouldered much of the weight of President Obama’s argument against income inequality, illustrates the extent to which that case has collapsed under the weight of recent findings.
The Great Gatsby Curve, introduced to the nation by then-chair of the president’s Council of Economic Advisors, Alan Krueger, displays a strong relationship across industrialized countries between income inequality and sons’ intergenerational mobility. One of Krueger’s research assistants came up with the moniker, after he offered a bottle of wine to the one who came up with the best name. Hopefully it was a good bottle.
Google “Great Gatsby Curve” and you get over 95,000 hits. A LexisNexis search shows over 500 unique articles citing the phrase. Last June, the White House included a splashy infographic of the Curve on its Tumblr, and Krueger highlighted it in his valedictory speech at the Rock and Roll Hall of Fame. It has been central to the Administration’s case that rising inequality has diminished economic mobility.
To appreciate the political work the Great Gatsby Curve has done, it helps to review the early days of the Administration’s anti-inequality campaign. In late 2011, in a highly-publicized speech in Osawatomie, Kansas, the president issued what would be the first in a series of salvos decrying income inequality. In the speech, the President cited mysterious statistics suggesting that mobility had dropped alarmingly over time:
You know, a few years after World War II, a child who was born into poverty had a slightly better than 50-50 chance of becoming middle class as an adult. By 1980, that chance had fallen to around 40 percent. And if the trend of rising inequality over the last few decades continues, it’s estimated that a child born today will only have a one-in-three chance of making it to the middle class—33 percent.
One of us (Winship) tracked down the source of those novel estimates, which contradicted the research consensus that mobility probably had changed little over time. The figures came from David Card, Berkeley economist and frequent collaborator with Krueger. They turned out to be the product of an elaborate set of back-of-the-envelope calculations that Winship showed to be deeply problematic. Thereafter, the figures were never publicly cited again by the Administration, to our knowledge.
But then as now, it clearly wanted to show that rising inequality had hurt mobility. In January 2012, the Center for American Progress hosted a speech by Krueger where he unveiled the Great Gatsby Curve. Derived from a chart developed over several years by Canadian economist Miles Corak, the Curve was put to innovative use by the Administration. We have reproduced the original Great Gatsby Curve here.
The chart plots a measure of inequality (the “Gini coefficient”) on the horizontal axis; countries further to the right have more after-tax household income inequality, and countries further to the left have less inequality. On the vertical axis, the Great Gatsby Curve shows countries’ immobility levels; countries with more immobility (less mobility) are toward the top, and those with less immobility are toward the bottom. Each of the dots in the chart represents a single country. The awkwardly-named “curve” is just the best-fitting straight line through the ten dots, shown in the above chart and also as Line One in the chart at the beginning of this essay.
Krueger was not simply illustrating that countries with more inequality tend to have less mobility, he was enlisting the Great Gatsby Curve to show that since inequality in the U.S. has been rising, diminished mobility will be the logical consequence. That is, absent consistent evidence that mobility has declined in the U.S., the Administration decided to project a decline into the future. By predicting tomorrow’s mobility from today’s inequality using the Curve, Krueger visually “showed” that today’s children will experience much lower mobility compared with contemporary levels. The dot in the upper-right corner of our prefatory chart shows the level of mobility he projected—a steep increase in immobility from that we have today.
This exercise, presented in an academic seminar in any of the nation’s research universities, would have elicited some serious raising of skeptical eyebrows, if not guffaws. The projection was a wild guess. It put a faux-precise number on a hypothesized relationship using a visually striking prop. Administration-friendly economists swooned. A post by Justin Wolfers on the Freakonomics blog conveys the extent to which the hopes of liberals hung on the Great Gatsby Curve.
It’s striking just how closely related inequality and mobility are. And it’s political dynamite. Why? If income inequality in one generation can be linked to unequal opportunity in the next, then income inequality can’t just be dismissed as the politics of envy.
At the time, Winship dismissed the forecast as “so rough that it is uninformative” and pointed out in a series of posts on Reihan Salam’s “The Agenda” blog the various specific shortcomings of the Curve. The best-fitting line failed to accurately predict a number of the data points, and therefore its ability to predict the United States’s future mobility levels was questionable. There was considerable imprecision around the individual countries’ mobility estimates and, for some countries, multiple estimates from which to choose. Using different estimates changes the steepness of the Curve and, hence, the projection.
Line Two in our prefatory chart, shown with more detail below, illustrates this sensitivity. For each of the eleven countries common to three Gatsby-Curve analyses, including Corak’s, we averaged the mobility estimates across the studies. We also switched to inequality estimates from the Luxembourg Income Study, which has taken great care to make each country’s measure as comparable as possible. Since it is flatter, projecting future mobility levels from Line Two yields a much smaller rise in immobility than the prediction from the Administration’s line.
Mobility as Re-Ranking
It turns out, however, that the most important shortcoming of the Great Gatsby Curve is an issue not raised in the “wonk fight” that followed the Krueger presentation. Unfortunately, understanding it requires a bit of wonkiness here. The mobility measure in the various published Great Gatsby Curves—called the “intergenerational elasticity,” or IGE—is an indicator of how much income gaps between children typically close in adulthood. These income gaps tend to close between generations because children are re-ranked between childhood and adulthood, so that richer and poorer children trade places. But even with re-ranking, income gaps will close less when inequality rises between childhood and adulthood than when it falls or is stable. Put another way, a given amount of re-ranking can produce different IGEs depending on how income inequality changes.
Similarly, two countries can have different IGEs either because one had more re-ranking than the other between childhood and adulthood or because each experienced different changes in income inequality. Indeed, it is possible for one country to experience more re-ranking than another but a bigger increase in inequality, with the result that the IGE shows less “mobility” in the country that has more re-ranking.
The IGE, therefore, reflects “relative mobility” (re-ranking) but also changing income inequality. We might expect, then, that a “mobility” measure that is affected by inequality trends would show a relationship to inequality levels. But another feature of the Corak/Krueger chart embeds the relationship even more deeply into the Great Gatsby Curve.
Most claims that income inequality produces immobility imply that it is inequality in childhood that is important, not inequality in adulthood, by which time re-ranking has played itself out. Since most IGE estimates come from data on adults in the 1980s or 1990s, ideally the Great Gatsby Curve should chart these mobility estimates against inequality figures from the 1950s, ‘60s, or ‘70s. However, comparable inequality estimates are available across countries only for the last few decades. The Great Gatsby Curve shows countries’ inequality levels in or around 1985, as do our alternate estimates.
That means that the Curve plots inequality in adulthood—at the end-stage of mobility. But recall that the IGE reflects the change in inequality adults have seen growing up. If nations with greater inequality in adulthood tend to have experienced larger increases in inequality leading up to adulthood, then inequality will be related to the IGE even absent any cross-national differences in relative mobility. More inequality growth after childhood in some country means both more adult inequality and a smaller reduction of income gaps as children age into adulthood. The upward slope of the Great Gatsby Curve is to some extent baked into the cake.[i]
More interesting than the question of whether a “mobility” measure that reflects inequality growth is related to inequality at the end-stage of mobility is the basic connection between relative mobility and inequality. Is it less common for rich and poor children to change places in adulthood when the gap between rich and poor is greater in childhood? That has been the source of both policy and academic interest. Answering the question calls for a version of the Great Gatsby Curve based on a pure measure of relative mobility.
A Quietly Devastating Paper
Fortunately, a new paper allows us to create such a chart. The paper estimates multiple measures of mobility and takes great care to use a consistent methodology for each country, but because the data challenges are so limiting it only looks at three of them—the U.S., Canada, and Sweden. However, the results are enormously informative nonetheless. Line Three in the chart at the top of this essay shows the Great Gatsby Curve for these three countries using the IGE from this new paper and the Luxembourg Income Study inequality estimates. It is a little flatter than Line Two, but not greatly different.
Relative mobility in the paper is measured by the “Spearman rank correlation coefficient,” which indicates the extent to which a son’s adulthood earnings rank (whether poorest, second poorest, richest, or anywhere in between) is predictable from his father’s earnings rank. Because this measure is unaffected by changes in inequality between the two generations, comparing countries’ rank correlations does not pick up inequality differences between them.
As shown in Line Four, when we plot the inequality levels of the U.S., Canada, and Sweden against their relative mobility levels, the Great Gatsby Curve indicates that higher inequality corresponds with less immobility, not more. The Curve points downward, and a projection from it would suggest a higher rate of mobility for today’s children rather than a decline.
These blockbuster findings come from Miles Corak, the paper’s first author. A version of it was presented nearly three months to the day after Krueger rolled out the Great Gatsby Curve and Corak highlighted an expanded version of it on his website.[ii]
It is worth emphasizing that this downward-pointing Great Gatsby Curve is only based on three data points, which would ordinarily make it next to meaningless. However, the U.S. had the most inequality of any country in the original Great Gatsby Curve and Sweden the least, yet their relative mobility levels are the same. The only other high-inequality/high-IGE countries in the chart were the U.K. and France. Evidence suggests that the U.K. and the U.S. have similar relative mobility levels. Recent analyses also provide a relative mobility estimate for Denmark that is reasonably comparable to those in the Corak paper (indicating remarkably high mobility).[iii]
Line Five in our opening chart gives the Great Gatsby Curve using relative mobility, assigning the U.K. the same mobility as the U.S. and Sweden and adding Denmark. It shows that higher inequality corresponds to very slightly higher immobility. The dot in the lower-right side of the chart indicates the projected mobility of today’s children implied by the Curve—barely worse than current levels and far better than the Administration’s projection in the upper-right corner.
Given how strong an influence Denmark, the U.K., and the U.S. have on the steepness of the original Great Gatsby Curve, it is unlikely that complete information on the relative mobility of the remaining nations in the Corak/Krueger chart would alter the conclusion. As Markus Jantti and Stephen Jenkins recently put it in the most comprehensive review of the income mobility literature to date, "[I]t is far from clear that intergenerational persistence and inequality are, in fact, as clearly positively correlated as [the Great Gatsby Curve] suggests.” They add, “[D]espite the public prominence of the ‘Great Gatsby’ curve, very little is known about how intergenerational income persistence and mobility vary across countries and how this relates to cross-sectional inequality.”
Cause and Effect
The predictive power of the Great Gatsby Curve, such as it was, did not quite require that inequality and mobility be causally linked, but the whole point of the exercise was to argue that rising inequality was lowering mobility. Right up to the end of his CEA chairmanship, including in his Rock and Roll Hall of Fame speech, Krueger argued that the “rise in inequality since the 1980s is likely to move us further out on the Great Gatsby Curve” toward higher immobility.
Corak has argued somewhat inconsistently. He carefully suggested, in a report highlighting the chart for the Zelig-like Center for American Progress in 2012, that it merely “invite[s] us to explore the underlying institutional and policy differences between the countries” before concluding that, “Persistent and growing inequalities will limit future opportunities.” In a paper last summer Corak argued that the “Great Gatsby Curve is not a causal relationship, but it is too glib to dismiss it by saying ‘correlation does not imply causation.’” But the paper concluded flatly that, “Inequality lowers mobility because it shapes opportunity,” The Curve is the primary empirical evidence Corak offers in support of this conclusion, and he otherwise falls back on theoretical conjectures about why the two might be related.
Since inequality has been rising for some time, if income disparities produce lowered mobility, we might expect to see mobility declining in the U.S. in recent decades. The academic literature on mobility trends is mixed, but the most common finding is that the change has been too small to detect reliably or to be substantively important. At the time of the Great Gatsby Curve’s unveiling, however, Winship’s preliminary analysis of Bureau of Labor Statistics data (finding flat to rising mobility) was the only attempt to look at children born in the period after the top one percent began pulling away from everyone else. That allowed Krueger, Corak, and others to argue speculatively that when children born in recent years become old enough to observe in the data, the inequality-mobility link will reveal itself.
Two papers released this month by Harvard economist Raj Chetty and his colleagues have severely weakened this conjecture. In one, Chetty and coauthors report that relative mobility in the U.S. was the same for children born in 1986 as in 1971, and likely the same for children born in 1993. Unchanged mobility in the presence of rising inequality, of course, casts doubt on the importance of the causal connection.
In a second paper, Chetty and his team find that there is little to no association between mobility and the share of income received by the top one percent, either across U.S. labor markets or industrialized nations. While they did find correlations within the U.S. between relative mobility and two other measures of inequality (including the one used in the Great Gatsby Curve), those correlations disappear when only the top one hundred labor markets are considered. That suggests that other factors differing between large and small labor markets may be behind the inequality-mobility relationship.[iv]
Indeed, this is a basic statistical point taught to all economics undergraduates: correlation does not mean causation. Even strong correlations are weak evidence of causality. In the cross-national context, there are so many cultural, political, economic, and demographic differences between countries that many, many variables are likely to be associated with mobility differences. Manhattan Institute scholar Jim Manzi found with little effort that a country’s population size was nearly as strongly correlated with mobility as its inequality level, and export-orientation and religious fragmentation were also strongly related to mobility. Sociologist Lane Kenworthy noted that Scandinavian countries have low inequality and high mobility but also distinctive education policies; it could be that these policies produce high mobility and that the low inequality levels are incidental. The Chetty paper on U.S. labor markets reveals an extensive list of factors correlated with mobility. The Great Gatsby Curve picks one factor from an unexamined list and elevates it to causal importance.
Strong correlations are one thing, but even before the Chetty studies there was ample evidence that inequality is only weakly related to economic and social mobility. Sociologist Deirdre Bloome looked at family income and found that across states, “The inequality to which children were exposed during youth has no robust association with the mobility they experienced as adults.” As Winship noted after Krueger’s introduction of the Curve, there is no relationship between wealth inequality and the earnings IGE across five industrialized countries, despite the fact that the correlation between income inequality and the IGE is strong across them. There is no relationship at all between income inequality and educational mobility across industrialized nations. The occupational mobility literature has tended to find only small differences in occupational mobility across countries, so they cannot be explained by inequality differences.
It remains true that we will not know whether today’s children will finally experience diminished mobility, and if they do, it, of course, remains possible that rising income inequality will be to blame. But the Great Gatsby Curve is of practically no use in assessing whether this fate will befall us. And the case for declining mobility is weaker than indicated by selectively cited social problems that are worsening.
The collapse of the Great Gatsby Curve, the weakness of the evidence that inequality and mobility are causally linked, and the accumulation of studies finding little change in mobility over time suggest two possible directions for arguments around inequality and opportunity. First, the Administration and its allies could continue their campaign against income inequality but point to other justifications than the harm it does to opportunity. Such a strategy is likely to run into similar problems however.
Second, Democrats could maintain their attention to insufficient upward mobility—a concern shared by a growing number of Republicans. But in that case, they should stop pushing income inequality as a primary factor behind unequal opportunities and desist from claiming that the American Dream is falling increasingly out of reach. Only three in ten adults today who started in the bottom fifth made it as high as the middle fifth. That is half the rate of upward mobility we would see if parental income and the things associated with it made no difference to child outcomes. That may be no worse than Sweden and no lower than fifty years ago. But it is too low. Is that not justification enough?
A debate on these terms between Democrats and Republicans around how to expand upward mobility would be enormously productive. And it would have the benefit of being intellectually principled.
Scott Winship is the Walter B. Wriston Fellow at the Manhattan Institute for Policy Research. You can follow him on Twitter . Donald Schneider is a research assistant in the Center for Policy Innovation at the Heritage Foundation. You can follow him on Twitter here.
[i] The timing of when inequality is measured—in adulthood rather than childhood—is another reason why the Administration’s projection of declining mobility was uninformative. The Curve “projects” the mobility today’s adults have experienced, not the mobility today’s children will experience. And we do not need to “project” what today’s adults have experienced, we can look at actual data.
[ii] The IGE estimates in the paper for the U.S., Sweden, and Canada are 0.40, 0.25, and 0.26, respectively, very similar to our averaged estimates from Line Two (0.44, 0.25, and 0.25).The rank correlations are 0.30, 0.30, and 0.24. Using estimates from the paper, we can improve on the new line we have plotted by using as the inequality measure not household income inequality for an entire nation in 1985, but earnings inequality, when the sons were living at home, among their fathers. The Curve is flat when sons’ relative mobility is plotted against the rich-poor (“90-10”) ratio of father earnings (not shown here). The Corak et al. paper also finds that the U.S. has upward relative mobility from the bottom that is comparable to Sweden and Canada, and downward mobility from the top comparable to Sweden (but lower than Canada). This finding contradicts a 2006 paper from Markus Jantti and colleagues that found the U.S. has less upward mobility among men than Sweden, as well as a 2009 paper by Corak finding American upward mobility is lower than in Canada.
[iii] Denmark’s rank correlation is plotted as 0.18 in our Line Five, which is technically the coefficient on parent income when child income is regressed on it. This differs from the estimates in the Corak et al. paper not only in the sense that this “rank-rank slope” is not technically the rank correlation but also in that the Denmark estimate uses incomes from tax returns (including spousal income if applicable and income other than earnings) rather than father and son earnings and includes daughters as well as sons. Nevertheless, the analogous U.S. estimate in the paper is 0.34, which is very similar to the 0.30 rank correlation in the Corak et al. paper. The IGE estimates for Denmark in the three analyses we average range between 0.12 and 0.15. Norway and Finland are the other two low-inequality/low-IGE countries, but it is unclear whether they should be closer to Denmark or to Sweden on the Great Gatsby Curve. Our Line Two puts them closer to Sweden, and if their rank correlations are also closer to Sweden, then Denmark’s rank correlation appears to be an unusually low outlier. As for high-inequality/high-IGE countries, the French IGE estimate is based on a single study using data that does not include direct measures of fathers’ earnings. Italy, which was not included in Krueger’s chart, is another high-inequality/high-IGE country, but its IGE, too, is based on a single study using data that does not directly measure father earnings. In short, no one should put too much stock in the IGEs for France and Italy, so it is impossible to say what their rank correlations would look like. Corak has produced iterations of the Great Gatsby Curve with additional high-inequality/high-IGE countries, but these countries’ mobility estimates are generally imprecise and based on statistical modeling by the original authors or Corak rather than direct computation. Most of them are developing countries, making their relevance to the U.S. questionable. And rank correlations for them are generally unavailable.
[iv] These one hundred labor markets include 70 percent of the American population. The two inequality measures are the “Gini coefficient” and the “interquartile range.” The latter was featured in an earlier paper by the team that did not include the former. The analyses in our earlier essay only included the interquartile range because the Gini estimates had not yet been made available. Our subsequent analyses have confirmed that the Gini coefficient is no longer correlated with relative mobility after restricting the sample to the top one hundred labor markets. Using a different approach that retains all of the labor markets but estimates a best-fitting line under the assumption that bigger labor markets should be weighted more strongly than smaller ones, Carter Price of the Washington Center for Equitable Growth found that the interquartile range is correlated with relative mobility. We suspect that he would find the same result for the Gini coefficient now that the new data is available. We are not convinced that bigger labor markets should be weighted more strongly, and we chose the approach used by Chetty that weights them equally. We restricted our sample to the one hundred largest labor markets out of concern that omitted variables related to labor market size are driving the correlation between inequality and mobility, a point which Price appears to have missed and which his approach does not address. The sensitivity of the correlations to these weighting decisions, we would argue, reinforces our point about omitted variables and strengthens the contention of our previous essay: The correlation between inequality and mobility is quite fragile, and bivariate correlations, should not, at any rate, be the basis for causal arguments.