Few things are more private than tax records and Facebook data, but Harvard economist Raj Chetty and his assorted coauthors have managed to conduct huge studies using both.
A few years back, they used anonymized tax data to measure economic mobility for kids from different parts of the United States—calculating, for example, the relationship between a low-income kid’s economic prospects and where he grew up. They buttressed these results by analyzing siblings whose families had moved: when a family goes to a better place, younger siblings get exposed to the magic of the new location for more of their childhood, and such kids indeed saw better outcomes. That helps to establish that some places cause higher mobility and don’t just happen to be home to populations that would be upwardly mobile no matter what.
Now Chetty & Co. are back with an analysis of Facebook’s behind-the-scenes data, presented in two different Nature papers. It turns out that, just as areas of the U.S. vary in their economic mobility, they also vary in their “economic connectedness”—meaning the extent to which people of low socioeconomic status (SES) are friends with people of high SES. There’s a spiffy website where you can look up your ZIP code, high school, and college to see how common cross-class friendships are there.
Importantly, these two lines of research are linked. Economic connectedness, measured in the two new papers, is strongly correlated with the upward-mobility patterns demonstrated previously.
On one level, the new results are obvious. Interacting with richer people can help the less fortunate in any number of ways, from familiarizing them with upper-class norms and manners to giving them specific social connections that may one day lead to jobs. And it’s hardly surprising that areas with high levels of cross-class mobility also have high levels of cross-class interaction.
But for public-policy purposes, it’s important to suss out exactly what the study is picking up on. If we wanted to leverage these findings to improve social mobility, how would we do it?
For most of their analysis, the authors look at U.S. Facebook users aged 25 to 44, who are fairly representative of all Americans in that age range. But while Facebook has a lot of information on these people, it does not keep direct records of its users’ income or wealth. Therefore, when the study talks about people with high or low SES—more specifically, above- or below-median SES—it’s ranking people according to a combination of other bits of information, from the college they attended and the census “block group” they live in, down to the price of the phone they log in from.
A major question is why different places’ economic connectedness is correlated with their upward mobility. Does the former cause the latter, or is the study at least partly measuring something else? For example, some subsets of low-SES Americans are particularly disadvantaged (say, by racial discrimination or destructive subcultures); it’s possible that such disadvantages make cross-class friendships harder and reduce mobility, rather than the lack of friendships causing the reduced mobility directly.
Fortunately, the authors recognize this problem and test various alternative explanations. For example, the results hold up when looking at racial groups, or at least places where specific racial groups predominate. They also hold up when connectedness is measured based on high school friendships alone. (Adult friendships can result from social mobility, so this approach addresses the possibility of “reverse causation.”) And they hold up, to an extent, despite geographic income differences: richer places have more high-SES people around for low-SES people to interact with, and that’s part of the story, but if two places have the same typical income but different levels of economic connectedness, the place with higher connectedness will tend to have higher mobility.
Perhaps most strikingly, the authors run a statistical model pitting their economic-connectedness measure against some other major predictors of places’ rates of upward mobility, to see which still perform well after controlling for the others. Economic connectedness remains incredibly powerful—though black population share and the share of single-parent households are also strong predictors, and racial segregation seems to matter as well. Test scores, income, and income inequality become quite weak when sharing a model with the other contenders.
In the second Nature paper, the authors turn to the question of what drives economic connectedness itself, drawing on Facebook’s rich information about the schools and other institutions that users report being a part of. It seems to be a roughly 50/50 mix of two broad categories: difference in raw “exposure” to high-SES individuals (such as attending the same church) and a “friending bias” in which, even when low- and high-SES people do share the same settings, they’re less likely to become friends.
Interestingly, friending bias seems particularly low in religious settings, a reason to lament the long-term decline of religiosity. Friending bias tends to be elevated, by contrast, in larger high schools (presumably because self-segregation becomes easier when there are more potential friends to choose from), as well as in schools with high Advanced Placement class enrollment and high racial diversity. And in some cases, friending bias and cross-class exposure apparently compensate for one another: in high schools, friending bias rises with the share of high-SES students until the latter reaches about 60 percent. Clearly there are difficult tradeoffs for those wanting to engineer more economic connectedness.
But for such social engineers, there are ideas galore and have been for decades. A libertarian approach would be to reform ways that the government currently segregates the classes: relax aggressive zoning regulations or offer school choice to families stuck in segregated schools. More coercive, backlash-inducing methods would include school busing or plopping down subsidized affordable housing in the middle of wealthy communities. Chetty and his coauthors mention several clever possibilities, such as restructuring lunchrooms to facilitate more cross-class interaction and even a program that recruits low-SES personal trainers to work with high-SES clients.
Despite the limitations, there’s something for everyone in these two new studies. Conservatives can cheer the findings on religion and two-parent families; liberals can read the results as a rallying cry for greater integration. Even the apolitical will get a kick out of looking up their old schools on the interactive tool.
Photo by ALASTAIR PIKE/AFP via Getty Images