Intro. [Recording date: August 8, 2022.]
Russ Roberts: Today is August 8th, 2022. My guest is economist Raj Chetty of Harvard University. He is the Director of Opportunity Insights, which uses big data to study the science of economic opportunity. At the youthful age of 33, he was awarded the John Bates Clark medal, which honors the best economist under the age of 40. And my joke used to be: it was for the best 39-year-old economist. So, you beat that by a few years.
He has authored numerous influential works on inequality and economic mobility. And, today we're going to focus on a set of recent papers with Johannes Stroebel, and Theresa Kuchler of New York University, Matthew Jackson of Stanford, and the Santa Fe Institute, published in Nature. The title is "Social Capital: Measurement and associations with economic mobility," parts one and two. Raj, welcome to EconTalk.
Raj Chetty: Thanks for having me, Russ.
Russ Roberts: These papers are about what you call social connectedness--and some other factors, but that's the main focus. And, the claim of the first paper, which of the two parts, which we'll focus on is that poor people with rich friends do better financially in their ability to rise than poor people who are only connected to other poor people. Is that a fair summary?
Raj Chetty: I think that's a fair rough summary Russ, if you will, but I would want to caveat that a little bit in terms of precisely what we're able to say with the data we have, which is just to put a little bit more flesh on this for those who may not be familiar or haven't seen the papers yet.
So, we measure various notions of social capital: the extent to which people are interacting with different types of people; the extent to which communities are tight knit; the extent to which people are participating in civic activities like volunteering; and so forth. And, what we document specifically is, if you look at a measure of what we term 'economic connectedness'--what folks like Bob Putnam would call 'class bridging social capital', the extent to which basically low-income and high-income people in a given area are connected with each other--that is a very strong predictor in a correlational sense of differences in economic mobility across zip codes or across neighborhoods in the United States.
So, in particular what we show is, if you grow up in a community where low-and high-income people are interacting more, you are more likely to rise up in the income distribution. That's the fact we're able to establish in the data.
The key caveat, Russ, if you will, that I would emphasize is: it's not an individual-level statement that we're able to establish. Though, the way you phrased it was, if I'm a given person and I go and make more high-income friends, my prospects are going to change. That may well be true. We haven't directly shown that in this study. This study is about community-level relationships, and I'm happy to talk more about exactly what we find there.
Russ Roberts: And, in a minute we'll talk about how you could possibly measure that, which of course is quite an achievement, and it's quite an extraordinary data set and empirical project that this is about.
But so, the caveat you're emphasizing is that neighborhoods that have more economic connectedness tend to have more successful experiences down the road than neighborhoods that don't.
Raj Chetty: That's exactly right. That's exactly right. And, what we're able to show, specifically, is building on some earlier work. If we look at kids who move at younger ages to such neighborhoods, they have better outcomes in the long run than kids who move to those same neighborhoods at later ages. And, under a set of identification assumptions--a set of assumptions that we think make our estimates of these neighborhood effects as good as causal--the way you can interpret this is: If I were to take a given child and put that child at a young age in a connected community, I would see better outcomes for that given child than if I were to put that child in a different community.
Now, that--again, it's a leap to then say from that statement, now if I come and make more high-income friends for a given person, that's going to change their outcomes. And, why do I say there's a leap one has to make there? It's because there could be many other things that are different about the connected communities, right? They do have more cross-class social interaction, but that might also be associated with different institutions, other changes in behavior that are then ultimately affecting kids' outcomes. We investigate that a bit and I can talk about that, but I think that distinction is important scientifically.
Russ Roberts: Well said. You do the best you can: obviously there are a lot of unobserved things about neighborhoods you inevitably don't have data on. You might have something on, say, the quality of schools if we could measure such a thing, but that would be an example of something that would be potentially a confounding factor.
Russ Roberts: Now, I think most people would find this intuitive, this idea, and there've been a lot of movies made about it. The poor person who falls in socially with the rich person and who connects them, and then they get all kinds of opportunities they wouldn't have had. So I don't think it's surprising that interacting with successful people is good for a person who comes from, say, a poor home.
What stands out for me in this work, and I think will be the test of it in terms of its real impact, is the size, the magnitude of this economic connectedness. Give us a feel for that.
Raj Chetty: So, I totally agree with the high-level assessment that your friends matter. I think most of us can think of, from introspection of examples where people we know have influenced us, have shown us a pathway, have shaped our aspirations, have directly connected us to jobs. There are lots of mechanisms through which you might think this type of connectedness matters.
Before I talk about magnitudes and how we thought about fleshing out magnitudes, let me just point out that in the prior work that talked about social capital and economic mobility, there are many theories as going back a hundred years in sociology and economics on many different types of social capital that matter.
What we're finding is that it's this one particular type--connection to people from different socioeconomic backgrounds--as opposed to being in a more tight-knit community where everyone's friends with everyone or in a community where people trust each other more or volunteer more. When we think about magnitudes, the first thing to notice, it's this one particular type of social capital that seems to matter.
And then, I think the next step in understanding magnitudes relates to what you were pointing out earlier about potential confounding factors. If we want to isolate the relevance of this factor, it's very important to think about: What are other factors that might be correlated with economic connectedness that could actually be responsible, perhaps, for the entire magnitude of the association we find?
And so, the way we approach that is, as I'm sure, Russ, this is a pretty developed literature at this point where lots of people have been trying to study what explains differences across countries, across areas in terms of levels of economic mobility. And so, there have been numerous papers that have identified various strong predictors of differences in economic mobility across areas.
Let me pick one of them, poverty rates. So, it's been well established that if you grow up in an area with more concentrated poverty, you have a lower chance of rising up in the income distribution. So, when we first documented this correlation with economic connectedness, the first question we had as well: we see in the data that if you're living in a less economically-connected neighborhood where you're interacting less with high-income folks, it also tends to be typically a poorer neighborhood.
That makes sense, right? Because you tend to be friends with the people around you. So, if you live in a very low-income area, you tend to have very few high-income friends.
To give you a concrete example, take a place like central Los Angeles [LA]. Some of the lowest-income neighborhoods in LA, those are also very disconnected places in our data. And so, you might ask, well, is it because of the disconnection that people have lower chances of rising up? Or is it just because these are really poor neighborhoods where the schools don't have a lot of resources, there are high rates of crime, there are numerous other things that might matter?
So, one simple thing we can do given that there are a number of clearly-identified observable factors like this, is ask: When we do a horse race--when we control with good data on rates of poverty, on things like measures of inequality, segregation, racial composition, and so forth--and look at the relevance of economic connectedness, conditional on those factors, I think what was most striking to me is not only does this economic connectedness variable continue to matter quite a bit, the univariate correlation is something like 0.7. And, even after you control for all of these factors, that correlation remains very similar. But, moreover, Russ, it actually explains a lot of the other relationships that people have found.
In particular, we see that once you control for economic connectedness, poverty rates are now much more weakly associated with economic mobility. So, that is to say, poorer neighborhoods seem to have poorer prospects for upward mobility insofar as they have less interaction across class lines. If you look at places that have very different levels of income but similar levels of economic connectedness, they don't have very different levels of economic mobility.
But, if you go in the other axis, take a bunch of places that have similar levels of income, but in some the low-income folks are interacting much more with high-income folks and in others they're more disconnected, there's a very strong association with economic mobility.
Several analyses like that, where we look at the same thing with respect to racial segregation, with respect to inequality that people have identified in the past as potentially being an important factor, really isolates connectedness as being important.
So, finally now to come to your question about magnitude: so what do we conclude in terms of the size of this association? We show that--here's one way to think about it--if low-income kids were to grow up in neighborhoods where the average high-income kid typically grows up, they have that level of economic connectedness, right? So, if you close the gap in connectedness between low and high-income kids by moving them to comparable neighborhoods, the incomes of low-income kids would increase by about 20%. Okay. So, 20%, that's a fairly sizable change, right? We think about educational interventions, we think about other types of interventions. It's not going to fundamentally change everything--no one factor I think determines everything--but it seems important.
Russ Roberts: One way to think about that, I think for listeners at home, is this: If you're a poor person and you could live in that high-income area and have lots of economic connections, social connections to the people around you, even though you are going to the same horrible school that you were in before, the implication is that those social and economic connections could overcome the poor human capital that you would be acquiring, that traditionally has been the way economists have often looked at your ability to thrive economically.
Raj Chetty: I agree with that, Russ; I would, again, just caveat that a little bit.
Russ Roberts: Go ahead.
Raj Chetty: 'Overcome,' I think, may be stronger than I'd want to say, because my view is schools are extremely important; human capital is extremely important. You know, 20%, let's put numbers on that. Let's say you have an income of $30,000 a year. $30 becomes $36. So, why do I pick $30,000? That's like a typical income for kids growing up in low-income families. $30,000, 20% gain puts you at $36,000, maybe $40,000. Right? So, you're doing better and I think that would be important progress in terms of economic mobility in the United States.
Does it by itself give you kind of the silver bullet? No. I think improving schools and other things is also going to have quite a significant effect.
Russ Roberts: No, but I only mention that because I think it's hard to imagine an educational intervention that has a 20% payoff, at least that we've measured successfully.
Raj Chetty: I would argue, if you look at estimates of the rates of return to additional years of education, people would put that somewhere between seven and 10% per year now. So, if you think of this as going to college for two or three years, if we can really do that on scale--which is not a minor thing--that would have a significant impact as well.
One more point here as we talk about this: We should note that some of the impacts, some of this 20%, it's not just coming in isolation. Actually the mechanism for that may be that kids are more likely to attend college when they grew up in a community where they see lots of other kids attending college, right? So it's not literally the friendships in and of themselves. It's the friendships' changing your behavior, which then change the educational choices you make, which then change whether you go to college, the type of job you get. So, there's a lot of downstream stuff that's contributing to this.
Russ Roberts: It's not just that your neighbor offers you an internship in a place you otherwise would have no access to.
Russ Roberts: Now, let's talk a little bit about how you try to establish this claim. The economic connectedness measure you're using comes from Facebook data, which is rather extraordinary. You have an enormous amount of data from Facebook that you've acquired in a privacy-protected way. Talk about how you're using that. And, if you can--we can get to it in two steps--but, how you can use that information from Facebook to look to prospects for children in poor families, right? It's not just what's happening to them today. It's you're trying to forecast where they're going to be or where they were. So, talk about how you do that, because that's kind of a magic trick.
Raj Chetty: Let's take that in two steps, as you said.
First, what are we doing with the Facebook data? Just to say a little bit about how I got into thinking about this a few years ago. So, as we were doing this work on economic mobility, seeing it varying across communities, across different types of people and so on, as I said earlier, we were investigating various hypotheses for what might drive that variation. And, a number of scholars, sociologists folks like my colleague, Bob Putnam, suggested social capital seems like it could be quite important. And, that seemed intuitive to me as well. But, being a social scientist who is interested in empiricism, interested in quantifying things, measuring things, and seeing how we might manipulate them, I started to think: Well, how can we really measure social capital systematically?
And, naturally in this day and age, you think of social networks as providing potentially a large amount of data on how people are interacting.
And so, we began a collaboration with Theresa Kuchler and Johannes Stroebel, who have done some of the most innovative work in the past using Facebook data on other topics, and Matt Jackson at Stanford, who has really pioneered network theory in economics; and approached Meta--the owner of the Facebook platform--and Mark Zuckerberg to think about whether we could use their data to study these issues systematically.
And so, what we did is took data on everyone between ages 25 and 44 in the United States on Facebook. So, why that age range? As I'm sure you're familiar, Russ, in older ages and younger ages usage of the platform falls. In that sweet spot, 85% of people in the U.S. population are on the Facebook platform. So we feel that, while it's not 100% perfect representation, it's pretty good as a starting point.
And so, then we do a series of benchmarking exercises to basically convince ourselves that the Facebook data are providing a reasonable representation of people's incomes, of their friendships across class lines. Anytime I start with these administrative big data sources, I like to start with the step of comparing to the traditional survey data: Are we patterns that line up? And, in this case, there's something called the Add Health Survey that tracks about 10,- or 15,000 kids and looks at their friendships. And, we show that at the national level, the Facebook data delivers results that look almost identical to the Add Health data, which is reassuring to us.
Now, what the Facebook data allows to do then is, because of the massive sample size--72 million people, 21 billion friendships between them--we can then drill down to the zip code level, to each high school, to each college, and construct estimates of the extent of cross-class interaction--the extent of what we call 'cohesiveness.' So, think of this as a simple measure of whether everyone is friends with everyone or whether a society is broken into separate cliques: it's a different way to think about social capital.
Or another example: measuring volunteering rates in every zip code. Traditionally economists and social scientists would use things like the General Social Survey [GSS] to measure a statistic like volunteering rates, but you really only have the sample size to do that at best at the state level. Here again, we show at the state level, the measures we construct with the Facebook data based on participation in groups--volunteering groups--align closely with the GSS at the state level. But then, you can zoom in now to the zip code level and look at patterns in a more granular way.
So, that's the construction of the data. Then, separately, drawing up--
Russ Roberts: Could we stop there? Let's just clarify a few things on that. I know Facebook uses the word 'friend,' which is very similar to the English word 'friend' for somebody you have a relationship with. But of course--mmmgrrrmmm [vocalized sound effect of skepticism]--there's a little--
Raj Chetty: [inaudible 00:18:36]--
Russ Roberts: Yeah. How connected am I to my Facebook friends? So, I think you gave me a little phrase there in the middle of there about the Add Health data. So, talk about why--I think most people would say, 'I'm not sure that's a great measure.' And, I also don't understand why it's related to my neighborhood. I mean, almost by definition, the essence of Facebook is that I'm friends with people outside my neighborhood. So, talk about that.
Raj Chetty: Yeah. Thanks for pushing on that, Russ, and certainly something we thought about quite a bit.
So, simplest answer: So, the worry is that the typical person has something like 500 friends on Facebook. Is that really what we mean by a friend in terms of the types of things we're interested in, somebody who might influence the type of career you choose, whether you go to college, and so on?
So, I was quite concerned about that. And so, what we did early on was a check where we subset to your 10 closest friends or your five closest friends. So, the folks you're interacting with the most is measured in various ways. Are you posting on their Facebook wall? Are you exchanging messages? There are various different ways that one can measure closeness of interaction.
And so, we did a series of analyses: We basically replicated everything we did, looking at your closest friends. Think of these as people who you really are interacting with in real life; and, connected to your neighborhood question, often you're physically proximate to. And, what we found is you get almost identical results when you use your closest friends, as opposed to your full network of friends.
And so, I was puzzled by that, because I think, like you, I had the intuition that it would really be the close friends who would matter, not people you've met once and happen to connect with on Facebook.
But, I think what's going on is the following: It's that your broader social net is representative of who your close friends are. So, if I'm just trying to measure, on average, how many high-income friends you have, when I take your full network of 400 friends, it tends to be very representative of how many high-income friends you have among the ones who really matter, like the 10 closest friends. Because those two things are really correlated, it turns out that using the full Facebook network actually gives you a pretty good proxy for what we're interested in. But, that's an empirical result ex ante that may or may not have been the case.
Russ Roberts: And I assume you are open-minded--or agnostic. That's all I meant by that. You're agnostic about how these connections actually manifest themselves in people's lives. So, I don't understand your conclusion to be: People on Facebook need richer friends--poor people on Facebook. But, you're using that as a proxy for a much wider, amorphous, hard-to-maybe-quantify network.
Raj Chetty: That's exactly right. That's exactly right. We view the Facebook data largely as a proxy for the types of offline interactions that people are having. We're not, per se, interested or able to really isolate the impacts of online interactions themselves. And I think that's very important to point out. Our interest is not in: Create more online interactions of a particular type and that's going to lead to something. This is a way to get data on who people are interacting with systematically.
Russ Roberts: But, to sort of come full circle on this piece--and then we'll go on to the next piece about measuring mobility--what you're talking about here is that there are neighborhoods or communities ideally--although of course, you're stuck with zip codes and other geographic measures, or schools--but, there are certain communities that have wider connections outside of the socioeconomic status of that community. And, there are others that don't. And, those communities that have those wider connections are going to do better; and we're using the Facebook social network, meaning internet network, as a way of proxying for that wider range of interactions. Fair? Good summary?
Raj Chetty: That's a good summary, Russ.
And I would want to emphasize one thing, which I think you picked up on: the connections are not just within your immediate neighborhood, right? So, lots of the connections that people are forming are across the United States; and importantly in the Facebook data, while we're aggregating the data to the zip code level, and then we're going to correlate with zip code level measures of economic mobility--a lot of the connections that people are making when they live in a given zip code are of course to people outside their zip code.
Especially, it turns out, it's interestingly for higher income folks. And, this is intuitive. I think your friends tend to be very spread out.
For lower income folks, neighborhoods, physical proximity turns out to be very important. You have a lot more of your friends close by to where you live.
And, that is actually part of the reason that very low-income communities tend to be more economically disconnected because low-income folks are just friends with those around them.
But, interestingly there are exceptions to that. We find for instance, parts of West Virginia--a relatively poor state, as you know--the northern half of West Virginia has a surprisingly high level of economic connectedness. And, a lot of those friendships somehow are outside the immediate areas where people are living. So, it's an interesting thing to investigate and further work how is that happening.
Russ Roberts: I would just reference for listeners, episodes we did with Chris Arnade and other examples, where we've talked about--I mean, this goes back, I don't know, forever, on this program where we talk--the solution to poverty is luggage: is to let people get out of town and find, go to places where there's more opportunity. And, with Chris and others, we've talked about the fact that some people either can't or don't want to go out of town. They want to maintain the closeness to their--Megan McArdle and I talked about this in a recent episode. They want to maintain family connections and networks that are very important to them.
Whereas there are other people--and Chris calls them the front-row kids--who are just going to go out and find the best opportunity. They're going to go to the best college they can go to. They're going to take the best first job they can go to. They don't care if they live near home; they don't care if they stay close to their parents and visit them often, and if their kids interact with them, etc., etc. And, that's kind of--that's part of what you are measuring.
Raj Chetty: I think that's exactly right. And, I think part of this addressed[?] more broadly is an effort to try to understand: short of people moving to a different place, which I agree, a) is not a scalable solution, and b) many people don't want to do, what is it that's making some communities offer better opportunities than others? And, can we learn something from that in order to in essence bring opportunity to people where they are, rather than bringing people to opportunity? We're hoping this is a step in figuring that out.
Russ Roberts: It reminds me of the person who grows up in a very intellectually impoverished environment, but something's different. They either get turned onto books early, or their parents are unusual. And, their main goal in life is to get out of that community and to find people more like them. They start finding them as soon as they can. They might be represented in Facebook and elsewhere online, now that that's a possibility, but before they'd just go to New York or wherever was the large city that drew them. Some of that is, of course, cultural--in the air and the water, hard to understand. And I salute you for trying to get at that.
Russ Roberts: So now go to the next part. Talk about: How can I possibly, once I know this about different people--I have serious measures of their connectedness--how am I going to use that to get a measure of their economic mobility, given that you don't have their income? Which is a challenge.
Raj Chetty: Exactly right. In the Facebook data, we don't measure mobility directly because if you think about, 'What is mobility?' What I'd like to do is take a set of kids who grew up in low-income families, they're now working as adults. You need to be something like age 30 to have a reliable measure of your income. If you look at people in their 20s, some of them are in college. It's too early, basically. If I want to look at a 30-year-old now, they have to be born obviously back in 1990, or even before that to get a good sample size. And, you're not going to be able to measure that within the Facebook data directly.
So, what we do is lean on our earlier work, where, completely separate from Facebook, we use information from anonymized tax returns to measure economic mobility by zip code, by census tract, by county and various levels of aggregation in the United States. Essentially taking all kids born in the early 1980s in America, linking them to their parents, using information from tax returns and then measuring in each area--there's some statistical methods in the background, which we can come back to if there's interest.
But, in a nutshell, the simple summary is: Take kids growing up in low-income families--say, the 25th percentile of the income distribution--ask where they themselves end up as adults about 30 years later; and estimate that for every zip code. And, then we're going to take those zip code measures and compare them to the Facebook measures of economic connectedness and ask how correlated are these two things.
Russ Roberts: It's a fantastically ambitious enterprise. I'm very skeptical about it, so I'm going to give you a chance to convince me. One of the challenges for me is that this idea that neighborhood is destiny--that there's something inherent in the neighborhood itself, in some sense almost independent of the people who happen to be there and their characteristics. And, especially given that we can't observe all the things we care about. I find it hard to understand how you can take that leap and say, 'Well, if you grew up in one of these rocket pad neighborhoods where you can be launched into economic success versus one of these staid, backwater, swampy places,' and that as if no one ever moves, as if they don't change, as if they can't choose where to live--it seems to be--it's a bold stretch, isn't it?
Raj Chetty: I appreciate you raising that Russ, because it gives me an opportunity to clarify what I think are some misconceptions about the work and maybe over-interpretation of what I think we're able to say.
So, first, just the phrase 'zip code is destiny,' which for some reason people like, I actually disagree with.
Russ Roberts: I thought I made it up, but okay.
Raj Chetty: No, I think it has appeared in various headlines and so forth in the past. I think it's maybe a convenient way to summarize what people think they're seeing in the data.
But, point number one, destiny seems to suggest that somehow your fate is fully determined by the zip code or the neighborhood in which you grow up, which is absolutely not true in the data. Right? What we're seeing is there's significant variation across neighborhoods, but there's a lot of variation within neighborhoods. Right? There are plenty of kids who in the high upward mobility places on average, who end up not doing so well. And, there are plenty of kids in the places which we show in the data have low upward mobility on average, do end up doing well.
So, to use a little bit of statistical jargon, the R-squared--or the explanatory power--of neighborhoods is nowhere close to 100%. It's probably more like 5% or 10%, meaning lots of other things matter well beyond your neighborhood--family, education, the types of things you were talking about, that one kid gets inspired in a particular way. Numerous things like that.
And so, the first very simple point I'd want to make is: we're definitely not saying neighborhoods determine your destiny. We're saying that's one level at which things seem to vary in a significant way.
Now, second point, sometimes people interpret the neighborhood focus as saying, 'You think there's a physical characteristic, some fixed thing about this neighborhood that's launching people on a better path than other places,' but that clearly doesn't make sense.
I don't think it's, like, about the quality of the soil in a place or agricultural productivity or something in this day and age. It's about who's living there and the set of social institutions. What varies at the neighborhood level are exactly things that people have emphasized in the past. Things like the quality of schools, or rates of crime, or culture, family structure, all of these other factors. And, for me, what's interesting about thinking about neighborhoods is, it's sort of a package where a whole set of things shift at the same time.
So, if you're interested in the thesis that environment might matter for human development, I think neighborhoods are just about as good an aggregation as anything else. We can look across schools--and there's been good research done, say, in charter schools versus public schools, etc. But, often a lot of these factors vary at a geographic level, right? You go to a different school if you live in a different neighborhood, typically.
And so, one thing I'd want to emphasize to your listeners is, sometimes people interpret the neighborhood findings as saying they're sort of at the expense of other factors--or independent of other factors--like parenting habits or family structure or schools and things like that. And, what I would argue is actually those things are varying at the neighborhood level and that's why we're seeing these neighborhood effects.
And importantly, what that means is these things are not necessarily fully stable over time.
There's actually some very nice work that's been done showing, for instance, a nice paper in the AER [American Economic Review] by one of our former students at Harvard, Ellora Derenoncourt, who is now at Princeton, showing that the causal effects of neighborhoods on upward mobility, in particular for African Americans, changed substantially in the context of the Great Migration. As a result of an influx of African American immigrants from the South, there was a disinvestment in public goods, schools, and so forth, changes in segregation patterns which then changed neighborhood effects.
So, all of that I think fits with the type of reasoning you had. And I just think we should think of neighborhoods as a useful source of variation to study.
And perhaps, given that there is variation across neighborhoods, for folks who want to move to higher-opportunity areas, that could provide a pathway out of poverty, at least for some people.
Russ Roberts: You know, when you think about culture, you know, people make generalizations, say, about American culture: It's more risk taking than certain other cultures. There's more emphasis on competition or individual achievement, or so on. And, that this changes how Americans do in the world. But of course, there's enormous variation. Part of the value, I think, of your work is reminding people there's enormous variation within the country, and even within neighborhoods, and so on.
The other thing to think about, though, that I would like to hear your reaction to is: reverse causation or selectivity issues--selection bias issues. So, again, coming back to these concerns about, say, areas of--I would call them areas of despair, particularly Rust Belt places where manufacturing, which used to be a very healthy way for people to do well when they grew up without a college degree. Those places disappeared. And, a lot of the kids who grew up in those areas left because opportunity was scarce there.
The ones who were left behind, either are the ones who are less ambitious, or who care more about their family, or a whole bunch of complicated factors there.
But, the neighborhood is constantly changing in response to these external changes in both the return to, say, certain types of education, manufacturing versus knowledge-based jobs, and so on.
Raj Chetty: Yeah, absolutely. So, you know, if you take the kind of explanation I was giving earlier--that it's not a fixed characteristic of the neighborhood, it's really about the people who live there and what emerges, and insofar as there's a causal effect, it's coming from that social composition. If you take the most recent work on connectedness or even the quality of schools, it's going to be determined by maybe your peers and the types of choices that people, investments people are making and so on. Naturally, in that view of the world, you have to worry about sorting--or selection versus causal effects.
So, are we seeing that--to go back to my earlier example--certain parts of LA [Los Angeles], we see higher levels of upward mobility than other parts. Is it actually, if we take a given child and put that child in that different neighborhood, we're going to see different outcomes for that different child, the causal effect argument?
Or is it just that, as you were saying earlier, it's a different set of folks living there? They have a different level of ambitions, a different set of experiences. And maybe that's why we're seeing differences in economic mobility across places.
So, there's been an extensive effort in the social sciences, which our team has contributed to in recent years, on trying to disentangle these two different explanations. I basically see this as a test of whether environment is important or is it really just about different types of people living in different places and we need to think at the person-level of what's driving those differences.
And so, in our work, we've done a series of studies looking at people who move across neighborhoods, with kids of different ages. Essentially--the simplest example I like to give, is: Let's say you've got a family with a seven-year-old and an 11-year-old who moves from Neighborhood A to Neighborhood B. What we're basically doing is comparing the outcomes of the seven-year-old to the 11-year-old in relation to the outcomes of the people who were already living there. So, if you move to a place where we're seeing higher levels of upward mobility, do we see better outcomes for the seven-year-old instead of the 11-year-old?
And, the reason we make that age comparison is because we basically have a dosage view. That's one way to think about it: the model--the theoretical model underlying this--is every extra year of exposure to an environment that might change your aspirations, that changes culture, that gives you access to better schools, etc., might shape your outlook, shape your choices down the road. And so: Are we seeing differences based on the age at which kids move, looking at siblings within families, for instance?
And, the answer is: Yes, you see very clear evidence that if you move to the higher-upward mobility places at a younger age, kids do better.
But, to be clear, that doesn't mean there's no selection. We estimate that something like 60% of the variation in the data across neighborhoods is due to causal effects. And, 40% at the Census tract level is due to selection. Just the fact that different types of people are living in different places.
So, by no means are we saying selection doesn't matter, but we're seeing there is an important causal component. And I can elaborate a bit more if you'd like on why we think that's a pretty robust finding based on subsequent work.
Russ Roberts: But the issue there--and we'll come back and talk about this, I hope, in a little bit. But, the issue there is an ambitious parent--a parent is a better way to say it--a parent who is ambitious on behalf of their child or their children, is going to want to move earlier too, also. Right?--
Russ Roberts: Because they think--they've read your work. They know it's important. They believe in the dosage idea. That's the challenge, of course, is teasing that out.
Russ Roberts: Let's talk about a more general issue, which I hope we have time--we'll come back to the paper you wrote with co-authors on that you called the "Fading American Dream." But, in that paper, which, without going into the details yet--one of the things I love about that paper is that there's an Appendix. And, in that Appendix, you're very open about how sensitive the results are to different assumptions.
And, I've been in the kitchen of econometrics. It's a messy place. There's a lot of freedom in that kitchen to swing the cleaver at different times in different ways and in different amounts. How reliable do you think these results are in terms of their sensitivity to various assumptions? I mean: There's an enormous number of assumptions here, is the challenge.
Raj Chetty: Yeah. Russ, I'd love to talk about that. Let me just quickly make a point about the earlier statement you made about ambitious parents moving earlier, which is of course correct, and something we worry a lot about in the paper and discuss.
I find it useful to take the siblings example for exactly that reason. Right? Because, take your example of the ambitious parent moving earlier, that may well be true, but it's not obvious why that would create a differential effect within their family, looking at the three-year-old, versus the seven-year-old, versus the 10-year-old. Right? So, that is, I think, key to sibling comparisons.
But then one more point on that: There have been a series of subsequent studies--and this is what really gives me the most confidence. Any one study, we're trying to do our best. I mean, we lay out the identification assumptions. Of course, you can question--as you're correctly doing--do we really believe this or not? So it's useful to triangulate by using different approaches.
So, there have been subsequent studies using a series of different methodologies. One example is the famous Moving to Opportunity experiment that gave people housing vouchers to move to higher opportunity areas. And, you find incredibly similar patterns: that kids who moved, this time purely driven by random variation--so we actually know it's a randomized experiment, right? The kids who moved at younger ages to the lower poverty areas have significantly better outcomes than the kids who moved at older ages.
Another example, Eric Chyn has a nice paper in the AER, where he uses as a quasi-experiment the demolition of public housing projects that exogenously forced people to move, right? You didn't have a choice. And again, he shows that the people who moved with kids at younger ages to better neighborhoods, the younger kids did better while the older kids didn't gain as much.
So, there have now been a series of, I would say, eight or 10 replications using different methods, different countries, different data sets that always recover this dosage pattern. So, I feel quite confident that, you know, even if you question the particular identification assumptions we had in our initial paper, my sense is this seems like a pretty robust finding.
Russ Roberts: Say something about immigrants. We recently had Leah Boustan and Ran Abramitzky talking about their work, a similarly ambitious bit of empirical work with a very large sample. And of course, immigrants typically come to poor neighborhoods. Many of them, though, come very well connected to their fellow ethnic group folk. And, they obviously are selected to some extent on ambition, at least, again, on behalf of their children. Have you looked at them separately at all? I don't remember seeing that in the paper.
Raj Chetty: Yeah. So, we've looked at immigrants and actually some of the recent data Ran and Leah are using in their fantastic work. They look at the historical era, as you know, and the modern era. And, the modern era, the data they're using is partly coming from the estimates we've put out publicly. They're able to draw on that.
And, what's really interesting, Russ, about the patterns with immigrants is--I think it connects to two aspects of our conversation. First, on average, immigrants have higher levels of upward mobility than natives, a point that Ran and Leah emphasize, over generations. Even controlling for other things, it shows that many things matter beyond--zip code is not destiny, as I was saying earlier.
But then, they also show--and this to me is another point of validation on these neighborhood results--they show that the immigrants, a large part of the gap in terms of rates of upward mobility can be explained by the fact that immigrants move to neighborhoods that even for natives have higher levels of upward mobility.
So in particular, they show that immigrants' outcomes are very correlated with natives' outcomes across areas, which is another point suggesting that this package of things that's varying across neighborhoods--schools, neighborhoods, schools, culture, family, etc.--they're having some common effect across these groups, which I think is basically one more piece of evidence that all of this seems to matter.
Russ Roberts: I'm curious how much--you know, immigrants tend to move to large cities. Large cities are going to be more dynamic than rural areas on all these dimensions we're talking about. There's a lot more opportunity for choosing what parts of the world you want to connect with--school, neighborhood, social hall, religious community, and so on. To what extent--is there a way you've looked at how much of the results, of your results, are driven by really large effects in certain areas, both at the top and the bottom? Obviously, there's some very poor areas, that don't[?] have so many people in them, almost by definition. But, there's some very poor areas that might be driving on the results as well. How sensitive are the results to particular outliers? Or is that not relevant?
Raj Chetty: Yeah. You know, there's actually a bit more nuance there than one might think. So, I also had the intuition that it would be the big cities that give you a lot of opportunity--opportunities to connect and so forth. But, actually when you look at these statistics, both in the original Opportunity Atlas data on upward mobility and in this new Facebook data on economic connectedness, there are many cases where it's the rural areas where you see higher levels of upward mobility and greater economic connectedness. And, the way I think about that is: example, my wife grew up in a small town in Southern Illinois and her dad was the town doctor. There was one school; everybody went to the same school, and so everyone connected with everyone in that setting.
What you find in the data is, places like rural Iowa, much of North Dakota--you know, some of these smaller towns, they have very high levels of economic connectedness. They also have high levels of economic mobility.
Importantly, the kids when they grow up, many of them are not staying in those small towns when we're measuring their incomes in adulthood, right? They've moved to Chicago, they've moved to New York. So, there's a difference in terms of labor market opportunities versus the type of shaping of aspirations and so on, that might occur in those smaller communities.
Now, to your question about outliers: Yes, there are outliers. My sense is these patterns are fairly linear and hold across the distribution. So, you can kind of throw out the extreme top, the extreme bottom, look in the middle. It looks like basically a steady progression as far as we can tell. It's not just about being in the most connected place or the most disconnected place, etc.
Russ Roberts: You're very--maybe I'm wrong, but I sense you're very loathe to draw strong policy implications from your work. Is that true? Let's talk about this work. Do you see policy implications for it that you want to champion?
Raj Chetty: Well, I like to be, [?] close to the data in terms of what I think we can actually say. I leave it to others to interpret, bring their priors, their views into the mix and also take the data and figure out what the best way forward is. My sense from this work is, what we can say is that growing up in these more-connected communities seems to matter for kids' outcomes. I feel confident about that. I think what that warrants is at least an inquiry into understanding how changing connectedness directly might affect kids' outcomes. Your initial question when we started on suppose I could somehow connect people to more high-income friends, is that a viable pathway to improving people's outcomes--given what I've seen here, I would want that to be a priority in terms of understanding that question, maybe there would be a pilot to test interventions.
We're starting to hear from lots of folks doing interesting work on the ground to create those sorts of connections, to break down barriers to interaction. And I certainly think that warrants further study and analysis.
I think these things are complicated enough that before you jump to, 'This is what we should do in terms of national policy,' you want to go step by step in terms of 'Let's pilot something. We've done things like this in other contexts. We're helping families use housing vouchers to move to different neighborhoods,' and so forth. I think there's a progression there.
Russ Roberts: Let me try a narrative on you, see what you think. I've been a little bit critical of your work, or challenging of it, which of course is part of the deal. But, the part I really like is it forces you to think about things you might not think enough about. I think that's really important.
I haven't thought enough about the fact that there's so much economic segregation in the United States. We know there's racial segregation, but there's a lot of economic segregation. When I think about that, I tend to think about it in terms of housing and how housing prices have made it hard for low-income people to move to higher-opportunities places. If you're in West Virginia, the idea of moving to New York City might be appealing and it happened a lot more in 1950 than it does today.
And so, I'm going to pick a few things that I think make it a lot harder for people to rise. So, one of those is the fact that we make it hard to build new housing in American cities.
Second is really important. Go back, listeners, to the episode with Alain Bertaud on zoning. The fact that it's against the law in many cities to offer a small apartment as an option--there are minimum sizes that make it harder for poor people to be able to get a toehold in a place they might want to live in, even though it would be a very small apartment.
Then we have the way we've organized education in America--it's free, but only if you are in your neighborhood school, which means that poor people are going to tend to live, go to school with other poor people, and rich people tend to go to school with other rich people.
And then finally, there's the way we've made healthcare part of employment in the United States.
And so, all these factors have made--it's a fact, and I'm not sure we totally understand that. I'd be curious on your feeling about it. But, there's a lot less mobility--physical mobility--in the United States. People move less. And, as a result of those factors, which are complicated and multifaceted, there's a lot more economic segregation in the United States.
Forget the financial consequences of that, which is what your work focuses on: you're looking at how people thrive financially as they grow older. But, it--socially, it seems to me, extremely destructive for a country that purports to be a democracy to have people only hanging out with--whether it's on Facebook or in real life--with people like themselves: that their religious institutions are economically segregated, that their school is economically segregated--from the very beginning, by the way, from K [kindergarten] through graduate school, almost, with--there are a few exceptions, obviously. But, it just seems to me that's really bad for a country. And, that one of the lessons of this work is that we ought to have tried to change some of those policies that make it hard for people to intermingle and understand each other.
Raj Chetty: Well, I agree with all of that. That was very well put, Russ. If I can just add a bit of color on a few dimensions.
Russ Roberts: Please.
Raj Chetty: So, first on housing prices and segregation: so, that, I think, is an important factor, but what struck me in these data is you don't actually need to move from West Virginia to New York. Often you can move within your own community to a place where we're seeing much better outcomes for kids. And, when we first started to look at that data, at that finer geographic level, our initial instinct as economists was, 'Well, it must be that people aren't moving to these higher-opportunity places because it's too expensive. The prices are too high in the places that have the good schools, etc.' And, there's some truth to that for sure.
But, it turns out that there's a lot of residual variation. By that I mean lots of places that don't look that much more expensive actually, but where we're seeing in the historical data, kids have much better outcomes. And, along the lines of what I was saying earlier, kids who move to those neighborhoods at earlier ages do better. You can establish that with experiments, etc., etc.
And so, this was puzzling to us. You can think of it as basically places that look like opportunity bargains, so to speak, right? They are places where kids do much better; you can afford to live there; but for some reason you're not living there.
And so, this is what motivated this experiment we did in Seattle, with the Seattle and King County Housing Authorities that we call "Creating Moves to Opportunity," where we basically sought to understand: Why is it that people are not moving to these neighborhoods where we think their kids would do better?
In order to try to figure that out, we tested whether giving people assistance in moving to these high-opportunity places would make a difference in where they choose to live.
So, just as a bit of institutional background, in the United States we spend about 45 billion a year on affordable housing programs, the largest component of which are housing choice vouchers, which go to a couple million families a year. They provide rental assistance to the tune of about $1,500 a month in Seattle. So, quite a substantial amount. The idea is to try to give people stable housing, break the cycle of poverty, and so forth.
But, what we and others noticed is despite receiving those vouchers--which should make a number of neighborhoods more affordable--these families were still living in high-poverty, lower-opportunity areas.
And so, what we sought to do in this experiment was to test: Why is that the case? Is it that these families have a preference to stay in these neighborhoods because it's closer to their jobs, closer to family, and so forth? Or, are there barriers that are preventing them from making those moves? Do they maybe lack information? Are they having trouble navigating things with the landlord? Are they having trouble finding the listings? This is not an easy market to find housing in.
The other thing I'll note is: it's also a place where I think issues of behavioral economics, I'm reminded actually a little a bit about the recent New York Times column you had about Wild Problems. For these families, this is that kind of problem, right? You've not done this before. You've not lived in a totally different neighborhood, etc. Right?
Russ Roberts: Scary.
Raj Chetty: And so, we connect you with a counselor. Think of it as like a nonprofit broker who helps you navigate this process. If you want to move to a high-opportunity place, they'll help you figure that out.
And, in a randomized trial, we show that this dramatically changes where people choose to live: 60% of people in the treatment group end up moving to a high upward mobility place, compared to 15% in the controlled. And so, effectively what you're achieving, Russ, is a little bit of that desegregation that you were calling for, where you're starting to see people living in different communities.
And, what I find encouraging about that result is it suggests that the roots of segregation in America might not be deep-rooted preferences and things that are super-hard to change, but rather something that we were able to change with the program that costs $2,500 per family--not a trivial sum, but relative to the amount we spend on trying to fight poverty and the amount we spend on housing vouchers, not a very large amount of money.
And so, kind of stepping back, my sense is all of the structural factors you described are certainly important. And, I think changes in things like zoning laws can be very valuable to create more supply, basically. You have to have units in these places that have better outcomes. But, I think there are things that we can do in addition to that, where this problem is more tractable than it might seem.
Russ Roberts: I recently moved to Israel about a year ago. It's an example of the kind of intimidating social environment that is alien, right? I've been to Israel many times before I moved here. And, in my mind, I kind of know what Israel is like. But I don't. It's in the Middle East. It's a different world. Norms are different, expectations are different. Wandering into a store and knowing how to interact with the other people in line and the person behind the counter--I don't know the rules.
And so, I think one of the interesting pieces of this social problem or phenomenon--we could debate whether it's a problem--but the social phenomenon of segregation that is so fascinating, is that once it gets entrenched, it's very hard to get people to move. Because if they wander into that new neighborhood, they're thinking, the cars don't look like their cars in their neighborhood. The people don't dress the way they dress. The people interact with their children differently. And, all of a sudden you don't feel comfortable and you want to go back to that old place.
The idea of a counselor or a mentor, which is of course part of what a wider range of social interactions often gives you, it's not just that, quote, "you know rich people": it's that you know people who know about things you don't know about, sometimes, and opportunities, and how to behave. Once we've gotten that level of segregation, it's often going to be a lot trickier to break it down than simply giving people say a lower price of housing, because you've changed the zoning law. You're still not necessarily going to take advantage of it.
Raj Chetty: I think that's exactly right, Russ. I couldn't agree more. I also want to come back to the question you raised earlier on "The Fading American Dream." I'd like to comment on that.
But, before I do that, let me just pick up on what you said, on, I think, the challenges of segregation. Just--to me, to push a little bit further in the policy vein. One thing that I think is quite important is: as economists we tend to focus on incentives and resources--you know, think about the housing voucher example I gave. A lot of the policy debate traditionally is on: How many housing vouchers should we have? And, what should the budget be for this program? And, how big should the voucher be in terms of dollars?
But, I actually think there's a missed opportunity in terms of how you couple those resources with the kind of social-capital, connection type of intervention that can make people harness these resources in a much more effective way.
So, we're seeing this in a number of different domains, where if you look at programs that couple resources with a form of social support--the counselor, somebody who helps you navigate the system--that can have a much, much bigger impact than just saying, 'Here's a check,' or 'Here's a housing voucher. Here's a job training program that gives you a certain set of technical skills and now go on your way and figure things out.' Because, figuring things out is actually a bit more complicated than just having those resources.
Russ Roberts: I think that's one of the insights I think that Geoffrey Canada and Harlem--is it Harlem Children's Zone? Is that the name?
Raj Chetty: Harlem Children's Zone. Exactly.
Russ Roberts: So, he said: Well, look, the schools here aren't very good. So, you'd think it would help to improve the schools. Or, nutrition is not very good, so it would help to offer nutrition. I think his insight to the extent that--I don't know if it scales and you can debate how successful it's been, and so on. But, the part that I love about what he's done is he said: 'We can't just fix one thing. We need a full court press. It has to be a multifaceted effort.' And, as economists, we tend to look at all other things held constant. And, I think part of, one of the lessons from these kind of programs is: You don't want to let other things be held constant. You better try to move on a bunch of different fronts.
Russ Roberts: Let's talk about the American dream paper, "The Fading American Dream." I've been critical of it and I want to let you defend it instead of me trying to defend it. Critical--I try. Go ahead.
Raj Chetty: Let me say a few throws[?]. First of all, I appreciate your reading the Appendix. All of these papers, as you know have lots of appendices because that's the nature of empirical economics, right? You have to make a bunch of choices about specification, about how you're going to define things. And, you can define things in five different ways and they're going to give you different results. So, one thing I try to do--and I think many empirical researchers try to do--is at a minimum be transparent about what the different choices are going to give you. And so, that's what we try to do in that particular appendix.
For those who may not be familiar, what is the claim this paper is making? It's trying to measure a simple statistic: What fraction of kids go on to earn more than their parents did? One notion of what the American dream is about, the idea of doing better than your parents. And so, the idea for this paper was: Well, let's try to measure how that's changing over time. Is it the case that kids are doing better than their parents today, the past, and so forth?
So, at the surface you might think: Well, that's a simple calculation, like, 'Let's just go calculate what fraction of people earn more than their parents.' Turns out to be a lot more complicated than that, because you've got to make a lot of choices. One simple thing you have to figure out is how you deal with the fact that prices change over time. You can't just take the nominal dollars that your parents earned and that you earned and compare the two. You've got to account for the fact that it costs a lot more to buy milk now than it did 40 years ago, etc.
And so, what we do in this paper is start from a baseline where we're making a certain set of choices. We're taking the consumer price index off the shelf that the government puts out. We're comparing total household income of kids to their parents. That is also a choice because the size of families has changed over time; maybe you want to adjust for that in some way. We're comparing all kids to their parents. That's also an important choice, because as you know female labor force participation has changed a significant amount over the past half century. So, do you count women coming into the labor force as part of the picture? Or do you just compare boys to their fathers, which has been a tradition in this literature in the past? Because people don't want to deal with essentially the changes in female labor force participation.
Let me say what we did in our baseline, what we find, and then where I think some of the sensitivity is.
So, we take the conventional consumer price index. We compare total family incomes for kids to their parents, and the headline conclusion that lots of people focus on is back for kids born in the middle of the last century--1940s, 1950s--something like 90% of kids ended up doing better than their parents. If you take the baseline measures, that number looks like 50% for the current generation. So, big change over the past half century or so: the title "Fading American Dream." Right?
Russ Roberts: I just would add, 50% is a particularly depressing number, because it means that 50% do worse than their parents. [inaudible 01:02:10]. It's a very important number. It's a very depressing number.
Raj Chetty: That is a depressing number. You can actually get more depressing than that. So, if you take an approach that many people pushed us to take actually: compare boys to their dads and just look at their own earnings, right? So, you're taking women out of the picture on the female labor first participation thing. Then you get a 40% number for the current era.
But, then you can go in the other direction as well. So, suppose your view is, as many people argue, inflation indices--this may not feel right in the current time where prices are going up a lot in the past few months--but more broadly economists have worried that we might be overstating inflation, mainly because of technological innovation--
Russ Roberts: Quality--
Raj Chetty: You didn't have the iPhone 50 years ago. Now you do. That's worth something. How do we deal with that in prices? So, we do some sensitivity analysis in the paper where we say: Suppose inflation was actually 1% lower than what the CPI [consumer price index] says it is. Where does that end up? Well, that's going to give you--that 50% number is going to become 60%.
And, then suppose you adjust for changes in family size, so rather than comparing boys to their dads, take everyone in the family. But, one thing people sometimes do in the literature is divide by the number of people in the family--like, per capita resources--or by the square root of the number of people in the family. There's a debate about how you should do this. You can end up with a number as high as 70% in that case for the recent era.
So, what is actually robust here? You might have the view, 'Well, this number could be 40%. It could be 70%. What are we supposed to conclude from this?'
What I think is actually robust are two things. One, and part of our purpose in putting all this out is different people focus on different specifications. There are folks like Scott Winship at a AEI [American Enterprise Institute], others, who've done nice work emphasizing different aspects of this. Nobody, I think, disputes the view that rates of absolute upward mobility are lower in the current era than they were for kids born in the 1940s and 1950s. So, maybe the 90% became 70%. Maybe it became 50%. Maybe it became 40%. I don't think anybody thinks it's stayed 90% or went up.
Part of the reason for the relative robustness of that conclusion, like the key--in my mind, the key uncertainty is, how do you think about inflation? You've basically got to compare people living in different eras and translate dollars today to dollars 30, 40 years ago. That inherently is kind of a fraud exercise. You have to make a bunch of assumptions there.
But, here's the key thing, Russ: Unless you think inflation changed--the extent to which we're mismeasuring inflation changed--a lot over the past 60 years, that comparison of today versus the past is going to be somewhat stable. What I mean by that is: Insofar as we think we're overstating inflation today, if we were also overstating it back in the 1950s, 1960s, and 1970s--and I don't think there's a dramatic change that would lead us to think we're mismeasuring inflation a lot more today than in the past--then that 90% number is also going to change when you change the inflation index in the present era. And so, you're not going to end up in a situation where you conclude that rates of absolute upward mobility have not fallen.
So, that's the kind of reasoning that makes that comparison over time fairly robust.
The other thing that I think is pretty robust, there's been some nice recent work done by folks like Robert Manduca and co-authors Yonatan Berman, where they basically take what we did in the United States, they replicate that, and then they do it for many other countries. And, what they show is the United States looks pretty different from most other countries in terms of this trend. So, the United States and United Kingdom in particular tend to have pretty rapidly declining rates of absolute upward mobility, but a number of other developed countries exhibit much less of a decline.
And, that also, I think, appears to be a relatively robust conclusion where you can make different specification choices. It's going to move all of those series in terms of the numbers you get for the current period. But, once you start comparing across countries or to different periods, those conclusions are more stable.
And so, that's why what I emphasize is the fading of the American dream and not so much am I absolutely sure the number is 50% and not 60% or 40%. I think it's hard to be super-precise about that in the current era.
Russ Roberts: Obviously, to some extent, it's half full, half empty. 50% is not half full or half empty. 50% is awful, but you can still say that if it's 72%, because these other assumptions were really the, quote, "right ones," that you could still argue that's not high enough: it's certainly less than 90%. You could argue that 90% was easy when people were coming out of the 1940s and 1950s: people growing up then it was a different time.
For me, it's really a question of, how easy or hard is it for people to rise, and what are the reasons for why it may have gotten harder? I think our American school system has done a horrible job, especially for the poorest people in helping them be equipped for a different economy that came along in the 1980s, 1990s, and 2000s.
And I guess--I am a little more worried than you are, I guess, about inflation measurement. It seems to me that the last 20, 30 years, the amount of improvement in quality, compared--the iPhone is an example, if you think about music storage, just that--it didn't just continue to improve slowly at 2 or 3% a year. It went through the roof, and then some, to the point where it's such a cornocopia a pleasure, if you like music. These are important, empirical questions and obviously we should try to do better at teasing out the magnitudes.
But, to me, the deeper question really is, can people get ahead?
And, a lot of the longitudinal evidence is that they do get ahead. And, in particular--and this is in your data also, which I think is often not noted--the poorest people do better than some of the richest people in terms of getting ahead of their parents. Partly because if your parents are poor, it's easier to get ahead of them. If your parents are rich, it's harder to get ahead of them. I don't think any--I don't know if you have children, do you have any children?
Raj Chetty: Two kids. Two small kids.
Russ Roberts: I don't think any of them will get the John Bates Clark Award at 32. If that's their goal--to beat their dad--they're going to struggle.
So, obviously there's biases; there are measurement challenges.
But, I guess I would say it in a slightly more social way, a cultural way, that: Pessimism sells, and people like waving it about. So, I would just--50-72% a huge range; whether it's high or low at 72 is still a legitimate, interesting question. It's a big range. It's an imprecise estimate, unless you want to say, 'No, no, no. These assumptions I made, this is why it's 50%. I'll stand by that.'
Raj Chetty: That's fair. I can tell you a little bit about where we landed where we did, and you're right that people pick and choose what to focus on in the media and so forth, which is a little bit out of our control as researchers.
But, why do we--how do we choose what's in the Appendix and what's in the main paper? To put it like that. So, the reasoning here was: It seems reasonable to start with the standard inflation index: that's what the government puts up; let's see what that gives us. Okay. So, we started with that.
Then, there was this issue of addressing for family size and how to think about the female labor force participation issue. And I've had conversations with other folks who say, 'You guys basically incorporate the fact that women started to work, and that makes things look rosier than they are. What you should really be doing is comparing sons to their dads; and the number is really 40%, not what you report.'
So, the way I look at it is: Look, like, that's on the one hand. If you look in the Appendix you can find that version. You can find the version where we adjust for family size; if you leave out the inflation part, that gets you to 60%. The 50% is in the middle. And, we try to emphasize the fading--the change aspect--which is really the thing that I think is robust, because of--what I was saying earlier--as opposed to, 'We know for sure this is the level.'
Now, people take that and some people like to emphasize one version of it, some people like to emphasize other versions of it. I guess I'm at least encouraged that people are talking about something to do with data and measurement in science, as opposed to--you know, there's some anchoring of the discussion.
Russ Roberts: So, let's close with a slightly more philosophical issue. You've been an advocate for economic education that emphasizes empirical technique more than it has in the past. And, certainly the field has moved dramatically in that direction over the last 20 years, away from theory. When I say 'theory,' I don't mean theory: I mean just even the use of standard economic technique. There's much more of an emphasis of, 'Just let the data speak.'
One of the problems I have with that--I have a lot of problems with it, but that's not your problem. This is the problem I want you to address, the [?scenario?] one, which is this: So, we don't know whether it's 50% or 72%. You did the best you could. And by the way, as it is with the previous work, the newer work we've been talking about previously, it's a very complex set of calculations. It's not just, like you said, going out and say, 'Well, who makes more than their parents?' It requires a lot of different assumptions. It requires challenges in measurement and in the data.
One of the things I find disturbing about this passion and intensity toward more empirical work, is: It's really hard to know how credible it is. I know it's called the credibility revolution--obviously in Psychology, less so than in Economics, but in Psychology a lot of the empirical work has been discredited. It hasn't been replicable. It has not replicated. Economics has a similar problem, I believe, that will become more upfront. We'll see. Maybe I'm wrong.
But, the idea that--let's take the work we've been talking about from the Facebook data: You move this person from a less connected community to a more connecting community, on average they're going to gain 20%. You can't verify it. It's not like--the example I use is: Where's John Glenn going to come down in his spacecraft if we change the angle? He's either there or he isn't. We can find out if the model is right; we can find out the calculations are accurate. In economics, econometrics is so complicated. Do you feel like we're getting closer to the truth? And: How competent should we be?
Raj Chetty: Those are all--I think it's a healthy skepticism, Russ.
Let me say a few different things. So, I guess, I don't think any one empirical study by itself that always relies on certain assumptions, is that going to stand the test of time forever? Can we be fully confident in all scenarios that that 20% number is, some universal constant? You know, obviously, no.
Where I do think we can make progress, though, is: replication is feasible. We can run an experiment. Like, one of the things I've been very encouraged by related to the earlier work is: we take these experimental studies, like the Moving to Opportunity data, and we overlay that on the predictions we make from the Opportunity Atlas observational work we had done. And, there's one of these papers, a scatter plot showing the two.
And what's really cool and made me very happy, is: that looks really good. Like, we were able to forecast when, purely through random chance, these kids were given an opportunity to move here versus there. We were able to predict how well they would do on average.
So there clearly was some stability across studies over time in those estimates, which gives me some confidence that we're making some kind of progress here.
And so, that's not going to be the case every single time. There are going to be some empirical studies where that doesn't hold. I think we learn from that. I think the best we can do, given that we're in a very complicated field, is to be very transparent about the assumptions we're making, try to show the different results we get under different sets of assumptions, and then have--with access to greater data--more researchers studying these questions and triangulating their findings so, we can see what seems to be stable, what isn't, what's changing over time.
And, you know, maybe I'm too much of an optimist, but I feel like we can make progress in our understanding there.
I also want to say: You know, sometimes I think people frame my work as--they interpret it as saying, 'Oh, we should just look at the data.' You know, 'It's not about theory, it's about data.' I actually think that that's misguided and not my own interpretation of the role of data and the role of theory in the age of big data.
And so, just to say a bit on that: I still see a very important role for theory as we have more data, in three particular domains.
So, first, in the formulation of hypotheses: You can't just take a data set and not have a theory and make progress. Right? So, if you take the topic we focused on in the first part of this conversation, on social capital and mobility, the analysis we did was firmly motivated by theory. It's a set of theories that economists and sociologists developed over the years about how different types of interactions might matter. And, what we found in the data is some of the theories--like Glen Lowry's theory on cross class interaction and how that might shape aspirations--they seem to be validated to some extent in the data. Whereas other theories on how having a tight knit community where norms can be enforced and your friends are friends with each other and so on--that seems much less relevant.
But, each of those tests were motivated by theories that people had developed.
If we had no theory to stand on, I'm not even sure exactly how we would have organized the data to look at this question. So, I think: One, theory plays a very important role in hypothesis formulation.
Second is--related to what you were saying--the way I would think about it is on extrapolation. So, if you're only going to look at--if you're going to rely solely on the data we have in hand, what we have seen, then you're never going to be able to make a prediction about some new policy, some new environment. That fundamentally has to rely on a theory. It might be a purely statistical theory. It might be an economic theory, but there has to be some theory for why whatever we did in the past is going to translate in some way to a new intervention we're going to do in a school or a new way we're going to organize society. So I think theory is crucial there.
And, then a third dimension, which I myself hope to spend more time on in the coming years and I think is really a priority for economics: Our empirical toolkit in the context of the credibility revolution, which I do think has greatly advanced our understanding of the number of things--it's fundamentally partial equilibrium. We can say what happens when one kid moves to a different neighborhood or gets a different teacher or goes to a different school, etc., but it doesn't deal with what I see as one of the core contributions of economics--thinking about general equilibrium and how these things play out when prices change when people re-sort, and so forth.
And, if you just think about the basic paradigm of treatment control comparisons, it is very hard to make that work in a setting where you're interested in equilibrium, because there's going to be fundamentally contamination of the control group by the treatment.
And so, I think theory continues to be extremely important in thinking about equilibrium impacts. I think thinking about how we use data to build better models that fit the partial equilibrium data, but then can help us make better predictions in equilibrium, seems like a very valuable direction for the work as well.
Russ Roberts: My guest today has been Raj Chetty. Raj, thanks for being part of EconTalk.
Raj Chetty: Thank you, Russ. My pleasure.