One of the biggest challenges facing those who seek to understand U.S. elections is establishing an accurate portrait of the American electorate and the choices made by different kinds of voters. Obtaining accurate data on how people voted is difficult for a number of reasons.

Surveys conducted before an election can overstate – or understate – the likelihood of some voters to vote. Depending on when a survey is conducted, voters might change their preferences before Election Day. Surveys conducted after an election can be affected by errors stemming from respondents’ recall, either for whom they voted for or whether they voted at all. Even the special surveys conducted by major news organizations on Election Day – the “exit polls” – face challenges from refusals to participate and from the fact that a sizable minority of voters actually vote prior to Election Day and must be interviewed using conventional surveys beforehand.

This report introduces a new approach for looking at the electorate in the 2016 general election: matching members of Pew Research Center’s nationally representative American Trends Panel to voter files to create a dataset of verified voters.

See also: A Demographic Profile of 2018 Midterm Voters

The analysis in this report uses post-election survey reports of 2016 vote preferences (conducted Nov. 29-Dec. 12, 2016) among those who were identified as having voted using official voting records. These voter file records become available in the months after the election. (For more details, see “Methodology.”) Among these verified voters, the overall vote preference mirrors the election results very closely: 48% reported voting for Hillary Clinton and 45% for Donald Trump; by comparison, the official national vote tally was 48% for Clinton, 46% for Trump.

This data source allows researchers to take a detailed look at the voting preferences of Americans across a range of demographic traits and characteristics. It joins resources already available – including the National Election Pool exit polls, the American National Election Studies and the Current Population Survey’s Voting and Registration Supplement – in hopes of helping researchers continue to refine their understanding of the 2016 election and electorate, and address complex questions such as the role of race and education in 2016 candidate preferences.

It reaffirms many of the key findings about how different groups voted – and the composition of the electorate – that emerged from post-election analyses based on other surveys. Consistent with other analyses and past elections, race was strongly correlated with voting preference in 2016. But there are some differences as well. For instance, the wide educational divisions among white voters seen in other surveys are even more striking in these data.

Overall, whites with a four-year college degree or more education made up 30% of all validated voters. Among these voters, far more (55%) said they voted for Clinton than for Trump (38%). Among the much larger group of white voters who had not completed college (44% of all voters), Trump won by more than two-to-one (64% to 28%).

There also were large differences in voter preferences by gender, age and marital status. Women were 13 percentage points more likely than men to have voted for Clinton (54% among women, 41% among men). The gender gap was particularly large among validated voters younger than 50. In this group, 63% of women said they voted for Clinton, compared with just 43% of men. Among voters ages 50 and older, the gender gap in support for Clinton was much narrower (48% vs. 40%).

About half (52%) of validated voters were married; among them, Trump had a 55% to 39% majority. Among unmarried voters, Clinton led by a similar margin (58% to 34%).

Just 13% of validated voters in 2016 were younger than 30. Voters in this age group reported voting for Clinton over Trump by a margin of 58% to 28%, with 14% supporting one of the third-party candidates. Among voters ages 30 to 49, 51% supported Clinton and 40% favored Trump. Trump had an advantage among 50- to 64-year-old voters (51% to 45%) and those 65 and older (53% to 44%).

For a detailed breakdown of the composition of the 2016 electorate and voting preferences among a wide range of subgroups of voters, see Appendix. For the survey methodology and details on how survey respondents were matched to voter records, see “Methodology.”

2016 vote by party and ideology

Voter choice and party affiliation were nearly synonymous. Republican validated voters reported choosing Trump by a margin of 92% to 4%, while Democrats supported Clinton by 94% to 5%. The roughly one-third (34%) of the electorate who identified as independent or with another party divided their votes about evenly (43% Trump, 42% Clinton).

Similarly, voting was strongly correlated with ideological consistency, based on a scale composed of 10 political values – including opinions on race, homosexuality, the environment, foreign policy and the social safety net. Respondents are placed into five categories ranging from “consistently conservative” to “consistently liberal.” (For more, see “The Partisan Divide on Political Values Grows Even Wider.”)

Virtually all validated voters with consistently liberal values voted for Clinton over Trump (95% to 2%), while nearly all those with consistently conservative values went for Trump (98% to less than 1% for Clinton). Those who held conservative views on most political values (“mostly conservative”) favored Trump by 87% to 7%, while Clinton received the support of somewhat fewer among those who were “mostly liberal” (78%-13%). Among the nearly one-third of voters whose ideological profile was mixed, the vote was divided (48% Trump, 42% Clinton).

Religious affiliation and attendance

As in previous elections, voters in 2016 were sharply divided along religious lines. Protestants constituted about half of the electorate and reported voting for Trump over Clinton by a 56% to 39% margin. Catholics were more evenly divided; 52% reported voting for Trump, while 44% said they backed Clinton. Conversely, a solid majority of the religiously unaffiliated – atheists, agnostics and those who said their religion was “nothing in particular” – said they voted for Clinton (65%) over Trump (24%).

Within the Protestant tradition, voters were divided by race and evangelicalism. White evangelical Protestants, who constituted one out of every five voters, consistently have been among the strongest supporters of Republican candidates and supported Trump by a 77% to 16% margin.

This is nearly identical to the 78% to 16% advantage that Mitt Romney held over Barack Obama among white evangelicals in Pew Research Center polling on the eve of the 2012 presidential election.

Among white mainline Protestants (15% of voters overall) 57% said they voted for Trump and 37% reported voting for Clinton. Clinton won overwhelmingly among black Protestants (96% vs. 3% for Trump).

White non-Hispanic Catholics supported Trump by a ratio of about two-to-one (64% to 31%), while Hispanic Catholics favored Clinton by an even larger 78% to 19% margin.

Among all voters, those who reported attending services at least weekly favored Trump by a margin of 58% to 36%; the margin was similar among those who said they attended once or twice a month (60% to 38%). Those who reported attending services a few times a year or seldom were divided; 51% supported Clinton and 42% supported Trump. Among the nearly one-quarter of voters (23%) who said they never attend religious services, Clinton led Trump by 61% to 3o%.

Demographic and political profiles of Clinton and Trump voters

As the pattern of the votes implies, the coalitions that supported the two major party nominees were very different demographically. These differences mirror the broad changes in the compositions of the two parties: The Republican and Democratic coalitions are more dissimilar demographically than at any point in the past two decades.

In 2016, a 61% majority of those who said they voted for Clinton were women, while Trump voters were more evenly divided between men and women. Whites constituted nearly nine-in-ten (88%) of Trump’s supporters, compared with a smaller majority (60%) who voted for Clinton. Clinton’s voters also were younger than Trump’s on average (48% were younger than 50, compared with 35% for Trump).

Among Clinton voters, 43% were college graduates, compared with 29% of Trump voters. And while non-college whites made up a majority of Trump’s voters (63%), they constituted only about a quarter of Clinton’s (26%).

About a third of Clinton voters (32%) lived in urban areas, versus just 12% among Trump voters. By contrast, 35% of Trump voters said they were from a rural area; among Clinton voters, 19% lived in a rural community.

The religious profile of the two candidates’ voters also differed considerably. About a third of Clinton voters (35%) were religiously unaffiliated, as were just 14% of Trump voters. White evangelical voters made up a much greater share of Trump’s voters (34%) than Clinton’s (7%). One-in-five Trump voters (20%) were white non-Hispanic Catholics, compared with just 9% of Clinton voters. And black Protestants were 14% of Clintons supporters, while almost no black Protestants in the survey reported voting for Trump.

How did 2016 voters and nonvoters compare?

The data also provide a profile of voting-eligible nonvoters. Four-in-ten Americans who were eligible to vote did not do so in 2016. There are striking demographic differences between voters and nonvoters, and significant political differences as well. Compared with validated voters, nonvoters were more likely to be younger, less educated, less affluent and nonwhite. And nonvoters were much more Democratic.

Among members of the panel who were categorized as nonvoters, 37% expressed a preference for Hillary Clinton, 30% for Donald Trump and 9% for Gary Johnson or Jill Stein; 14% preferred another candidate or declined to express a preference. Party affiliation among nonvoters skewed even more Democratic than did candidate preferences. Democrats and Democratic-leaning independents made up a 55% majority of nonvoters; about four-in-ten (41%) nonvoters were Republicans and Republican leaners. Voters were split almost evenly between Democrats and Democratic leaners (51%) and Republicans and Republican leaners (48%).

While nonvoters were less likely than voters to align with the GOP, the picture was less clear with respect to ideology. Owing in part to the tendency of nonvoters to be politically disengaged more generally, there are far more nonvoters than voters who fall into the “mixed” category on the ideological consistency scale. Among nonvoters who hold a set of political values with a distinct ideological orientation, those with generally liberal values (30% of all nonvoters) considerably outnumbered those with generally conservative values (18%).

Voters were much more highly educated than nonvoters. Just 16% of nonvoters were college graduates, compared with 37% of voters. Adults with only a high school education constituted half (51%) of nonvoters, compared with 30% among voters. Whites without a college degree made up 43% of nonvoters, about the same as among voters (44%). But nonwhites without a college degree were far more numerous among nonvoters (at 42%) than they were among voters (19%).

There also were wide income differences between voters and nonvoters. More than half (56%) of nonvoters reported annual family incomes under $30,000. Among voters, just 28% fell into this income category.

CORRECTION (October 3, 2019): The text of the report has been edited to correct an error in the reported vote choice of non-Hispanic white mainline Protestants, 57% of whom voted for Donald Trump vs. 37% for Hillary Clinton. The graphics and tables were unaffected by this error.

CORRECTION: (August 9, 2018): In the chart “Among validated voters in 2016, wide gap among whites by education,” the “share of electorate” column has been edited to reflect updated percentages for gender by race to correct for a data tabulation error. Changes did not affect the report’s substantive findings. The associated detailed tables have also been updated accordingly.