The standard narrative of Donald Trump’s victory is simple: Trump won in states like Michigan, Wisconsin, and Pennsylvania on the back of unprecedentedly strong support from white working-class voters. Those making less than $30,000 swung to Trump by 16 percentage points relative to the 2012 election, narrowing the Democratic lead among low-income voters to a mere 12 points. Very few of the voters were people of color.
The natural upshot of this analysis is that class was a huge reason why Trump won, that poor whites found Trump’s attack on economic elites revitalizing. It was “the revenge of the forgotten class,” as ProPublica’s Alec MacGillis puts it.
But is that really true? A new study from the UK’s Resolution Foundation, a well-regarded think tank focusing on economics, looked at this question in some statistical depth — and the answer is more complicated than you might think.
Researchers Stephen Clarke and Dan Tomlinson looked at the effect of income on Trump’s support — and, at a surface level, found a correlation between income in a particular area and support for Donald Trump. But when they took a closer look, they found that this effect was less powerful than it seems. Instead, sociodemographic factors, particularly race and education, ended up being more important explanations of the unexpected Trump surge.
That suggests the vaunted swing among voters making less than $30,000 might not have happened simply because of the fact that they make less than $30,000. It might have more to do with the fact that these people tend to be less educated and white — factors that incline people to support far-right anti-immigrant politicians around the world, like France’s Marine Le Pen and the Netherlands’ Geert Wilders.
Now, the report doesn’t disprove that economics played a key role in the election. Indeed, if you read it, the authors are very clear that they think the economy played some role — and the data isn’t fine-grained enough to answer the question one way or another.
The point, instead, is that some observers are being too quick to explain this election through a simple economic lens, based on the mere fact that Trump won bigger among the white working class than Romney did. It’s not that simple.
What the study found
To figure out why Trump performed better against Clinton than Mitt Romney did against Obama, Clarke and Tomlinson looked at the degree of change in county-level results between Romney’s totals in 2012 and Trump’s in 2016. The goal was to figure out what characteristics of these counties correlated most strongly with a shift toward the GOP — whether, say, wealthier or poorer counties swung harder toward Trump.
The biggest factor, according Clarke and Tomlinson’s regression analysis, was education. In counties with large percentages of people with high school degrees or less, Trump got a lot more support than Romney did. In counties with more educated voters, Trump’s gains were far more modest. The correlation is extremely strong:
Interestingly, this makes it difficult to isolate the economic impact. In Clarke and Tomlinson’s initial analysis, it looked like there was a real correlation between household income in a county and its swing to Trump, with less affluent places backing him more than they had Romney. A cursory reading of the report would suggest this was one of its major findings.
But the problem is that their initial statistical study didn’t take into account the effects of education. Low-income places also tend to have lower levels of educational attainment, so it’s hard to say whether income had any effect independently of education. Their tests showed that once you factor in the powerful effect of education, income ceases to be a strong predictor of a Trump vote.
“When you add in the education variable, it just knocks a lot of stuff out,” Clarke told me when I called him to get more clarity on his findings.
This wasn’t the case with other demographic variables. Clarke and Tomlinson also found that counties with more white voters (surprise!) shifted far more heavily to Trump.
Similarly, areas with more foreign-born residents were less likely, all things considered, to shift toward Trump.
Clarke and Tomlinson also set up what’s called an interactive term, a way of testing how multiple variables interact, between a county’s whiteness and its education level. The idea here was to figure out whether less educated people in general, or only less educated whites, swung hard toward Trump.
The findings were very clear: Only less educated whites were significantly more likely to have swung to Trump. “As the share of the white population in a county increases, so does the effect of the share of people with only a high-school education,” Clarke and Tomlinson write.
Now, Trump won whites overall handily, regardless of education or income. But Mitt Romney also won whites nationwide; the question here is what Trump did to improve on Romney’s performance. And on that question, Clarke and Tomlinson’s data tells a consistent story: White people in less educated areas swung heavily to Trump, while people of color and more educated whites did not.
“Demographics — particularly race and foreign born — and education were the strongest predictors [of a swing to Trump],” Clarke says.
Why the link between education and the economy is so important
What to make of these findings? The principal takeaway is that the settled narrative — Trump won by winning working-class whites away from Democrats — is more complicated than it seems.
The “white working class” theory almost always has an economic underpinning. If Trump won because of a revolt of poor whites, the theory goes, then he must have won as a result of poor whites’ economic grievances. Trump’s economic populism — his attacks on elites and free trade agreements — must have resonated.
But the fact that education was actually more important than income, to the point where it entirely cancels out income’s role in Clarke and Tomlinson’s regression, suggests an alternative explanation: that one’s low levels of education, not low income, is the key explanatory variable.
A useful comparison here is the rise of European far-right parties. These parties, like the Front National in France or the Party for Freedom, have a very similar message to Trump’s. They are populist, in the sense that they’re highly critical of the current establishment, but are (like Trump) principally defined by their xenophobia and hostile approach toward minority groups.
Scholars of the European far right generally believe that these parties’ appeal is fundamentally cultural, not economic: that they draw support as a result of racial and cultural anxieties about immigration rather than economic anxiety created by globalization and the decline of the middle class. One of their key pieces of evidence is the priority of education over the economy — the kind of effect we see in Clarke and Tomlinson’s study.
Study after study finds that income isn’t a particularly strong predictor of far-right support once you account for education. Education, instead, is far more important. Less educated whites who do well financially — think a successful plumber or shopkeeper — are as, if not more, likely to support the far right as poor whites.
This finding, that education matters much more than income, is extremely consistent. Eric Kaufmann, a professor at the University of London who studies this phenomenon, dubs it an “iron law” of far-right studies.
This suggests that the reason people support these parties isn’t about income or employment. It’s something else, something strongly related to their education level — something that seems to affect their tolerance for cultural difference. While far-right supporters in Europe aren’t consistently poorer, they are consistently more hostile to immigrants than other white natives.
This could be the story of the 2016 election. It could be that Trump’s key innovation was a more open illustration of the power of white ethnic grievance — that his plan to ban Muslim immigration actively attracted low-educated Americans in the same way the Party for Freedom’s call to ban Muslim immigration attracted low-educated Dutch voters.
This kind of cultural explanation would make sense — it’s Trump’s white identity politics, and not his economic message, that most clearly distinguished him from previous Republicans like Romney. Studies from the GOP primary, and pre-election polls, found high racial resentment was a far better predictor of a voter’s likelihood of supporting Trump than any economic variable.
Clarke and Tomlinson’s research can’t resolve this question one way or another. County-level data isn’t specific enough to tell us how individuals in those counties behaved. It could be that wealthy people who live in low-education counties were less likely to vote for Trump than their working-class neighbors. Or it could be the other way around.
Regardless, it suggests that the push among some liberals, like Bernie Sanders, to respond to this election with a “populist” economic message may not be the right approach. If the swing toward Trump among whites really was about racial and cultural anxieties — as some good research suggests — then they will have misdiagnosed the problem.
The issue wouldn’t be that Democrats “forgot” white workers; it’s that Trump promised them the kind of white identity politics they’ve been yearning for.