The Economics of the Great Gatsby Curve

In an article on the Brookings Institution website that was originally posted by the National Review, Scott Winship questions the idea that greater inequality at a point in time is associated with less generational mobility over time — what the Chairman of the Council of Economic Advisors, Alan Krueger, called the “Great Gatsby Curve” in a speech given on January 12th.

Winship’s article does a disservice to a well-established literature on generational mobility by suggesting that the basic information Krueger used is in some sense invalid. Krueger’s Great Gatsby Curve is in fact well-rooted in the labour economics literature, and debate would be better placed addressing the policy implications he draws than to suggest that President Obama’s top economist feels compelled to create his own facts.

So in the spirit of moving evidence-based public policy forward here is a quick review of the underpinning of the Great Gatsby Curve in both theory and practice.

[Warning: when I started this blog I promised myself that it was intended for non economists, and that I would not be speaking to other economists. But this post is an exception. However, if you persevere, there are important lessons in this discussion for what good labour economics is, in my mind, supposed to be about.]

1. Economic theory predicts that inequality will lead to less generational mobility

The workhorse theoretical model for intergenerational dynamics is presented in a couple of papers by Gary Becker and Nigel Tomes, the original one published in the Journal of Political Economy in 1979. Their analysis was extended by Gary Solon in a way that offers a framework for making comparisons across time and space: indeed, Solon called his article “A model of intergenerational mobility variation over time and place.” (The source is here.)

If you want to understand differences between countries or differences within a country over time, then Solon’s theory suggests you organize your thinking around three broad factors: how families function and how effective parents can be in determining their children’s human capital; how labour markets function, in particular the earnings return to human capital; and how public policy functions, in particular the extent to which government policy is of relatively more benefit to the relatively disadvantaged (what Solon calls the “progressivity” of public investment).

His reworking of the Becker-Tomes model makes the prediction that he summarizes in the last sentence of his paper: “an era of rising returns to human capital or declining progressivity in public human capital investment is also an era of declining intergenerational mobility.”

It is important that discussions of the Great Gatsby curve be informed by economic theory.

2. All of Winship’s statistical concerns with estimates of the generational elasticity have been addressed since the beginning of this literature

The Great Gatsby Curve has been in circulation since 2004, and the estimates of the generational elasticity it uses address all the concerns Winship raises, and in fact some of his concerns seem to stem from a misunderstanding about the appropriate use of statistical method.

Solon takes the returns to human capital to be a marker for the degree of earnings inequality in a society, and on this basis I offered what to my knowledge is the first version of the Great Gatsby Curve in a 2006 publication (figure 3), which was in circulation since about 2004. Here is the associated PowerPoint for a presentation called Do poor children become poor adults? (slide 33) that I first made in 2004 while still working at Statistics Canada, and countless times since.

This first version of the curve follows Solon’s model closely, and uses the return to a college education relative to a high school education on the horizontal axis. It also uses estimates of the intergenerational elasticity I culled from a meta-analysis of all publications in the literature at that time.

Winship says:

Each of the points in the Great Gatsby chart represents an immobility estimate taken from independent earlier studies. It is very difficult to get comparable estimates of immobility for different countries. One needs measures of income defined in a common way across countries. Ideally, one would have multiple years of income data for each generation, and the analyses would use real data, as opposed to model-based estimates, on adults and on their parents when they were children. This last requirement is perhaps the hardest to meet. Many governments do not conduct studies tracking children’s income as they grow older (or have only started to recently), so researchers must estimate childhood income using an algorithm obtained from a separate data set and compare the result against actual adult-child income.

It is correct to say that multiple years of data are needed. Averaging over multiple years is one way of reducing measurement error. We have known this since Gary Solon published a 1989 article in the Review of Economics and Statistics, since he and David Zimmerman published separate articles in the same issue of the American Economic Review in 1992, and in fact we have known this since Tony Atkinson and his colleagues published a book using UK data in 1983.

But the issue of measurement error is handled in my literature review, where I begin by pointing out that if existing studies are not critically assessed on this dimension there is no basis for making international comparisons: at the time of my literature survey the published US estimates ranged from less than 0.1 to over 0.6. Following Nathan Grawe, I also pointed out that estimates could also differ because of life cycle biases arising from the parent’s earnings being measured at different ages.

I proposed a method for choosing the best estimates of the intergenerational elasticity for the purposes of international comparisons that accounts for differences in study designs: including, the degree to which measurement error is addressed, differences in the point of the life cycle the measurement is made, and differences in the statistical method used.

Further, all of the estimates use “real data” as opposed to what Winship calls “model-based estimates”. It is not clear what he means by this term, but the insinuation is that Krueger is somehow using simulated data. In some studies statistical methods—Instrumental Variables—totally appropriate for addressing measurement errors are brought to bear. But “real data” is being used; it is being used in a way that recognizes limitations in the data; it is being used with statistical methods appropriate for the challenges at hand. Even if in some sense these methods generate “predicted” values, this is an improvement over the so-called “real” data upon which they are based.

Two-Sample Split IV is also used, but this is a great way of combining information from cross-sectional surveys to obtain intergenerational estimates when panel data do not exist. But to suggest that many “governments do not conduct studies tracking children’s income as they grow older (or have only started to recently), so researchers must estimate childhood income using an algorithm obtained from a separate data set and compare the result against actual adult-child income” is at best misleading.

In fact, many countries do collect panel data, and many countries make administrative data available. And if a country does not, labour economists have appropriate ways to address the issue that lower the need for high cost surveys that take a long time to bear fruit. This is a good thing.

But either way my literature review corrected for the difference in statistical method by recognizing that IV based estimates tend to be higher and represent an upper bound.

This information has been in circulation for eight years, and in a peer-reviewed publication for six years.

3. Winship’s questioning of the robustness of the empirical literature on the intergenerational elasticity by citing one study is misleading

To the best of my knowledge the first person to change the variable on the horizontal axis to the Gini coefficient was Jo Blanden when she was working at the London School of Economics. I don’t know if her research has been published, but you can see the Great Gatsby Curve in Figure 8 of her 2009 discussion paper.

There is also a 2009 paper by Dan Andrews and Andrew Leigh, called “More inequality, less social mobility,” that shows the curve using data from comparable information sources.

More recently in a forthcoming paper, highlighted in my previous post, I offer another version of the curve using intergenerational earnings estimates on 24 countries by repeating my original literature review.

All of these studies are based on defensible estimates of the intergenerational elasticities, but Winship says:

Because of these technical difficulties, for some countries—particularly the United Kingdom—researchers estimating immobility come up with widely varying estimates. Building a Great Gatsby chart then requires choosing an immobility estimate to represent the country. Different scholars choose different estimates, with the result that their best-fitting lines differ. In the version I trust the most, there is a relationship between inequality and immobility, but it is entirely driven by three countries—the United States, Italy, and France. In particular, the estimates of immobility for the United Kingdom range widely across different versions of these charts.

The reference he is making is to a paper by Bjorklund and Jantti using information on 11 countries, that shows a weak positive relationship. Winship never says why he prefers this study to all of the others. But he argues that if you drop the three countries in this study with the highest Gini coefficients the relationship is flat. It is not clear why you would do so if the estimates are valid? Drop more than one-quarter of the sample at one end of the distribution in any study and you are likely to get different results. This criticism is no more valid than suggesting that the weak positive relationship the authors uncover is driven by the three countries with the lowest elasticities. Drop them and you have a very strong positive relationship.

A claim that “Different scholars choose different estimates, with the result that their best-fitting lines differ” suggests a degree of arbitrariness that inappropriately questions the care researchers have used in this literature.

Take a look at the Andrews and Leigh paper and you see all the country estimates derived by the authors using exactly the same type of data, with exactly the same definitions of variables, with exactly the same statistical methods. The Gatsby Curve is still there.

Take a look at my recent literature survey using as large a sample that is available from comparable published estimates. The Gatsby Curve is still there.

Or forget the intergenerational elasticity altogether and use parent-child correlations in years of schooling from the excellent paper by Tom Hertz and his colleagues, as Blanden does in her Figure 6. The Gatsby Curve is still there.

A fair statement of this literature is that the finding of a positive relationship between inequality at a point in time and the degree to which it is transmitted over time is robust.

4. Which Gini coefficient to use is debatable, and besides it likely won’t matter that much

Winship’s final point is what I will characterize as splitting hairs, something I actually mentioned to him in a Twitter conversation before his post.

He says:

Admittedly, all of the charts like this that researchers have created show a relationship between inequality and immobility. There is one other big problem with them, though, particularly when a researcher is estimating future mobility, as Krueger did. If one believes that inequality diminishes opportunity, one should look at how inequality experienced in childhood affects mobility between childhood and adulthood. Krueger’s immobility data are, for the most part from people who were in their 30s during the 1990s, so they should be matched with inequality data from the 1960s, when those people were children. But instead they are matched with inequality data from 1985. Cross-national differences in inequality were much bigger by 1985 than they were in the 1960s (see Figure 1.1), so it is likely that the relationship across countries between childhood inequality in that decade and immobility experienced by the 1990s is weaker than the Great Gatsby chart suggests; that is, if countries had relatively similar inequality levels when the 1990s adults were children, then inequality would be a less salient factor in explaining mobility differences thirty years later.

Yes, to some extent one wants to use the Gini prevailing during the time children are being raised as it might capture the capacity of parent’s to invest in their children’s human capital. But Solon’s model is more subtle than just that; parents make their investments based upon expectations of the return in the future, when their children are adults. So it is not self-evident as Winship suggests that Gini coefficients from an earlier period are entirely appropriate.

Krueger uses Gini’s for the 1980s: not perfect from Winship’s perspective but not unreasonable, and not clear why this would be characterized as a “big problem”. Further, Winship has not made clear why he would use Gini’s for any one particular year, rather than some sort of average over the entire time a child is being raised, or why he would ignore the importance of future expectations.

I don’t think this is a fundamental concern that would reverse the result.

5. Making projections is tricky, but not because of anything Winship says

The point of all this is that there is a robust positive relationship between inequality at a point in time and its transmission across generations: it is a relationship that is predicted by theory, and it is evident in descriptive cross-national comparisons.

But can we make projections on this basis? What are the policy implications?

I am struck by the type of comparisons that have been focused upon in some of the discussions on this issue, for example Denmark and the United States in the New York Times.

There is no way the United States can mimic the outcomes of Denmark in the way Danes have made that accomplishment: a geographically small country, ethnically homogenous, with high levels of trust, and a labour market notably more structured is not a guide for American public policy.

Getting the slope of a least squares regression “right” is not going to make any difference.

Cross national comparisons don’t tell us how to go from here to there. Their value is to highlight the institutions and public policies that Solon pointed our attention to as determinants of generational mobility, and in this way force us to ask how they operate in our own country.

The Great Gatsby Curve is a great communication device to start this sort of conversation, and that is the way I interpret Krueger’s “projections” of what intergenerational mobility will be in the United States when current levels of inequality are reflected in the adult outcomes of today’s children.

If you listen carefully to his presentation at the Center for American Progress, you will note that he gets interrupted from the floor with a question when he starts making this projection.

The question was about steady-states, and this is a concern from an academic perspective. But the projection is legitimate because it motivates the second part of the talk, which is about the type of society our children are likely to inherit; about the type of society we want; about what public policy can do about it all.

This is exactly the way to think about the “projections”: think of them as calling for a review of how families function, of the nature of the labour to which they must connect, of how government policy supports them. To bring us to this stage is the real purpose of the Great Gatsby Curve.

I tried to do this in a limited way in a paper recently published by the Russell Sage Foundation, but the best example of the kind of research moving public policy discussion forward is a book by Scott Winship’s colleagues at Brookings, Ron Haskins and Isabel Sawhill, called Creating an Opportunity Society.

If you want to form credible “projections”, then read this book.

6. The Bottom Line: the facts are right

In my mind, Alan Krueger is first a labour economist, and only second the head of President Obama’s in-house think-tank. In his current position he has to deal with the challenges of communicating academic rigor to a wide audience, but he also has to do so as a labour economist. The Great Gatsby Curve does just that.

I think of good applied labour economics as a three-legged stool. The stool supports an important question that relates to how people live their lives. The first leg is academic rigor: a respect for theory, innovative yet appropriate use of data, and a thorough understanding of the existing literature. The second leg is policy relevance: research that is tied to new facts or trends, framed in terms of potential policy levers, and produced in a timely way that can impact policy debate. The third leg is the broad communication of research findings in an accessible way.

Labour economists contribute to evidence-based public policy by getting this balance just right. The facts need to be right, but they also need to be communicated so that democratic debate can focus on the implications for public policy.

Alan Krueger is entitled to his own views on public policy, but not to his own facts. But this also holds for public policy entrepreneurs in the blogosphere.

In this case Krueger got his facts right. In this case he linked them to public policy. And in this case he communicated them in a catchy and memorable way.


31 thoughts on “The Economics of the Great Gatsby Curve

    1. I don’t think the way Krueger defined it, as 50% around the median, is inappropriate or something to complain about. Krueger’s general point seems right, but I don’t address this issue in my post because it is not my immediate area of expertise.

  1. I would like to understand the differentiation between what is defined as “poor”, “middle” or “rich” in each country.

    Of course countries with high GINI coefficients introduce a barrier to moving from one income quintile to another – that barrier is called math.

    If the US has quintiles defined as (guesstimated mean of each income group; poor = 1/5 quint.; middle = 3/5 quint.; rich = 5/5 quint.)

    The US is about this:
    The poor = $15,000 per year
    The middle = $45,000 per year
    The rich = $300,000 per year

    Whereas France may be this:
    The poor = $15,000 per year
    The middle = $40,000 per year
    The rich = $200,000 per year

    It is harder to move from poor to middle to rich in the US because it takes more money to move than in other countries. Economists measure movement between classes, while not measuring the 1) money increase or 2) the effect of earning more money on the quality of life.

    1) In the example it takes $25,000 to move from being a “poor” French to a “middle” class French. That same $25,000 increase doesn’t clear the bar in the US. Using GINI as a measure does not tell the whole story. In each, the family has an additional $25K income. In the case of the French family, they are counted as moving 2 quintiles – from poor to middle. While the US family, has only moved 1 quintile, and hasn’t obtained what economists call “middle” class.

    2) The monetary $25K increase, above, will have a direct and debatably equal increase of the quality of life for both the French and American family. They will both have more money to get larger living quarters, better transportation, and increased quality of goods, food and clothing. While the “rich” French family will not achieve the same level of security due to the lower standard of being “rich” in France compared to the US.

    Specious definitions such as this is why average Americans do not trust their leaders. Obama is using tortured economics to wage class warfare, pure and simple.

    1. Doug T, I think you’ll find that what you’re saying here is that if the US were less unequal, it would have more social mobility, which I think is precisely what Krueger and Corak are saying.

      Less facetiously, social mobility could mean lots of things, but we usually think about it in terms of people’s relative positions: their place in a social ‘rank order’. People could get richer or poorer without ever getting ahead of those higher up in the ranking (or the reverse), but I don’t think it would be what most of us understand by ‘mobility’.

      An African-American brought up by a single mother making president – now that is what I call mobility.

  2. A very interesting piece from a non-economists perspective, but I think you have pretty well established why economics can’t be used to solve social problems. Even well established ideas like the Gatsby Curve are meaningless if anyone with an opposing opinion can invalidate it in the public’s eyes by raising a few, already addressed, criticisms. The same thing happen in the more scientific field of biology, so what chance does economics have?

    Does economics lead us toward the truth? Maybe, but much like religion its only in the mind of the individual. For those who choose to believe it is true, for those who don’t, it is not.

    1. Excellent!
      These are really important questions. I guess the mindset I have been trained into is that economics is a science, and our approach to the subject should be disciplined by the scientific methods. Indeed, every first year student is introduced to the distinction between “positive” and “normative” economics, what is versus what should be. But as you so correctly point out the boundary between these two is easy to blur.
      In the end though my own experiences have not been as pessimistic as you suggest.
      Thanks again for raising this.

  3. Congratulations Mr. Corak for this excellent text.
    I’d like to know what series for Brasil were used by you to estimate the intergenerational elasticity between father and son earnings.
    Many thanks

    1. Thank you for the kind words.

      You can find the references for all of the estimates on a January 12, 2012 post called: Here is the source for the “Great Gatsby Curve”, which contains this document. As you can see the estimate for Brazil is adapted from a research paper by Christopher Dunn in the 2007 issue of the Berkeley Electronic Journal of Economic Analysis & Policy.

      I should also let you know that I have paid attention to the fact that researchers in different countries may use slightly different methods according to the nature of the data available to them. In putting all of these countries together I have paid attention to these differences in methods, and tried to correct for them. The way in which I do so is described in a 2006 paper available on my IZA web site, here. As such the estimates are the best estimates for international comparison, not necessarily the best estimates for a particular country when looked at in isolation.

      1. Many thanks for your kind help, Mr Corak.
        If it is possible, do you see any problem in send me the paper of Christopher Dunn in the 2007, considering the fact that I’m not able to obtain it in The B.E. Journal of Economic Analysis & Policy.

  4. Pingback: Anthropology News

What do you think?

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.