The 2018 midterm election dramatically shifted the balance of power in the House of Representatives, from Republican to Democratic control. Many expected that, given President Trump’s relatively low approval rating. Historically, that had meant the president’s party would lose many House races. Pre-election polling largely confirmed the likely Democratic takeover.

But here’s what we haven’t yet known: Which groups supported the Democrats in this election? How do these patterns compare to previous elections?

Below, you can see five charts that help to explain what happened. For the most part, these charts are based on data from the Cooperative Congressional Election Study (CCES), a large-scale academic survey conducted in every election year since 2008. For the 2018 CCES analysis, we used pre-election interviews with respondents weighted to be nationally representative of the adult population. We then applied a likely voter model trained on previous election cycles to create estimates for the 2018 electorate.

1. How did different age groups vote?

First, let’s look at vote patterns among different age groups in House races over the past decade. While this year all age groups voted more Democratic than they had in 2016, those under 50 years old shifted more. In particular, 18- to 29-year-old voters chose Democratic candidates over Republican candidates by a 2-to-1 margin in 2018. And while we won’t know for some time, some indications suggest they may have made up a larger share of the electorate than in typical midterm elections — or in other words, that young people turned out to vote in especially high numbers.

2. How did suburbanites vote?

Suburban districts were among the hardest-fought battlegrounds in this campaign. The CCES data show Democrats did well in those districts. The chart below shows the House vote among individuals living in the suburbs, broken out by U.S. region. Suburban voters supported Democratic House candidates by a healthy margin over Republican candidates in every region except the South, where the party breakdown was even.

3. As expected, women and men voted very differently

Another pattern everyone was watching was the gender gap — which, as the next chart shows, was the largest we have seen in at least a decade. While nearly 60 percent of women who voted for one of the two major parties voted for Democratic candidates, only 47 percent of men did. That’s a gender gap of 13 points.

4. Let’s break down women and men by race and education

But of course, women and men are incredibly broad groups, made up of every U.S. demographic. So which subgroups of women and men were furthest apart? The next chart plots the two-party vote share among white voters (since voters of color went overwhelmingly for Democratic candidates, no matter the gender), by gender and education.

As you can see, white women without college degrees moved modestly toward the Democrats, more so than white men without college degrees. But a far greater proportion of white college-educated women swung to the Democrats, even beyond their previous support for Democratic candidates in previous elections. How much did they swing? In 2018, white college-educated women increased their support for Democratic candidates by eight percentage points over 2016. In previous cycles, this group has accounted for about 15 percent of the electorate, so large Democratic margins among this demographic surely helped fuel Democratic candidates’ success in 2018.

5. Why did college-educated white women swing so far toward the Democrats?

What explains college-educated white women’s big shift? The final chart comes from analysis I conducted for Data for Progress. In that piece, I compare the role of voters’ attitudes about women in this election to the role it played in 2018.

Among other factors, I looked at what researchers call “hostile sexism,” a set of antagonistic attitudes toward women that stem from a belief that women want to control men. While hostile sexism was a strong predictor of support for Trump, it did not affect how people voted in their House races in 2016. That changed in 2018.

The chart below shows how higher levels of sexism are related to voting for the Republican candidate in both 2018 and 2016, controlling for other factors such as ideology, partisanship, racial attitudes and demographics. In 2016, a voter’s agreement or disagreement with sexist statements (which you can find on the x-axis) did not matter much for whether they voted for the Republican House candidate. In 2018, however, people who were more likely to disagree with sexist statements (i.e., had less hostile sexism) were much less likely to vote Republican.

In essence, less-sexist voters punished Republican House candidates in a way they did not in 2016. What’s more, Republicans did not gain any more sexist voters to offset that loss.

Overall, these five charts suggest Republicans might wish to be concerned about being tied to a president whose rhetoric is often divisive and offensive. Doing so is turning off younger voters at historic rates, while also driving away women (especially those with college degrees). If the Republican Party brand becomes increasingly synonymous with Trump, these patterns may persist in 2020 and beyond.

Read more:

Brian F. Schaffner (@b_schaffner) is the Newhouse Professor of Civic Studies in Tisch College and the Department of Political Science at Tufts University.