How Much Are the Right Friends Worth?

My guest today is Harvard economist Raj Chetty, who in my opinion, is doing the most important and transformational research of any social scientist on the planet.

CHETTY: You don’t need the incredibly complicated statistical tools or more opaque models that can be useful in lots of context in understanding things. But sometimes you can just show people simple averages and that can tell a lot of the story.

*          *          *

Welcome to People I (Mostly) Admire, with Steve Levitt.

*          *          *

I first saw Raj Chetty present his work when he was only 23 years old and somehow already completing his Ph.D. He was speaking about his dissertation, which was impossibly ambitious. It combined cutting-edge techniques from both macro and macroeconomics. And it also tried to make advances in both economic theory and data analysis. Academic talks like these almost always failed disastrously. It’s just hard to do so many things well. And economists pride themselves on taking apart freshly minted PhDs who think they know it all. The room was packed. Standing room only. I waited for Raj to get destroyed, but to my amazement, he fended off every criticism. People loved what he had done. It was obvious at that moment that Raj Chetty would become a superstar. But I don’t think anyone could have predicted the type of research that would ultimately define Raj and change the way economists think about the world.

LEVITT: Raj, it’s one of my few regrets as an economist that you and I have never been able to do a project together, because I just love to observe up close how great economists think.

CHETTY: Thanks so much, Steve. It’s a pleasure to be here with you.

LEVITT: For the first six or seven years of your career, you’re working on topics like social insurance, corporate taxation, interest rates, investment. And you were very successful. You’re getting published in the top journals. You got tenured Berkeley in just four years. But watching from a distance, it seems like a time came where you altered sharply your research approach, switching to pretty simple research methods. Essentially, you decided to mostly be a data-focused microeconomist. Is that right?

CHETTY: Yeah, Steve, I started out with interests in a variety of different topics and methods, but there was a point where I wanted to broaden scope a little bit. And it was around that time in 2009 when I, and some of my colleagues, got access to information from anonymized tax returns covering the entire population of the United States. And that would really open the scope to ask some basic questions that economists had been interested in for a long time on the determinants of inequality, what makes kids succeed, and things like that. A lot of what theory and econometrics is about is trying to fill in the gaps that we can’t see with the data that we currently have. And this, for the first time, was really complete and rich data. And so, I tried to take the approach of just playing to that strength.

LEVITT: And I would say that the hallmark of Raj Chetty 2.0 — the economist I think you’ve turned into — is that you are asking these big, important questions that other people have given up on answering because they’re too hard. And the first example I’m aware of in this new vein of research you’re doing, was work trying to figure out whether the particular teachers you had in school — say, your fourth-grade teacher — whether they had a long-term impact on your life. Did that teacher affect whether you attended college? Did he or she affect how much money you earned 20 or 30 years later? And I was working in the area at that time, and I never thought about trying to answer that question because it just seemed completely impossible to get one’s hands on the data necessary to answer that question. So, I’ve always wondered: How did you get access to that I.R.S. data?

CHETTY: We — and many others over the years — had recognized that longitudinal administrative data — so, by that I mean, think of the exhaust pipe of the modern digital economy, where we all do all kinds of things that leave a trail of information about our behavior, about our choices, that can be very useful for social scientists. That kind of data is very different from the traditional way of collecting data, which is to survey people and ask them, how much money do you make? Where do you live? What’s your educational background? And so, what this new — if you’d like, “big data” — paradigm gives us is a data set that allows you to follow people systematically over time. So, if you take the question of how does your kindergarten teacher affect how much you’re earning when you’re 30 years old, you need to follow people systematically over 25 years and you need to follow lots of people, right? Because many things happen between kindergarten and when you’re 30 years old. So, there’s going to be lots of other variables that play into determining your earnings outcomes. And so, you need big samples, and that’s what these new tax data offer. And so, recognizing that potential, I and several others — Emmanuel Saez, John Friedman — had been thinking about, you know, there’s Scandinavian countries like Denmark and Sweden, where people were starting to work with those data. I had written some papers using Austrian data, Danish data, but being based in the U.S., and interested in the questions in the United States, I kept making inquiries with the federal government. Could we get access to similar data in the U.S.? And eventually there was a breakthrough in 2008, 2009, where we were able to work out a contract with the Internal Revenue Service to use these data for research purposes on a large scale for the first time.

LEVITT: Now, I’m amazed you had a breakthrough with the I.R.S. because probably 10 years before, I went to the I.R.S. and I wanted to help them catch tax cheats. And their response to me at the time was: “No way in the world would we ever let you look at the data, even if your only objective was to help us catch people who are cheating on their taxes.” When I found out that you had succeeded, I was stunned that you had the talent and the perseverance to pull that off. And that has been an input to some amazing research findings. So, what did the data show on this question? Do teachers matter?

CHETTY: So, what we did was took data from an experiment that I’m sure you’re familiar with, Steve — the Tennessee STAR experiment, the famous experiment back in the late 1980s that was conducted in Tennessee that basically randomized children in 79 schools to classes of different sizes. So, they basically flipped a coin. And if you got lucky, you were placed in a smaller classroom, and then if you didn’t get lucky, you were placed in a larger classroom. And that experiment has become the basis for many social science studies to understand how classroom environments early in childhood affect how kids do.

LEVITT: And these were kindergartners, first graders, second graders?

CHETTY: Exactly kindergartners, first graders, second graders, and third graders were involved in this experiment.

LEVITT: Okay.

CHETTY: And so, the many previous studies that have been written about this experiment — there are hundreds — have looked at short-term impacts of being randomized to a small classroom. There’s a famous paper by one of our late colleagues, Alan Krueger, who looks at the causal effect of being assigned to a small classroom on how kids do in later years. And the kind of disappointing finding that emerged from that literature is that you see these short-term impacts that look really exciting. Kids who are assigned to smaller classrooms, to more experienced teachers are doing better at the end of kindergarten and a little better at the end of first grade and so forth. But you see a dramatic pattern of what’s called “fade out,” whereby the time you look at these kids in third grade or fourth grade or fifth grade, they basically look like they’re doing just about as well — if you look at things like standardized test scores — as kids who didn’t have those better classrooms when they were young. And so, when you look at that data, you can see why people were skeptical about the potential impacts of these early childhood classroom interventions. Yes, maybe you set kids on a better trajectory temporarily, but it doesn’t really seem to last.

LEVITT: Okay. So, you’re then going to go to an enormous amount of trouble to take these same kids and see what happens to them 20 or 30 years later. But why would you think that’s an interesting question? We already know the answer. It won’t matter. It didn’t matter three years later. Why would it matter 30 years later?

CHETTY: Well, I think part of the answer, honestly, Steve, is a bit of naivete. Going into it, I had the maybe simple intuition that, “Wow, I can think about some teachers I had when I was a kid who really seemed to matter in my life.” It seems intuitive to me that this could be important in determining kids’ long-term outcomes. I figured, okay, let’s give this a shot. Let’s take that Tennessee STAR data and link it to the tax records and follow these 12,000 kids from kindergarten to when they’re about 30 years old. And somewhat surprisingly, those kids who were in the smaller classrooms with the better teachers early on in childhood, they’re doing significantly better as adults. And so, that first paper we put out was “How does your kindergarten classroom teacher affect your earnings?” And the answer was they affect your earnings quite a bit. The type of class you’re randomly assigned to in kindergarten matters a lot for whether you go to college and so forth.

LEVITT: So, what’s a lot? Can you just quantify what does “a lot” mean in this context?

CHETTY: Yeah. So, to give you one example of magnitudes here. So, I think there was a New York Times article after we put out the study that was titled something like, “The case for the $320,000 kindergarten teacher.” What they’re basing that off of is if you have a very good teacher, or are in a better classroom, you earn as an individual child something like $15,000, $20,000 more over your lifetime, multiply that by the size of the average classroom, say with something like 20 kids, and you end up with a number like $320,000 value for a top-notch kindergarten classroom environment, or a highly skilled teacher relative to an average teacher.

LEVITT: So, a good teacher compared to an average teacher, is that the comparison you’re making? That’s worth about $300,000 to his or her students over their lifetime in future incomes.

CHETTY: That’s right. And importantly, that’s the value of that having that teacher for a single year. Right? Now, at one level, folks in the general public react by saying, “Yeah, that makes sense. teachers matter, that seems very intuitive.” But then from the perspective of this evidence that we were just talking about on the fade out, you’ve got this kind of puzzling pattern now that we labeled “fade out and reemergence,” that these same teachers whose impacts seem not to last, based on things like subsequent standardized test scores, actually their impacts are reemerging when kids become adults in the form of higher college attendance rates, lower teenage-birth rates, higher levels of earnings, and so on.

LEVITT: So, I went back and looked at our email exchange that we had back around the time of the paper. And I saw you present it very early on, and I wrote to you, and I said, “Raj, love the paper. It’s so hard for me to believe the results, but I can’t see what’s wrong with it.” And your response to me was, “Well, I wasn’t sure either, but I just got my hands on a new set of data and it’s bigger and better, and it shows the same thing.”

CHETTY: That’s right. What we then did was got a much larger data set from the New York City public school district and linked all of that data, on two and a half million kids who went to New York City public schools over a 20-year period, to the tax records once again. Now, what we had as a benefit is a much larger data set where we could look for these patterns more systematically and use what economists now call quasi-experimental methods — rather than literally an experiment where we’re flipping a coin, take advantage of lots of kind of pseudo-experiments that arise in big data sets. So, let’s say you and I go to the same school and there’s a really good teacher in fourth grade who happens to go on maternity leave when I come along in the next year and enter fourth grade — let’s say I’m one year younger than you. So, effectively we have something that looks like an experiment where you happen to have this really good teacher and I don’t because of a sort of random event from the perspective of the students. Presumably the teacher’s maternity leave is not related to the quality of students that happen to be around. And so, using that kind of variation in the set of teachers who happened to be around, we’re able to look with this much larger data set at the causal effects of teachers on kids’ long-term outcomes. And remarkably, and I think I was doing this research right when I got your email, we were seeing exactly the same patterns — a fading out of impacts on test scores. And then this reemergence very clear evidence of impacts on outcomes in adulthood.

LEVITT: In what you just described, you focused on these quasi-experimental methods, but, honestly, one way to look at the data is just the simplest way — is just to look at the teacher I got, and does it matter? Even though it’s not randomly assigned, and the results that you’re talking about, those emerge loud and clear in those data with the simplest possible approach, right?

CHETTY: That’s right. So, the simplest way to look at the data is to create what people now call “measures of teacher value-added”. We’re going to take a bunch of teachers and look at their students’ test scores before they had that teacher and after they had that teacher. So, to make it concrete, let’s say you’re a fourth-grade teacher. You teach 25 kids. Essentially, what I’m going to do is take your students’ average test scores at the end of fourth grade, minus their test scores at the beginning of fourth grade. And if that number is very positive and large, I’m going to say you’re a high value-added teacher. And if that number is negative, I’m going to say you’re a low value-added teacher. And we show that if you, as a student, happen to be assigned to a high value-added teacher, you’re doing much better in the long run than a student who happens to be assigned to a low value-added teacher.

LEVITT: And this analysis has all sorts of far-reaching implications. the first, is it certainly suggests that there’s a value in high stakes testing at a time when critics of these tests are numerous and vocal.

CHETTY: I’d agree with that. I think with some caveats in the sense that we are looking at one measure of teacher quality, and I would not necessarily conclude from that that the only, or even the best way to measure teacher quality is purely based on these high stakes tests. You could think about principals trying to evaluate teachers, even students trying to evaluate teachers. You could think about other forms of testing that are not the standardized tests, but socio-emotional skills. So, my main takeaway is that teachers really seem to matter, figuring out how we get higher-quality teachers and keep higher-quality teachers in our classrooms can have huge payoffs for society, especially kids from lower-income backgrounds who often may not have access to high-quality teaching. I think this is one potential way to quantify how well teachers are doing and measure their performance, but should be thought of in the context of a suite of other techniques as well.

LEVITT: Yeah, I’m curious, do you know the work of Nolan Pope? He was one of my all-time favorite students, and he’s now a professor at the University of Maryland. He measures teacher value-added on dimensions other than test scores, just like you just talked about. Some teachers are better at getting kids to show up at school than others. Some are better at keeping kids from getting suspended. He first finds that the correlation for teacher value-added on test scores and these other dimensions is really low. It’s only about 0.15. So, the teachers who are good at one thing, aren’t necessarily good at another thing. And he also — just like you, he finds these long-term effects of these other dimensions of good teachers. So, he looks at grade-school teachers in Los Angeles and he can see effects on high school graduation 10 years later of these teachers who are good at doing other things. It exactly corroborates your view that test scores are just one element of what teachers do, but an important one.

CHETTY: That’s exactly right, and part of why I was saying what I was saying, Steve, is drawing on the great work that Nolan has done that folks like Kirabo Jackson have done, people are increasingly finding that teaching, like many other things, is multidimensional. It’s not just about teaching kids to do better on standardized tests. There are many different skills that are being instilled and it’s not the same teacher that does all of these things in the same way. There’s actually a tie-in back to where we started on these issues, because you might remember, you know, this whole puzzle that we’ve been focused on of the fade out and the reemergence. Folks might be wondering, how is it that teachers don’t have lasting impacts on how well students are doing on standardized tests, but do have lasting impacts on things like earnings? And I think the answer actually relates potentially to some of these other dimensions, where we and others have accumulated evidence over the years that there are important impacts on things like non-cognitive skills. How disciplined are students? How well do they get along with others? Things that may not immediately manifest themselves and say how well you do on your sixth-grade math test, but probably are pretty valuable in the labor market, right? Getting along with other folks, showing up at work. And so, I think that multidimensionality starts to give us a picture into why we get what seemed like this really odd statistical finding at the beginning.

LEVITT: So, good teachers matter. That’s the conclusion. It seems like the one easy, clear policy that emerges from it, although not necessarily a popular one, is you should just fire the worst teachers, right? That this system where really terrible teachers have their jobs for life is just a disaster. Would you agree?

CHETTY: That’s one way one could look at it. I worry that’s a bit draconian and might have downstream effects that we’d need to worry about in equilibrium, who wants to become a teacher? Are people worried about how they’re being evaluated? And so, on. The way I look at it is on two dimensions. First, I think setting aside the lower tail of teachers who may have the lowest value-added. I think there’s a real problem in terms of retaining the high value-added star teachers, of whom they’re relatively few to begin with. Those folks have lots of other great options, typically in the private sector, where they can make a lot more money and potentially have more stable and rewarding careers. And so, I think one major challenge is focusing on how we recruit and retain the most talented teachers in our current system.

LEVITT: So, one implication of that would be a much more skewed salary distribution among teachers. Obviously, teachers unions stand directly in the way of that.

CHETTY: Look, one challenge in the current way that pay is structured in the United States in most public schools is that there’s very little scope for principals to reward teachers who are doing an exceptional job and who might have an attractive offer from another private school or maybe someplace outside education. And indeed, there’s some recent evidence which suggests that a lot of the great performance that we see in certain types of charter schools is precisely because they have the sort of latitude to retain the best performers. But zooming out a bit, simply increasing flexibility and pay may not be enough. If you look at a lot of the countries that have very effective educational systems — places like Finland, for example — my sense is that equilibrium is completely different. If you are a top performer in school, one of your aspirations as a kid would be to become a teacher because teaching is held up as a high-status position that’s very attractive. And I worry that in the United States, we’re not in that equilibrium. Fixing that and pay is, presumably part of that, could go a long way. The second thing I was going to say is, I think there’s also a role to be played in focusing on teachers who may be struggling at present and figure out how we can improve their quality. We certainly have some evidence that teacher quality can change. There’s a very clear pattern in the data that teachers who are very inexperienced are much less effective than teachers who just have a couple years or three years of experience, but more broadly, there’ve been many efforts to try to implement training programs, to improve teaching quality, and they haven’t shown a tremendous amount of success yet. I don’t think people have quite cracked the code in terms of figuring out how you make someone a good teacher. But I think insofar as we can figure that out, that also would have tremendous value.

LEVITT: So, your results suggest that parents should jockey like crazy to get the best teachers for the kids. That’s an implication of what you’ve got.

CHETTY: Look, among many things parents can do, having your kid in a good environment and getting your kid to a good teacher is likely to be valuable. Now, that said, it’s not like any one teacher or any one intervention really determines anyone’s destiny. The way I look at it is there are lots of small inputs. Think of it as like a hundred different things that affect kids from who their friends are, to where they’re growing up, to which teachers they have, to the size of their classroom. And you get a hundred of these draws. And if you get more positive shocks on each of them, then you’re going to do better on average. And if you get more negative shocks, it looks like you’re going to do worse on average. And for parents, it’s not that, you know, if your kid doesn’t have the right kindergarten teacher, your kid is somehow doomed. Far from the case. But on the margin, if you can get your kid to a better school and to a better teacher, that’s an investment that can really pay off.

You’re listening to People I (Mostly) Admire with Steve Levitt and his conversation with Raj Chetty. After this short break, they’ll return to talk about Raj’s work on social mobility.

*          *          *

LEVEY: Hey, Steve. So, our listener Gord wrote in with an idea. He wants to start a marketplace for medical appointments. For example, if patient A has an appointment time to see a doctor that patient B wants, there would be an opportunity for patient B to pay patient A for the appointment. Patient A would have the choice to accept or reject the offer. Now this raises some ethical issues, so there’d have to be a system that kept patient confidentiality in place, of course. But what do you think of this idea?

LEVITT: Well, I like the spirit of a market. Markets are often good when it comes to allocating scarce goods. And in general, we think that markets are efficient in accomplishing that goal. I think at this particular setting, there would be two big possible downsides you’d have to worry about. The first is the ethical issues, not about confidentiality, but just about who will end up with the appointments. It reminds me a lot of the market for organs and how people are nervous about having a market for kidneys, because they’re afraid that rich people will end up with kidneys and poor people will end up with small amounts of money for having given those kidneys, but will regret their decisions. Now we’ve talked at length about why I think there’s a solution to that, which is to pay a lot of money for kidneys, but that’s not the point here. The point here is that almost for sure, if you create this market, the people willing to pay the most for the appointments will be more affluent people. So, there will be a transfer of appointments to rich people and money to less-rich people. And the medical ethicists would flip. They would hate that. Now you could say, “Oh, it’s a transaction between consenting adults. Why don’t you let it happen?” But it just is a fact that within medical ethics that would make people very angry.

LEVEY: Right, and you talk about what a fan you are of a marketplace for kidneys during the second episode we had with game designer, Jane McGonigal, in case listeners want to hear more around your idea. What is the second reason you think could be a hold up to this idea?

LEVITT: In the current system, there’s no reason to set up an appointment that you don’t need. But in this new market-based system, people will have a strong incentive to make appointments, even if they don’t really want them. Of course, you can try to make it based on you need a specialist to recommend you or whatnot. But I think there will be many people on the margin who don’t really want to take the appointment, but who want to make some cash. And I suspect they’re going to seriously clog up the appointments and you’re going to be left with a situation where essentially nobody gets an appointment unless they pay for it. And in a situation where appointments are even more oversubscribed, then the ethical issues become even stronger because now you basically might not even be able to get an appointment unless you’re willing to pay top dollar to have it.

LEVEY: I can think of a third reason that this might be a difficult idea to put into place.

LEVITT: Let’s hear it.

LEVEY: Often when people need to go to a specialist it’s because they have a pertinent medical issue that really needs attention. And I think if patients started swapping their times, this could lead to issues with some undiagnosed, for example, brain tumor being delayed because the patient keeps getting paid for their appointment time. So, as a result, the specialist is never able to actually deal with their medical issue until it’s maybe too late.

LEVITT: And those undiagnosed brain tumor issues will be concentrated among low-income people, who are likely the ones selling it over and over. And I think that’s the kind of argument that feeds the ethical fears around this kind of program. But I will say, Morgan, I’ve seen a system like this work at least once. And I told this story when I had Richard Thaler on the show a while back. At the University of Chicago, the way that they allocate slots into classes is by lottery. And I had a class that was heavily oversubscribed, and it seemed so inefficient to me, that we wouldn’t have a better mechanism for allocating slots in my class. So, I said offhandedly, as far as I was concerned, if anyone in the class wanted to sell their spot to someone who would pay more for it, that would seem like a useful transaction. And I can’t see anything wrong with it. And a few days later, the Dean of Students called me and said, “I’m really sorry to have to tell you this, but there’s been a terrible event on campus. It turns out that one of your students has sold her spot in your class to another student and we’re taking the appropriate steps to punish her for what she’s done.” And I said, “Wait, you might not want to punish her, because I encouraged the students to do that.”

LEVEY: Is it true that you hired that student as a research assistant later on?

LEVITT: I did, there was such a hubbub about the whole thing when the Dean of Students threatened her with punishment, she got cold feet, and she withdrew the offer and she ended up taking my class. And we had so much communication around it. And she was really nice and interested in economics that I ended up hiring her as a research assistant.

LEVEY: Gord, thanks so much for writing in with your idea. If you have an idea or question for us, our email address is pima@freakonomics.com. That’s P-I-M-A@freakonomics.com. It’s an acronym for our show. Steve and I read every email that is sent and we look forward to reading yours.

 *         *          *

In the second half of our conversation, I want to trace the evolution of Raj Chetty’s work on economic and social mobility. From his first studies done over a decade ago to his most recent project, which uses Facebook data in an incredibly creative way. If you want to see big data put to good use Raj is your guy.

LEVITT: You co-direct an organization at Harvard called Opportunity Insights and on the webpage, you cite two facts based on your research that are about as stunning to me as any social science facts I’ve ever seen. So, you claim that of all American children born in 1940, 90 percent would grow up to earn more than their parents. But now, in the U.S.A., 50 years later, doing that same analysis, you find that less than half of the people grow up to earn more than their parents.

CHETTY: So, Steve, the American dream is really faded. That’s what those numbers say to me. Back in the 1940s and 1950s, if you were a kid growing up in this country, you could pretty well expect that you were going to achieve the American dream as defined as upward mobility, rising up relative to where your parents were. And if you look at kids entering the labor market today, those prospects don’t look as good.

LEVITT: What data do you analyze to get those numbers? Is this I.R.S. data again?

CHETTY: So, this is I.R.S. data, but there’s a little bit of statistical machinery in the background to make this work. So, the I.R.S. tax data, which allows us to follow kids over time and look at how they’re doing relative to their parents directly, at present, that’s only available for relatively recent years.

LEVITT: And is that because back in the 1940s and ’50s — well, for one thing, you didn’t have to put social security numbers of your dependents.

CHETTY: There are two problems. One, you’ve got to be able to link kids to their parents. That got much easier after 1987, when by law, in order to claim a dependent exemption, you had to write your kid’s social security number on your tax return. Because kids basically became worth money in the tax system, 99 percent of kids get claimed by their parents after 1987. And you can then make these intergenerational linkages. And then the second issue, obviously, back in the 1940s and 1950s, people were not e-filing. So, things are on paper and there’s a lot of work to be done to link everything together that has not yet been done.

LEVITT: So, it’s straightforward once you have it, you’re done, or —

CHETTY: Well, so, in the recent period, we were able to use the tax data to measure the link between parents and kids’ incomes and calculate that 50 percent number relatively in a straightforward way. Think of a big Excel spreadsheet of kids’ incomes and parents’ incomes, and I’m just going to count what fraction of times the kids end up earning more than their parents. The problem historically is we don’t have that spreadsheet. So, what do we do? Historically, you have data on what economists would call cross-sectional income distributions. You can’t link the kids to their parents, but you can look at the kids separately and you can look at the parents separately. But it turns out — and this is really the crucial observation of this paper that was published in Science — that back in the 1940s and 1950s, there was so much growth that basically all kids ended up earning more than all parents did in the prior generation. It doesn’t matter which kid is linked to which parent. There was so much broad-based growth that virtually all kids ended up earning more than virtually all parents. And so, we are able to get to that 92-percent number using a methodology that we develop in this paper.

LEVITT: So, I’m still trying to wrap my head around that 50-percent number that people coming of age today are only a coin toss as to whether they’ll be more affluent than their parents. We know the economy’s been growing over roughly the last 30 years. That G.D.P. growth in the U.S. has been something like 2 or 3 percent per year. So, if you factor in compounding, the overall size of the U.S. economy must be about twice as big. But population’s grown as well, maybe 1 percent per year, but that still leaves G.D.P. per capita roughly 50 percent higher than 30 years ago. So, are you saying essentially all of those gains have gone to the upper tail?

CHETTY: That’s exactly it. We have an enormous amount of growth over the past 30 years. And on average, we have a lot more total output. People are richer, on average, than they were 30 years ago. So, how can it be that only 50 percent of kids are doing better than their parents? And the answer is, take an extreme example. Suppose one person — say, Jeff Bezos or Bill Gates —got all of the gains in G.D.P. over the past 30 years. Then you could have a tremendous amount of G.D.P. growth and you’d still have only 50 percent of kids doing better than their parents. We’ve had a lot of skewed growth where people at the top of the distribution have gotten most of the gains and people in the middle and the bottom have essentially gotten no gains since the 1980s. And that’s what leaves most kids in the same, or sometimes even a worse position than where their parents were.

LEVITT: You’ve also studied how the degree of social mobility varies as a function of where you grew up. Can you describe your initial forays into that?

CHETTY: Yeah, absolutely. So, to tie the threads of our conversation together a bit, we talked about getting access to the tax data and how this could potentially illuminate many issues that economists cared about. And it struck me that there were really two key dimensions where we had a lack of evidence. One was following people for long periods over time — that’s what led to the teacher’s work and education work. And then the other key dimension was zooming in to look at specific subgroups. So, usually when economists work with survey data, traditionally, you’ve got data on a few thousand folks. And so, you make statements about broad national averages. What is happening in the U.S. as a whole, in terms of average incomes or inequality or other trends. There’s never enough data to say what’s happening in Chicago versus Boston, or in Hyde Park versus other parts of Chicago. And tying back to the discussion we were just having on the fading American dream, seeing trends like that made us wonder, “Well, what can we actually learn about the determinants of economic mobility? And so, one way to learn about that is to break the data down and go beyond the national picture and ask, “What do a kid’s chances of rising out of poverty look like in Iowa versus California versus Massachusetts? What do they look like in one part of a city versus another part of a city?” And the tax records give you enough precision to be able to literally measure economic opportunity. A chance that a kid born to a low-income family rises to the middle class or beyond, neighborhood by neighborhood across America. And so, what we did is created what we call an Opportunity Atlas, which folks can actually look at on the web, opportunityatlas.org. You can just type in your address and look up the data for your own area. And what that gives you is an incredibly precise picture of how children’s chances for rising up — actually, it turns out very dramatically across the U.S., even in the present day, there are some places where kids have a great chance of rising up. Much of the central Midwest, the Great Plains, kids born to low-income families there, even today, have great chances of rising up beyond where their parents were. And then there are other parts of the country, much of the Southeast, industrial cities in the Midwest, like Detroit and Cleveland, where if you’re born to a low-income family, you don’t have very good prospects of rising out of poverty.

LEVITT: So, you didn’t, at that time at least, have the right kind of data to make strong causal claims about why Iowans had upward mobility, and people who grew up in Cleveland didn’t. But you could at least see what features tended to be more or less common in those high-end mobility places, right?

CHETTY: That’s right. The beauty of this is you see this map, and folks can see this on the web, where you see much higher rates of upward mobility in some parts of the country than the others. And that kind of creates a canvas for a conversation that researchers can have, that policy makers can have, to evaluate various hypotheses for what might be driving these differences. So, why is it that some places have higher levels of upward mobility than others? And so, what we and many others started to identify is that this is not just random variation. There’s some pretty systematic predictors of these differences in upward mobility across areas. For example, places that are more mixed income, where you have less concentrated poverty, tend to have higher levels of upward mobility. Places with more two-parent families tend to have higher rates of upward mobility. Connecting back to our earlier conversation, places with better schools tend to have higher levels of upward mobility. So, there are a bunch of different things that seem to at least predict these differences across areas.

LEVITT: And when you look specifically at African Americans, do you see similar patterns or something different?

CHETTY: It quickly became evident from looking at these maps, that the cities with larger African American populations, places like Detroit, Cleveland, and so on tended to be the places that had lower rates of upward mobility. And so, we began to wonder, given the long history of racial disparities in America, how much of these differences is really about differences across race, as opposed to differences across place? And so, to get at that, we were able to link data from the census to information from tax returns to look at the racial breakdown, look at upward mobility separately for Black Americans and white Americans. And what we found is that there are enormous differences in rates of mobility for Black and white kids and Black and white boys, in particular. So, if you take a Black boy and a white boy who are starting out in a family at the same income level, same wealth level, same family structure, even going to the same school, we see that the white boy has a much better chance of reaching the middle class or beyond than the Black boy does. In contrast, and interestingly, if you look at Black women versus white women, you see essentially the same rates of economic mobility. So, there are enormous differences by race, but they seem to intersect with gender.

LEVITT: And what about Hispanics?

CHETTY: The Hispanic population in the U.S. is a very interesting contrast with the African American population in the sense that if you just look at snapshots of data and look at average incomes, you would find that for the current generation, Hispanic Americans don’t look all that much richer or poorer than African Americans, they’re in the same place. But if you look at mobility across generations, Hispanic Americans are on a totally different trajectory. Their rates of economic mobility look much closer to those of white Americans than Black Americans. And so, what we show in this paper published in 2020 is that if you run that forward and make a prediction about where Hispanic Americans are going to end up, their incomes are going to end up being fairly close to white Americans over time as they make the climb up the income ladder. Whereas for Black Americans, given the enormous differences in rates of economic mobility, unless we change something fundamental in the U.S., those differences are going to persist across generations.

LEVITT: Let me skip ahead to your latest research, which was just recently published in the journal Nature, which feels to me almost like the grand finale in this 15-year mission you’ve been on to understand patterns in mobility. And it’s different from your earlier papers, for instance, it relies on Facebook data rather than I.R.S. data. And it definitely has a decidedly sociological feel to it rather than straight economics. So, can you just give a quick summary of this new research?

CHETTY: I hope it’s not the grand finale. I hope there’s more to come and that we’ll learn more. But, yes, it certainly is coming out of sociology. So, as we were doing this work, looking at this map, trying to understand the Opportunity Atlas, why do some places have higher mobility than others? We never felt like we really understood what made some places higher mobility than others. So, I would talk to lots of folks about this and a theme that often came up was the idea that social capital — who you’re connected to, who you’re hanging out with, who might influence your aspirations, or give you a job or an internship — maybe that’s important. It resonated with me from my own personal experience where I could point to folks who I thought really were critical influences and connections who shaped my own life. And so, naturally, in the current era, one starts thinking about social networks and online social network data. So, we set about to study in collaboration with Facebook in a project supported by Mark Zuckerberg and the Facebook data-science team, to use anonymized data from their platform to analyze how people were interacting with each other. For example, the extent of which low-income and high-income people were friends with each other, the extent of which a community is tight knit. Is everyone friends with everyone, or the community’s fragmented into different cliques? So, we took a bunch of these measures of social capital that sociologists had thought about for nearly a hundred years and tried to measure them systematically zip code by zip code, high school by high school, college by college.

LEVITT: And just to be clear, you don’t have any actual income data on these people from Facebook? So, you make some assumptions.

CHETTY: Exactly right. In the Facebook data, we don’t directly observe people’s incomes. Think about things like your residential zip code or the quality of the phone model that you’re using. If you have the latest iPhone versus a much older model, that’s probably correlated with your income. What college you went to. Various other pieces of information like that that we could glean from people’s public profiles on Facebook. So, we’ve measured folks’ incomes and various other characteristics, and we construct these different measures of social capital. And we make all of these measures publicly available in what we call the Social Capital Atlas, which folks can, once again, look up. And then we asked, “Okay, are these measures predictive of these differences in economic mobility across neighborhoods that we had documented before?” And it turns out the answer is yes, but in a very specific way. There’s one and only one measure of social capital that’s extremely highly correlated with differences in economic mobility across places. And that is this measure that we call “economic connectedness.” To put it in simple terms, it’s a measure of what fraction of the friends of a low-income person have high income. So, how connected are low- and high-income folks in a given area. If you grow up in a place where there’s more cross-class interaction, you are much more likely to rise up in the income distribution as an adult.

LEVITT: So, you’re saying something deeper than, “Well, if I’m a relatively low-income kid who has a bunch of rich friends, then good things tend to happen to me.” You’re saying, systematically, if you grow up in a neighborhood where that happens to a lot of kids, that’s good. And if you grow up in a neighborhood that almost never happens, that’s bad.

CHETTY: And we’re making that statement at the community level, rather than the individual level. So, we’re deliberately saying, “Let’s look across places where we systematically see kids doing well. What’s different about those places?” Well, it turns out they tend to be places where low- and high-income people on Facebook tend to be friends with each other. The second key thing to point out is we make a causal claim by looking at people who move across areas at different ages — we show that if you move to a more connected place at a younger age, you yourself are more likely to earn more as an adult. And the impacts are large. We estimate that if you grow up in a well-connected place, you may earn 20-percent more as an adult on average, than if you grow up in a place that is relatively disconnected.

LEVITT: This is if you’re relatively poor as a kid, right?

CHETTY: Kids growing up in low-income families. Yes.

LEVITT: And that number sounds so big that it makes me think there are very few other variables we’ve been able to identify that would have that kind of impact. Is that true?

CHETTY: So, that 20 percent number, it’s not solely coming through only the effect of social capital. So, part of what these connections might be doing is changing other downstream things that in turn have an effect. So, simple example, you might be more likely to go to college if you hang out with a bunch of families where their kids tend to go to college. Now, part of the reason you’re getting the earnings gain is not just from hanging out with those kids, it’s because you yourself went to college and going to college tends to have a positive impact on your earnings. This might be an upstream factor that then has an influence on lots of other things that end up, contributing to a higher level of earnings, which helps I think rationalize a little bit why you end up getting that big number. The second thing I would add is — and this is what was most surprising and striking to me — this variable is very highly predictive even when you control for many other factors that we and other researchers have looked at over the years as potentially strong predictors of economic mobility. So, for instance, neighborhoods with higher levels of poverty tend to have lower levels of economic mobility. And other example is neighborhoods that are more racially segregated, that are predominantly Black, for example, tend to have lower levels of economic mobility. So, it turns out, all of these relationships are completely explained by this new social-capital economic connectedness variable. In particular, once I account for the level of economic connectedness in a place, there’s basically no relationship between rates of poverty, levels of racial segregation, levels of inequality and so forth, and economic mobility. What’s going on is that those neighborhoods that are more racially segregated or are poorer, they are more socially disconnected as well.

LEVITT: What do you do about it? This seems like a hard one to pull policy levers on.

CHETTY: Yeah. Traditionally, at least economists have focused on things like how do you give people money? Or how do you change maybe the quality of schools, or how do you change incentives? And while that itself is challenging, changing kind of the social atmosphere and these sorts of connections is an order of magnitude harder. And so, in the second paper that we published in Nature, we became curious about, is this something that’s actually manipulable? The first thing we did is just try to lay out conceptually, what is it that determines the level of interaction across class lines? And we isolated two different factors. The first is what we call exposure. Just the simple idea that if low- and high-income folks go to different schools, live in different neighborhoods, attend different colleges, if they never meet each other, then they can’t be friends with each other. So, a lack of exposure can drive social disconnection. But there’s another factor, and that’s what we call “friending bias.” That’s the idea that you and I might go to the same school. We might actually be in the same room, but if we’re from different backgrounds, we might still not interact with each other. If you just think about going to, say, a big party, you probably have the tendency to gravitate towards people you’re familiar with, and you can see how that leads to this sort of separation. And so, what we show in the second paper is that the social disconnection between low- and high-income folks in America is half explained by this exposure-segregation phenomenon. But the other half is explained by friending bias. And the reason that’s so important, Steve, from a policy perspective, is that it suggests that one meaningful approach to reducing social disconnection can be to create more mixed income neighborhoods, to integrate schools, to create more socioeconomic diversity on college campuses and so forth. But that’s not enough. It’s also important to think about how we tackle the friending-bias issue. How do we get people to actually interact even conditional on being in the same institution?

LEVITT: And do you have some ideas for that one? Having seven kids, that seems like a hard one. My kids resent it deeply when I try to set them up with friends.

CHETTY: Yeah, so that is a harder one. One thing we find is that this friending bias is not just determined by your preferences. It also seems determined by structural and policy choices that we might be able to change. First, the amount of friending bias that people exhibit varies greatly across the settings in which they make friends. Friendships that originate in religious institutions, for example, are much more likely to cut across class lines and exhibit less friending bias than friendships made in neighborhoods or in schools. So, people behave very differently in terms of the friendships they make in religious institutions or in recreational groups. So, what that suggests is it’s not just about the person. Apparently, if you take that one person and put them in different settings, they start to make very different types of friends. Another piece of evidence that points in the same direction — in some schools, there’s a lot of friending bias. In other schools, there’s much less friending bias. I’ll give you a concrete example. In Chicago, the big public high school around Northwestern University, Evanston Township High School, is an enormous, diverse school that we see in our data, and has been documented, historically, exhibits a tremendous amount of friending bias. There’s a lot of splitting across class and racial lines in that school. But there are many other schools in Chicago that are pretty diverse that exhibit much less friending bias. So, what’s different about those schools? One pattern we find is they tend to be smaller. In small groups, people exhibit much less friending bias. They tend to break apart less than in large groups. Again, let me come back to the party analogy. If you go to a dinner party with 10 people, you probably will talk to everybody by the end of the evening. If there are 500 people, you’ll probably gravitate towards the people you knew. And the same kind of phenomenon seems to apply in schools and in other settings. So, all this is to say, it seems to be changeable in the same way that we spend a lot of time trying to tackle segregation in America. And so, I think that’s another area worth focusing on and studying in the years to come.

LEVITT: You’ve accomplished almost everything imaginable in academics. You’ve won or will win just about every award that’s offered. Do you ever think about trying something different?

CHETTY: I think my aspiration, Steve, and my hope is to try to have an impact with this work, where we will actually see a change in some of these outcomes and go beyond the journal publications and the recognition and academia, which I’m, of course, very grateful for, but really use that as a launching pad to have an impact through research on changing people’s lives in a meaningful way, in a scalable way. I personally don’t feel like I’m there yet and I think there’s a lot more to be done in society. For now, for me, the path to that is doing more academic research, and we’ll see where it takes me down the road.

A few months back, I had psychologist Dan Gilbert on the podcast, and we bonded over the fact that both he and I have gotten tired of academic research. Dan used the words “I’m full” to capture the idea that he doesn’t feel the need or desire to strive for more. When Raj Chetty, just now at the end of our conversation, said he still yearns to do more in academics, my initial reaction was to want to protest, to challenge him: “Raj, there must be other things that are more interesting to you than yet one more academic paper.” But I’m glad I caught myself. I think it’s a natural human trait to want others to view the world the way I do. For them to want the same things I do, to reassure me that I’m okay. But why should Raj do anything different than what he’s doing? He’s the best in the world right now. How fantastic is it that he still loves what he does after all these years? Honestly, I’m a little jealous. I wish I still had his passion for research, but I just don’t. The silver lining, though, is that it frees up my time to do other things I am passionate about. Like this podcast. And in two weeks we’ll be back with a brand new episode where I’ll be talking with the legendary documentary filmmaker Ken Burns. His newest series explores the United States role before, during, and in the aftermath of the Holocaust.

KEN BURNS: Here’s the problem: The antisemitism is always there. The racism, it’s always there. The nativist and xenophobic tendencies, it’s there. Right now, like in other periods, it’s been given more permission — the genie out of the bottle, the ills out of Pandora’s box, whatever, you know, metaphor, you want to make. Right now, it matters because so much of the playbook resembles the early days of the previous playbook. We got to do a hell of a better job.

*          *          *

People I (Mostly) Admire is part of the Freakonomics Radio Network, which also includes Freakonomics Radio, No Stupid Questions, and Freakonomics M.D. All our shows are produced by Stitcher and Renbud Radio. Morgan Levey is our producer and Jasmin Klinger is our engineer. Our staff also includes Neal Carruth, Gabriel Roth, Greg Rippin, Alina Kulman, Rebecca Lee Douglas, Zack Lapinski, Julie Kanfer, Eleanor Osborne, Jeremy Johnston, Ryan Kelley, Emma Tyrrell, Lyric Bowditch, Jacob Clemente, and Stephen Dubner. Our theme music was composed by Luis Guerra. To listen ad-free, subscribe to Stitcher Premium. We can be reached at pima@freakonomics.com, that’s P-I-M-A@freakonomics.com. Thanks for listening.

LEVITT: Now that implies your projects sometimes fail. I got the impression your projects always worked. I mean, you have, whatever, seven or 10 projects around inequality and mobility, and every one of them says the exact same thing. I mean, how lucky is that?

CHETTY: Well, you haven’t seen the the hypotheses that haven’t worked, I guess.