Chetty offers this nugget from previous research as as starting point:
In particular, let’s think about the chance that a child born to parents in the bottom quintile—the bottom fifth of the income distribution—has of rising up to the top fifth of the income distribution. If you look at the data across countries, there are a number of recent studies that have computed statistics like this, called intergenerational transition matrices. You see that in the United States, kids born to families in the bottom fifth have about a seven and a half percent chance of reaching the top fifth. That compares with nine percent in the United Kingdom, 11.7% in Denmark, and 13.5% in Canada. ... One way you might think about this is that your chance of achieving the American dream are in some sense almost two times higher if you’re growing up in Canada rather than the United States. I think these international comparisons are motivating. It’s a little bit difficult to figure out how to make sense of them.
Start with a basic question: What do we need to know if we want to measure intergenerational economic mobility? Basically, you need a measure of income for two separate generations, choosing an age when both older and younger generations are reasonably well-established in a career. Then you can compare where in the income distribution the younger generation falls, relative to their parents, and in that way have a sense of the chances of how likely someone from one part of the income distribution is to move (higher or lower) to a different part of the income distribution. Best of all, you would like to have a really big nationally representative sample of this data, so that you can look at how such effects might vary across different locations, different sexes, and races/ethnicities.
There is no single data set that will give you this kind of information. So Chetty and a team of researchers that has now grown to a couple of dozen people figured out how to combine existing datasets to get the necessary data. Specifically, as Chetty describes it:
In particular we use the 2000 and 2010 decennial census, as well as the American Community Survey—the ACS—which covers about 1% of Americans each year. We take that census data and link it to federal income tax returns from 1989 to 2015, creating a longitudinal data set. ...Those of you who have kids and live in the United States know that you have to write down your kid’s Social Security number on your tax return to claim him or her as a dependent. That allows us to link 99% of children in America back to their parents. What I’m going to describe here, the set of kids we’re seeking to study, are children born between 1978 and 1983 who were born in the United States or came to the United States as authorized immigrants while they were kids. That’s our target sample.
In practice, we have an analysis sample of about 20.5 million children once we’ve done the linkage of those various data sets, and that covers about 96% of the kidswe expect to be in our target sample. It’s not 100% because there are some kids for whom you’re not able to find a tax return or who weren’t claimed or who you weren’t able to link the census data to the tax data. But it’s still pretty good—at 96% you have essentially everybody you’re looking to study.
Just to be clear, this data is "de-identified": that is, stringent precautions are taken so that researchers don't see anyone's actual name or Social Security number or other personal information. Constructing this dataset is a substantial task, and it's a task that would not have been possible for researchers of previous generation.
Here's one result. The horizontal axis shows the percentage of the income distribution for the older generation; the vertical axis the average for where the next generation ended up in the income distribution. The marked point shows that when the older generation was in the 25th percentile, the average outcome for the younger generation is the 41st percentile of the income distribution. Overall, the flatter this line, the more intergenerational income mobility exists: a perfectly flat line would mean that no matter where the older generation started, the expected result for the younger generation is the same.
But while a lot of the previous research on intergenerational mobility had data that was nationally representative, the US-based research in particular has not had enough data that it could make plausible statements about intergenerational mobility at more local levels.
As a concrete example to help the exposition, Chetty sticks to this case where a parent's household is in the 25th percentile of the income distribution. Then he can ask: in different metropolitan areas across the United States, where is the average income of the younger generation higher or lower?
Here's a map showing the pattern across the US where the younger generation is white and black men. The blue areas in the "heat map" show where the income of the younger generation is higher than the older generation; the white areas show where it's the same; the red areas show where it is lower. For black men, on the left, the overall pattern is mostly red and white: that is, the younger generation of black men usually have the same or lower income than their parents. For white men, on the right, the overall pattern is mostly blue and white; that is, the younger generation of white men are usually doing the same or better than their parents.
Thus, Chetty and his team can tackle the question: what do patterns of intergenerational mobility look like not at the national level, or at the metro area level, but at the neighborhood level. Sticking to the earlier concrete example of children born to a family at the 25th income percentile, does their income mobility differ depending on the neighborhood in which they live? Chetty writes:
[T]he geographic scale on which we should think about neighborhoods as they matter for economic opportunity and upward mobility is incredibly narrow, like a half mile radius around your house. We find this not just for poverty rates, but many other characteristics. If you look at differences in characteristics outside that half mile radius, they have essentially no predictive power at all. I think that’s extremely useful from a policy perspective. We started this talk with the American dream. We now see that its origins, its roots, seem to actually be extremely hyperlocal.Do the hyperlocal neighborhoods that have more intergenerational income mobility tend to share certain characteristics? Yes. For example, the share of households in the neighborhood with two-parent families makes a difference, as does the "social capital" of the neighborhood as measured by whether it has community gathering places like churches or even bowling alleys.
At some level, all of this may sound like "common sense for credit," as economics has been described. Certain neighborhoods, not too far apart, are more correlated with intergenerational mobility than others. But are the neighborhood effects something that could potentially be used for public policy? To put it another way, if the neighborhood changes, does someone from the younger generation basically have the same level of intergenerational mobility--because mobility is mostly determined by your family--or does someone from the younger generation perhaps have a higher or lower level of mobility--because the neighborhood has a separate causal effect on eventual economic outcomes?
There are a variety of ways one might address this question. There's a major federal study called Moving to Opportunity. It was a randomized study: that is, a randomly selected half the sample got a certain program, but half didn't, so you can then compare outcomes between the results. Chetty describes:
Let me start with the moving to opportunity (MTO) experiment. This experiment was conducted in five large cities around the United States. I’m going to focus on the case of Chicago, just to pick one illustration. In this experiment, researchers gave families living in very high poverty public housing projects, for instance, the Robert Taylor Homes in Chicago, one of two different kinds of housing vouchers through a random lottery. One group received Section 8 vouchers, which were vouchers that enabled them to move anywhere they could find housing. This assistance was on the order of about $1,000 per month in today’s dollars. The other group, called the experimental group, was given the same vouchers worth exactly the same amount, but they came with the restriction that users had to move to a low poverty area. These areas were defined as census tracts with a poverty rate below 10%.It turns out that the randomly selected group who moved to lower-poverty areas as children did indeed have higher incomes as adults. The exact numbers of course have some statistical uncertainty built in. But as a bottom line, Chetty writes:
That is to say, if I moved to a place where I see kids growing up to earn $1,000 more, on average, than in my hometown, I myself pick up about $700 of that. Just to put it differently, something like 70% of the variation in the maps that I’ve been showing you seems to reflect, if you take this oint estimate directly, causal effects of place, and 30% reflects selection. So a good chunk of it seems to actually be the causal effects of place.Another statistical approach is just to look at families who move for any reason. Look at families with several children. When that family moves, some of their children will grow up in the new neighborhood for longer than others, and it turns out that the income gains from moving to the new neighborhood for children growing up in the same family line up with how long the child lived in that neighborhood. Chetty describes various other approaches to demonstrating that the neighborhood in which you grow up, where that is defined as the area within about a half-mile of your home, has a lasting effect on economic prospects.
What public policy recommendations might flow from this finding? It would be impractical and costly to pay for vast numbers of people to relocate from current neighborhoods. But there's a smaller-scale policy that might well be workable, which just involves providing information and connections to families with an interest in moving. Chetty writes:
A different view is maybe this is about some sort of barriers, frictions that are preventing families from getting to these places. Maybe they lack information, maybe landlords in those neighborhoods don’t want to rent to them, maybe they don’t have the liquidity they need to get to those places, and so on. We are conducting a randomized trial where we’re trying to address a bunch of those barriers by providing information, and by simplifying the process for landlords by providing essentially brokerage services, like search assistance, in the housing search process. We take that for granted in the high-end of the housing market, but it basically doesn’t exist at the low-end of the housing market. We take about 1,000 families, 500 of which receive the services, and 500 don’t, randomly chosen. ...
We found that this was an incredibly impactful intervention; we were extremely surprised by how much impact our services had on families’ likelihoods of moving to higher-opportunity neighborhoods. In the control group, less than one-fifth of families moved to higher opportunity areas. Eighty percent of families that received these vouchers chose to live in places that are relatively low mobility. In the treatment group, this completely changed. The vast majority of families in the treatment group are now living in these high mobility places.I was just in Seattle talking to some of the families who’ve moved. They’re incredibly happy and describe how this small set of services, which only comes at a 2% incremental cost relative to the baseline cost of the housing voucher program, dramatically changed their choices and their kids’ experience. ...
There are a couple elements. First, from an economic perspective, we provide damage mitigation funds. This is basically an insurance fund that says that if anything goes wrong, we will cover it. In practice, the amount of expenses incurred are essentially zero, but I think it gives landlords some peace of mind. Second, there’s a simplification in the inspection process, which traditionally involves a lot of red tape and delays. We shortened the inspection process to 1 day, making it much simpler. Third—this actually surprised me—apparently telling landlords that their units can provide a pathway to opportunity for low-income kids actually makes them much more motivated to rent their units to certain families. ...
In fact, now we find landlords coming to the housing authority saying things like: “I heard about this program,” or, “I had a really good experience with your previous tenant, I want to now rent again.” I think we can change that equilibrium if we do it thoughtfully.
What about taking steps to improve the prospects for intergenerational mobility in neighborhoods in a direct way? It's worth remembering that Chetty's evidence suggests that what really matters is the half-mile around where people live, so what would seem to be called for is projects that improve the neighborhoods at a local level. Chetty writes:
What specific investments can be useful? Of course, that’s the question you’d want to knowthe answer to. That could range from things like, most obviously, improving the quality of schools in an area to things like mentoring programs, and changing the amount of social capital, if we can figure out ways to measure and manipulate things like connectedness, reducing crime, and physical infrastructure. There are many such efforts that have been implemented over the years by local governments, nonprofits, and other practitioners.
You might ask which of those things is actually most effective; what’s the recipe for increasing upward mobility in a given place? I think the honest answer is that we just do not know yet. The reason for that is that there are lots of these place-based efforts where someone invests a lot of money in a given neighborhood. The neighborhood looks completely different 10 years down the road, but you have no idea whether that’s because new people moved in and displaced the people who were living there before, so the neighborhood gentrified, or if the people who lived there to begin with benefitted. And again, I think resolving that question comes back to having historical longitudinal data and being able to follow the people who lived there to begin with.
As you read this brief overview of a much larger body of work, you may find your mind raising questions about other possible connections or policies. That's natural. It's a genuinely exciting area of research that is opening up before our eyes.