There has been a heated debate over the role that that economic factors played in the improbable journey that took Donald Trump all the way to the presidency. “Economic anxiety” is the umbrella term that many pundits have seized on to explain what drove a substantial number of (white) working-class voters to Trump. While economic anxiety is a somewhat vague concept, a decrease in prospects for social mobility is part of the story many have in mind.
The New York Times columnist David Leonhardt offered a familiar version of this argument this week. Drawing on research by the Stanford economist Raj Chetty and co-authors that showed that only half of the people born in 1980 today make more than their parents did (at the same age), he writes how frustration over this state of affairs “helps explain not only this year’s disturbing presidential campaign but also Americans’ growing distrust of nearly every major societal institution, including the federal government, corporate America, labor unions, the news media and organized religion.”
However, our exploration of data used by Chetty and co-authors, in previous studies, suggests that the hypothesis that decreased social mobility contributed meaningfully to Trump’s victory, while plausible under a cursory look, is ultimately not compelling. This does not mean that the political implications of social mobility are unimportant: Much to the contrary, mobility patterns are strongly correlated with voting patterns — much more so now than they were as recently as two decades ago. But the direction of that correlation is basically the opposite of what conventional wisdom suggests. In short, places with higher social mobility have tended to vote more Republican than places with lower social mobility. They also voted more heavily for Trump than for Romney four years ago.
In short, places with high social mobility drove Trump’s victory.
This much is true: social mobility is generally declining in the US
Why is the hypothesis that “economic anxiety” drove people to embrace Trump’s anti-trade and anti-immigration message so plausible — so much so that both of us started our investigation expecting to confirm it? The idea that a child born into the lower rungs of the income distribution can realistically hope to climb upward is one of the key promises underlying the “American Dream.” If that is not as true as may have been assumed, it is natural to think that frustration over this state of affairs may have added to anti-establishment sentiment among significant swaths of the population.
It is indeed true that social mobility in the United States falls short of that “American Dream” ideal. Work by Chetty and his colleagues has established that the levels of social mobility that now prevail in the US are lower than in many Western European countries, and have decreased compared to previous decades. That body of work has also shown that there are substantial differences, in social mobility, across various parts of the country.
Now consider a map produced by Chetty and co-authors (scroll down), which depicts a measure of social mobility by “commuting zones” (geographical aggregations of counties that are similar to metro areas, but cover the entire US, including rural areas). The map is based on data about children born in the early 1980s, to parents who lived in a given commuting zone and who were at the 25th percentile of the US income distribution at the time; it depicts where these children were ranked in the income distribution three decades later.
In the places with lowest social mobility, the average child was barely above her parents at that time, while in the most mobile locations she had climbed to the top half of the distribution.
It jumps out of the map that parts of the country with especially low levels of social mobility, notably the South and the Rust Belt, were key parts of the Electoral College coalition that gave us President Trump. This naturally lends support to the social mobility hypothesis.
Analyzing the data, we did in fact establish that social mobility is a remarkable predictor of Trump’s electoral performance — just not in the way we expected. When we look at the change in the Republican share of the total vote from Mitt Romney’s campaign in 2012 to Trump’s in 2016, the full set of social mobility variables studied by Chetty and co-authors explains almost 80 percent of the variation across counties.
Put simply, knowing patterns of social mobility in a county goes more than three-quarters of the way toward predicting the increase in support for Trump compared to Romney. For the sake of comparison, the racial composition of a county (the proportion of white, black, and Hispanic residents) accounts for a mere 15 percent of the variation in that increase.
The influence of social mobility on voting seems to be growing
What is more, the explanatory power of social mobility has increased substantially over the past couple of decades. If we focus on only two simple pieces of information — namely, the income percentiles achieved by children born into the 25th and 75th percentiles (call them lower- and upper-middle class, for simplicity) — we can explain just about a third of the variation across countries in the Republican share of the total vote in 2016. (Not Trump’s performance relative to Romney, that is, but the total vote share.)
In 1996, in contrast, this number was more like 8 percent. There has been a steady increase in the effects of social mobility on voting, as illustrated in this figure:
The figures below, on the other hand, show what we found when we translated the data from the Chetty map into graph form. Consider our two measures of upper- and lower-middle class mobility. Looking at those two measures independently, we ranked counties according to the degree of social mobility, in 20 equal-sized “bins,” and plotted the increase in GOP performance for the average county in each of these bins. It shows rather clearly that Trump performed better, relative to Romney, in places with higher social mobility.
But perhaps there are other factors confounding the interpretation of these pictures? For instance, as another cursory look at the Chetty map indicates, places with low levels of lower-middle class mobility also tend to be those with relatively high African-American populations. The pattern above might simply reflect that Trump did particularly poorly with African Americans.
Indeed, accounting for racial composition eliminates the correlation between lower-middle class mobility and the increase in GOP performance between 2012 and 2016. However, the link between upper-middle class mobility and Trump’s relative performance remains strong.
Most importantly, both kinds of mobility remain strong positive predictors of Trump’s absolute vote share. This remains true even if we control for education levels.
But what about Wisconsin, Pennsylvania, Ohio, and Florida?
There’s one last possibility. What if low social mobility was important for the Trump vote in swing states in particular? Focusing on the 14 states that were classified as being in play in 2016, the measures of both lower- and upper-middle class mobility are indeed relatively low, on average. But the direction of the relationship — i.e., counties with more mobility went for Trump — is the same as for the country as a whole. (The explanatory power of social mobility is somewhat weaker.) In other words, even in swing states, there is no evidence that low levels of social mobility drove the Trump vote.
The message relayed by the data is therefore quite clear: Higher levels of social mobility predict stronger GOP performance. This has been increasingly true over the years, and is especially marked for the upward mobility prospects of the upper-middle class.
Why would that be so? One natural explanation may be that, in places with greater social mobility, people could be less willing to support redistribution: Maybe they would expect to be among those paying higher taxes as a result. However, interesting new research by Michael George has noted that this does not seem to be the case. People in places with greater social mobility vote more Republican, but are not more likely to say that they want less redistribution.
We are left with an open question. Obviously, establishing a causal impact of social mobility on partisan preferences would require much additional work; we have merely detected correlations. That said, the patterns in the data push against the interpretation of Trumpism as indicative of a realignment in which Democrats are “becoming the party of rising professionals,” while Republicans represent “downwardly mobile white Americans” (to use the words of George Packer in the New Yorker). It also pushes back against David Leonhardt’s claim that “nostalgia and anger over the fading American dream helped elect” Trump.
We don’t yet have a strong theory to put in place of the conventional wisdom about “economic anxiety.” But showing that this appealing narrative, at least as it pertains to social mobility, turns out not to fit the facts has some value, too.
Filipe Campante is associate professor of public policy, Harvard Kennedy School. David Yanagizawa-Drott is professor of development and emerging markets, University of Zurich.