Sabato's Crystal Ball

Another Look Back at 2016

Comparing the exit poll and the Cooperative Congressional Election Study

Geoffrey Skelley, Associate Editor, Sabato's Crystal Ball March 23rd, 2017


On election night in November, exit polls provided the first insight into how different demographic groups voted. But months later, other richer data sets are being released, and they provide researchers with new information about the election and the voters that participated in it. One such tool is the Cooperative Congressional Election Study, which is a large-sample national survey. The preliminary 2016 post-election version of the CCES study came out in early March, and it provides a treasure trove of information.

One way of looking at the data is to compare the findings of the CCES to the national exit poll to see how they differ and what they say about the makeup of the 2016 electorate. This is not to say one is more right than the other; if anything, the truth may lie somewhere in between. There are also differences between the two data sets that make comparing them an imperfect exercise. Besides methodological differences in how the surveys were conducted, variations in how questions were asked and possible answers also complicate things. For instance, the 2016 CCES included more racial categories for respondents to choose from, such as “Middle Eastern,” that the exit poll did not use. It’s also important to note that many studies suffer from over-reported turnout, and the CCES is no different. Most surveys have a difficult time combating overrepresentation of actual voters, as well as the misreporting of voting by people who didn’t actually vote. The CCES will mitigate this to some degree in later releases of the 2016 data by attempting to validate the votes of respondents. And the CCES, like all studies, weights the data for various criteria, such as age, education, gender, and turnout. Still, the data do not perfectly capture the actual electorate that showed up on Nov. 8, 2016. Nonetheless, the CCES is one of the best resources we have for analyzing elections. With these important caveats out of the way, let’s dig in.

The CCES provides further evidence of a healthy gender gap in American politics. President Donald Trump carried men by five percentage points and Hillary Clinton carried women by nine percentage points. But that 14-point marginal gap is smaller than the record-setting chasm in the exit poll (24 points), as shown in Table 1.

Table 1: 2016 vote by gender, exit poll vs. CCES

Note: Columns and rows may not add up to 100 due to rounding.

Source: CNN exit poll data, CCES Common Content 2016

Race is one of the most important explanatory factors in voting behavior, with white voters tending to vote more Republican and nonwhite voters tending to vote even more Democratic. Thus, the ratio of whites to nonwhites in the electorate is very important. The CCES data showed an electorate that was slightly whiter than what the exit poll found. That finding corresponds to other work that finds the exit polls tend to exaggerate the nonwhite share of the electorate.

The data in Tables 2 and 3 compare vote by race in the exit poll and the CCES. In Table 2, CCES respondents who didn’t identify as Hispanic or Latino on the initial race question but said they were of Spanish, Latino, or Hispanic background are included with the Latino cohort. The exit poll does something similar, which is controversial in the sense that those who identify directly as Hispanic or Latino are more Democratic leaning than those who only could be identified as such via the origin or descent question. In 2016, 8% of exit poll respondents chose Hispanic as their race while 3% identified as being of Hispanic or Latino origin. Similarly, 6% of CCES respondents identified as Hispanic or Latino, but an additional 2% said they were of Hispanic or Latino origin. As the exit poll included both groups together, we did the same with the CCES data to get a more apples-to-apples comparison.

Tables 2 and 3: 2016 vote by race, exit poll vs. CCES

Notes: Latino demographic includes those who identified as being of Hispanic origin but not as Hispanic on initial race question. Columns and rows may not add up to 100 due to rounding.

Source: CNN exit poll data, CCES Common Content 2016

Even with the more inclusive definition of Hispanic/Latino, a notable difference between the exit poll and CCES data is the smaller share of Latino identifiers. But this distinction could be due to sampling issues (i.e. language barriers), so it’s probably best not to over-interpret it. In terms of electoral preferences, the CCES data offer further evidence that Latino voters may not have backed Clinton as strongly as many expected (though that is a debate that may last until the next election — or never end). However, it should be noted that the 2012 CCES found Barack Obama won Latinos 63%-36%, so Clinton’s margin in the 2016 CCES data represented a slight improvement for the Democrats.

Just as the CCES data didn’t show as sharp a gender gap, it also doesn’t show as wide a race gap. Trump’s margin among white voters was 14 points according to the CCES versus the 20-point margin the exit poll found for him. As Table 3 shows, the CCES found that Clinton won nonwhites 73%-22%, almost identical to the exit poll (74%-21%).

When looking at race and gender, as Table 4 shows, the CCES corroborates the exit poll finding that a gender gap exists across race and ethnicity. Regardless of racial and ethnic background, men were more likely to vote Republican and women were more likely to vote Democratic. The gender gap among whites was not as pronounced in the CCES, but it was still there. As for the other groups, the CCES and exit poll largely agreed.

Table 4: 2016 vote by race and gender, exit poll vs. CCES

Notes: Latino demographic includes those who identified as being of Hispanic origin but not as Hispanic on initial race question. Columns and rows may not add up to 100 due to rounding.

Source: CNN exit poll data, CCES Common Content 2016

Another key demographic trait is age. Race and education complicate voting by age, but it’s still clear that young voters were more likely than older voters to cast a ballot for Clinton. (Regarding CCES data on voting by race, age, and gender, Alexander Agadjanian put together a revealing breakdown for Decision Desk HQ.) But while the racial and gender gaps are more pronounced in the exit poll than in the CCES data, the age gap is sharper in the CCES data than in the exit poll. Tables 5 and 6 show age data with the familiar four-category age cohorts as well as a two-way split at age 45. In Table 6, the exit poll found that voters aged 18 to 44 backed Clinton by 14 points while those 45 and older supported Trump by eight (together, a 22-point gap). In the CCES data, Clinton won under-45s by 19 points and Trump won 45+ voters by 10 points (a 29-point gap). Regardless of the source, it’s very clear that younger voters — in part because they are far more racially diverse — tilted much more Democratic than their elders.

Tables 5 and 6: 2016 vote by age, exit poll vs. CCES

Note: Columns and rows may not add up to 100 due to rounding.

Source: CNN exit poll data, CCES Common Content 2016

The relationship between education and vote choice dominated much of the post-2016 election analysis, and with good reason. As Tables 7 and 8 indicate, voters with higher levels of education tended to vote more Democratic while those with lower levels of education tended to vote more Republican. The main story in these tables is the contrasting overall share of the electorate that had a college degree of some type (associate’s or bachelor’s degree). Whereas the exit poll found that 50% of voters claimed to be college graduates, the CCES data found 41% of voters had some type of college degree. This substantial difference may principally be a result of weighting (education was part of the CCES weighting process), but may also be a matter of how people defined themselves and the nature of the survey questionnaires. But overall, the CCES figures more credibly matched the adult population. According to the 2015 American Community Survey estimates, 42% of the adult population has a high school degree or less, 31% has some college experience or an associate’s degree, and 27% has a bachelor’s degree or higher. The corresponding categories for the electorate according to the CCES (not in the tables below) were 36% high school or less, 34% some college or associate’s degree, and 30% bachelor’s degree or more. Considering the fact that education is strongly correlated to turnout, it makes sense to expect the actual electorate to be more educated than the adult population. With that being said, the exit poll probably exaggerated the relative education of the 2016 electorate, and it has received criticism for doing so.

Tables 7 and 8: 2016 vote by education, exit poll vs. CCES

Notes: For CCES data, “Some college” includes those who had some college education but did not complete a degree. “College grad” includes all those who had an associate’s or bachelor’s degree. Columns and rows may not add up to 100 due to rounding.

Source: CNN exit poll data, CCES Common Content 2016

Lastly, vote by race and educational attainment, as well as different racial, gender, and educational combinations, are worth exploring. These groupings reveal a sizable difference in voting among whites based on education and a small difference among nonwhites. The CCES found that Clinton narrowly carried white college-educated voters (48%-45%), the reverse of what the exit poll found. In the end, it seems fair to say that white college voters split about evenly between the major-party nominees. This is notable because it represents an improvement for the Democrats compared to 2012. Conversely, non-college white voters overwhelmingly chose Trump. While the two sources disagree somewhat, the difference between college whites and non-college whites is about the same: the exit poll found Trump did 34 points better among non-college whites versus college whites (+37 vs. +3), and the CCES found he did 30 points better (+27 vs. -3). Age also played a role in this, with younger whites voting more Democratic and older whites voting more Republican, but with increased education generally shifting the partisan tilt toward the Democrats across all age groups. Both the exit poll and the CCES showed that nonwhite voters with a college degree were slightly more likely to vote Republican than those without a college degree.

Regarding gender, race, and education, the exit poll and the CCES found almost identical marks for white college women and all nonwhites, which formed a large part of the Democratic base. The numbers for white non-college women were also fairly similar. But the two sources were at odds over how white men voted; the CCES showed them to be less Republican-leaning than the exit poll. But while the CCES data showed white men and white women without a college degree to be less GOP-leaning than the exit poll, the CCES found them making up a larger share of the electorate, increasing their impact. Considering the razor-thin margins in key Rust Belt states such as Michigan, Pennsylvania, and Wisconsin (all states with sizable non-college white populations), it’s easy to see why this mattered.

Tables 9 and 10: 2016 vote by education, race, and gender, exit poll vs. CCES

Notes: Nonwhite demographic includes those who identified as being of Hispanic origin but not as Hispanic on initial race question. Columns and rows may not add up to 100 due to rounding.

Source: CNN exit poll data, CCES Common Content 2016

As we’ve just reviewed, there are some notable differences between the exit poll and the preliminary CCES data, yet they broadly agree on the major demographic developments that affected the 2016 presidential election. The gender gap continues to be a fact of life in American politics, and the gap showed up across most every racial and ethnic group. Once again, whites voted substantially Republican and nonwhites voted heavily Democratic, but less so than when Obama led the Democratic ticket. The young and old were broadly in political disagreement, with younger voters leaning toward Clinton and older voters leaning toward Trump. Non-college white voters were considerably more Republican and proved decisive for Trump in the Rust Belt. More highly-educated white voters moved toward the Democrats, but not enough to make the difference for Clinton.

Much more analysis of the 2016 contest between Clinton and Trump will come out in 2017 as additional data become available. For example, the 2016 American National Election Study is expected around the end of March. While the Crystal Ball is beginning to look ahead to the 2018 midterm elections, we agree that there is still much to be said about 2016. Our own in-depth studyTrumped: The 2016 Election That Broke All the Rules — is available for pre-order, and the book is expected to be fully released by mid-April.

Initial 2018 House Ratings

List of competitive districts filled with Democratic targets that will require a wave to flip

Kyle Kondik, Managing Editor, Sabato's Crystal Ball March 16th, 2017


Democrats have a path to winning a House majority next year, but that possibility is highly dependent on variables over which they have effectively no control. That’s the takeaway from our initial ratings of 2018’s House races, a list that is heavy on Republicans who start this cycle only mildly endangered.

Historically, the president’s party loses ground in a midterm: That’s what happened in 36 of the 39 midterms since the Civil War. American political history is dotted with elections where the president’s party suffered big losses because of a bad economy (1930, 1938, 1958, and 2010 are all examples), unpopular wars (1950, 1966, and 2006), scandal (1974), or other factors. If a wave is developing, we may be able to track it through President Trump’s approval rating (particularly if it falls into the 30s) and/or the House generic ballot. Watch to see if the latter metric, which measures national opinions on how voters intend to vote in their local House race, starts to show a significant Democratic lead approaching double digits in polling averages. Those are the kinds of numbers it’s probably going to take for Democrats to crack the GOP’s House majority, which is protected both by the power of incumbency and by favorable district lines in many key states. Those structural advantages can help Republicans next year, although they do not grant the party an impregnable firewall, particularly if the economy slows or the public reacts poorly to whatever the White House and Congress ultimately decide to do about health care.

Democrats only begin with a short list of real targets next year. These are the seats that start in the Toss-up and Leans Republican columns, seats held by incumbents who have won tough races but who also mostly started their House careers during Barack Obama’s presidency and thus have not had to run a race with a Republican in the White House. But Democrats have a much longer list of potential targets — the Likely Republican seats, which they could put into play if the national mood breaks in their favor (and if they produce credible challengers, which is something that is in their control).

Last year, voters elected 241 Republicans and 194 Democrats to the House. Since then, vacancies have reduced the House to a 237-193 Republican advantage. Assuming the incumbent party holds all five of the vacancies (more on that below), Democrats would need to net 24 seats to take a House majority.

Our initial ratings include just 62 of the 435 House seats (14% of the total seats) as competitive to at least some degree. Of those, 41 are held by Republicans, and 21 are held by Democrats — which means that the remaining 200 Republican seats and 173 Democratic seats not listed on our ratings seem safe for the incumbent party to start.

Of the 62 seats rated as something less than safe, about half of them (30) are rated in the Likely Republican column. These are seats that are just potentially competitive: In a neutral environment, very few, if any, would be in much danger. Meanwhile, just 11 Republican-held seats are in the more competitive Toss-up and Leans Republican categories. Sweeping these districts wouldn’t even get Democrats halfway to the 24-seat net gain they need. So they are going to need to expand the playing field, which is where those Likely Republican districts come in.

Table 1: Crystal Ball House ratings

Out of an abundance of caution, we have included every “crossover” district in our initial list of ratings. Those are the districts that voted for different parties for president and House last year. There were 35 of these districts: 23 Republicans hold seats that Hillary Clinton won, and 12 Democrats hold seats that Donald Trump captured. These crossover seats make up a majority of the competitive seats on both sides (23 of the 41 Republican-held districts listed and 12 of the 21 Democratic-held ones). We did an in-depth list and analysis of these seats last month, and there’s not much to add at this early point. However, one thing to watch is whether any of these seats become open. For instance, many believe Rep. Tim Walz (D, MN-1) is gearing up to run for governor instead of another term in his southern Minnesota district, which Trump won by 15 points after Barack Obama carried it by a point and a half four years ago. If this seat is open, Republicans probably would have even odds to win it, but with Walz running the Democrats are favored to hold it. Another potential gubernatorial candidate is Rep. Kevin Yoder (R, KS-3), whose suburban Kansas City district voted for Clinton by a point after Mitt Romney had won it by 10 points in 2012. His district would also be a Toss-up if it became open, but he’s the favorite if he stays put.

Overall, retirements are an important metric to watch because they can be more susceptible to takeover by the other party.

Speaking of, and as noted above, there are five vacant House seats right now: Democrats should easily hold the Los Angeles-area CA-34, and Republicans should easily hold KS-4 and SC-5. Montana’s open at-large district is hypothetically competitive — Democrats can compete in the Republican-leaning state — although Democrats appear more focused on the other special election in GA-6, a suburban Atlanta seat that voted Republican for president by 20 points or more in recent years until Trump only won it by a point and a half last year. We rate the MT-AL and GA-6 specials as Likely Republican and the others as safe for the incumbent party.

Both the National Republican Congressional Committee and the Democratic Congressional Campaign Committee have released their initial target lists, and many of the districts they include do not make the cut as being competitive on our ratings. For Republicans, we’re skeptical that they can truly make inroads against some Democratic incumbents, such as Reps. Scott Peters (D, CA-52) and Raul Ruiz (D, CA-36), whose districts voted more Democratic for president in 2016 than 2012. Meanwhile, Democrats argue that with the right candidates they can put several Trump-friendly heartland districts into play, such as those held by Reps. Andy Barr (R, KY-6), Mike Bishop (R, MI-8), Bob Gibbs (R, OH-7), and Alex Mooney (R, WV-2). That seems like a significant stretch, but perhaps one or more of these districts makes it onto our competitive list depending on the emergence of strong candidates and the national environment. For what it’s worth, Democrats say that they are seeing considerable interest from candidates in a wide range of districts who are inspired to run by the Trump presidency. Whether those candidates actually run, and whether they can perform in red districts, is an open question, although Democrats did have success in some GOP-leaning districts in their 2006 midterm victory.


While the overall environment will dictate a lot of what happens next year, developments in individual districts and states matter, too. For instance, Texas might have to draw a new congressional map for next year: A federal court ruled late last week that the Republican-controlled Texas state legislature had improperly diluted the voting power of minorities when they drew the state’s congressional map after the 2010 census. If the map is eventually redrawn, Rep. Will Hurd (R, TX-23), a narrow winner in both 2014 and 2016, could find his district even harder to defend, and a safe seat or two could also be imperiled, like Rep. Blake Farenthold’s (R) TX-27. Court-ordered, mid-decade redistricting helped Democrats in Florida and Virginia last cycle; in fact, one could argue that redistricting was responsible for half of the party’s modest six-seat net gain last year. But it’s premature to factor a Texas remap into these ratings, which at the moment reflect the current districts.

Watch for retirements or for members who decide to run for other offices, particularly in districts that are already rated as something less than safe for the incumbent party. But also remember that filing deadlines for primaries are a year or more away in many places, and even after those pass there can be mechanisms to replace already-nominated candidates on the ballot. If things go south for either party a year from now, it’s possible that some endangered members could just decide to pass on running again.

Overall, though, voters’ perceptions of Trump and congressional Republicans will loom large next year — or at least history suggests those factors will be important. If perceptions are neutral or broadly positive, the GOP should have little trouble keeping the House. If they are negative, the House will be in play, and some of those Likely Republican districts — the districts that truly will make or break the GOP House majority — might start to slip away.

How Midterms Do (and Do Not) Differ From Presidential Elections

What recent history tells us about the likely size and makeup of next year’s electorate

Geoffrey Skelley and Kyle Kondik, UVA Center for Politics March 2nd, 2017


Editor’s Note: The Crystal Ball is taking off next week for the University of Virginia’s spring recess. We’ll be back on Thursday, March 16.
— The Editors

With politicos everywhere turning their eyes to the still-distant 2018 midterm election, we thought it would be useful to review some of the basic differences and similarities between the electorates in presidential and midterm cycles. Basically, midterm electorates are smaller, older, and less diverse than presidential ones, but the demographic voting patterns and divisions that we see in midterms are quite similar to presidential contests. What follows is a look at the similarities and differences between the two kinds of national electorates. For the most part, this analysis is based on exit poll data: We used the national exit poll data for the presidential race in presidential years and the national exit poll data for the national House vote in midterm years.


1. Turnout is always lower in midterm elections

The most fundamental difference between presidential and midterm cycles is that far fewer voters participate when there is no presidential contest. According to available data, the last time midterm turnout exceeded the previous presidential election was in 1838, when 70.8% of the eligible population voted versus the 56.5% who voted in the 1836 presidential election. Ever since, presidential turnout has always exceeded midterm turnout. Since the 26th Amendment to the Constitution expanded the franchise to all eligible citizens 18 years or older, 1994-1996 was the closest midterm turnout has been to presidential turnout — 41.1% of the voting-eligible population (VEP) cast ballots in the 1994 midterm and just 51.7% of the VEP voted in the 1996 presidential contest. Chart 1 and Table 1 lay out the substantial differences in turnout that depend on whether or not the presidency is on the ballot.

On average, midterm turnout has been about 17 points lower than presidential turnout in the 26th Amendment era. If that holds, we’d expect a midterm turnout in the low 40s next year given 2016’s 59.3% VEP turnout, but of course that may vary.

Chart 1 and Table 1: Voter turnout among the voting-eligible population (VEP), 1972-2016

Source: United States Elections Project

2. The midterm electorate is more educated

With lower overall turnout in midterm elections, groups that are more likely to vote thus exude a larger influence on the outcome. One demographic trait that is highly correlated with turnout is education, though there is debate over a causal linkage between higher education levels and higher voting rates. Chart 2 lays out exit poll data going back to the 2004 presidential election. The midterm election that followed each presidential cycle had a larger share of voters who said they were college graduates. And in the ensuing presidential election, the electorate had a slight decrease in the percentage of college graduates. This pattern endured even as the college-educated share of the population grew.

It is worth noting that exit polls may overstate the education level of the electorate, as the New York Times’ Nate Cohn has argued using other, non-exit poll sources. But since we’re comparing exit polls over time, and not comparing them to other sources, we feel confident in saying that the electorate’s education level tends to be a little higher in midterms even if exit polls paint a picture of an electorate that is more educated than it is in reality.

Chart 2: Exit poll shares of electorate that are college graduates, 2004-2016

Sources: Exit polls, 2004-2016

3. Midterm electorates are less racially diverse, though they too are becoming more diverse over time

Broadly speaking, nonwhite voters are less likely to cast ballots than white voters, whether in presidential or midterm elections. There had been a recent shift in presidential elections in this area because of increased black turnout rates in 2008 and 2012. But the 2016 exit poll and other data suggest that African-American turnout rates dipped again in 2016, leading to the reasonable conclusion that the historic candidacy of Barack Obama led to higher black participation.

Regardless of the presidential cycle differences, there’s little question that nonwhite voters tend to make up a comparatively smaller share of the midterm electorate than the presidential electorate. Chart 3 shows the data for the nonwhite share of recent presidential and midterm electorates. In every midterm cycle, the nonwhite portion falls, though there continues to be an overall increase as the American population becomes increasingly diverse. According to the 2015 American Community Survey estimates, the non-Hispanic white percentage share of the population of those 18-44 years old is 57% versus 73% for those 45 or older. And the under-18 population, which will age into the electorate, is 52% non-Hispanic white. Growing racial and ethnic diversity is a fact of American life, but for the moment we can expect midterm elections to still feature a smaller share of nonwhite voters relative to presidential elections.

This is hypothetically bad for Democrats given that they perform much better with nonwhite voters than white voters (which we explore more below). However, the political environment usually matters more than demographics: Democrats performed very well with a less diverse electorate in 2006, while Republicans won big victories in 2010 and 2014 despite the electorate becoming more diverse. An unpopular President George W. Bush in 2006 and an unpopular President Obama in 2010 and 2014 mattered more to the overall outcome than the electorate’s demographic makeup.

Chart 3: Exit poll nonwhite share of the electorate, 2004-2016

Sources: Exit polls, 2004-2016


But while the demographic makeup of midterm electorates differs at least slightly from that of presidential elections, the voting patterns of those who show up for both kinds of elections are quite similar. Here’s three ways that’s so:

1. There is always a gender gap

As we’ve noted before, there’s been a persistent gender gap in American politics for nearly four decades. Basically, women are more Democratic than men, and men are more Republican than women. That’s not just true in presidential elections — it has also held for the lion’s share of other recent statewide races, too.

Chart 4 shows the Democratic margin among men and women for each presidential and midterm election since 2004. Notice how women are more Democratic than men (and men more Republican than women) in every election.

Of course, the margins change from election to election, to the point where if a Democrat is winning men, that likely means there is a Democratic wave, as we saw in 2006 and 2008. Meanwhile, if a Republican is winning women, that’s likely evidence of a GOP wave, as we saw in 2010. And if men and women are voting in mirror opposite ways, like we saw in 2016, it probably means a very close election.

One other thing: Women make up a slightly larger share of the population, and they also make up a larger share of the electorate. Over the last seven national elections, the average makeup of the electorate has been about 48% men and 52% women.

Chart 4: Exit poll gender gap in the electorate, 2004-2016

Sources: Exit polls, 2004-2016

2. In recent elections, younger voters have been more Democratic while older voters have been more Republican

It has not always been the case that older voters are more Republican and younger voters are more Democratic. As recently as the 2000 election, exit polls did not find much difference among age groups in the close Al Gore vs. George W. Bush contest. Interestingly, though, Gore’s best group was actually voters aged 60 years or older (although he only won the group by four). Many voters age 60 or older in 2000 grew up in the time of the Great Depression and more often than not developed an affinity for Franklin Roosevelt and the Democrats, a party ID that endured in many cases. In the mid-1990s, the Pew Research Center found that younger voters were likelier to lean Republican, and older voters were likelier to lean Democratic. (Note that the exit poll shifted to 45-64 and 65+ for its two older age categories in its four-way age vote summary after the 2006 election, from 45-59 and 60+.)

However, in more recent elections, that dynamic has flipped, and generally speaking, the older a person is, the likelier they are to vote Republican (and the younger they are, the likelier they are to vote Democratic). This is a dynamic we’ve seen in both presidential and midterm elections since 2004, as shown in Chart 5.

One big difference between midterm and presidential electorates is that the midterm electorate skews older. On average, about 54% of the electorate was aged 45 or older in the most recent presidential elections, while an average of 64% of the midterm voters — nearly two-thirds — were 45 or older. Overall, 53% of the adult population is 45 or older, so older voters are significantly overrepresented in midterm elections. That’s a generic advantage for Republicans in midterms although, again, the overall political environment next year may be more important than the age of the participants.

Also note that there is a lot of variation in voting by age group even within the last dozen years. In 2004, voters 30 or older all voted somewhat similarly while the youngest voters were more clearly Democratic. As time went on, the 45-64 group and the 65+ group separated and then re-converged in 2016, while the difference between the 18-29 and 30-44 groups may be becoming less pronounced. That may be a result of the two younger age cohorts being similarly diverse (56% non-Hispanic white for 18-29s and 58% for 30-44s), while the two older age cohorts are much whiter (70% non-Hispanic white for 45-64s and 79% for 65+).

Chart 5: Net Democratic vote by age group, 2004-2016

Sources: Exit polls, 2004-2016

3. Racial divide defines both midterm and presidential years

While midterm electorates are not as diverse as presidential ones, nonwhites vote substantially more Democratic than white voters in both kinds of elections. Chart 6 shows the difference.

Notice the midterm difference between 2006, when Democrats lost by just four among whites, to 2010 and 2014, when Democrats lost by more than 20 points in each. While the 2018 midterm electorate should be the most diverse ever because of broader demographic changes, roughly three-quarters of the voters next year likely will be white. If Democrats are going to make big gains, they almost certainly will have to lose white voters by less than 20 points — in other words, a narrower gap than they achieved with whites in any of the last four national elections, according to exit polls.

Chart 6: Net Democratic vote for whites and nonwhites, 2004-2016

Note: For 2004-2014, overall nonwhite vote estimated based on size and vote of Black, Latino, Asian, and Other race exit poll categories.

Sources: Exit polls, 2004-2016


Midterm electorates are older, whiter, and more educated than presidential electorates. The first two characteristics make midterm electorates, on paper, better for Republicans. The last characteristic, education, might be helpful to Democrats given Donald Trump’s struggles relative to other Republicans among college-educated white voters (he barely won that typically Republican-friendly demographic). The voting patterns of midterm voters are similar to presidential voters, though: Democrats do better with women as well as younger and more diverse voters, while Republicans do better with men, older voters, and whites overall.

Meanwhile, the number of total voters will be significantly smaller next year than in 2016: Turnout will be much closer to 40% than to 60%. Who shows up is vital. Often, the party that does not hold the White House turns out more strongly in the midterm, leading to gains for the out party. But Republican strength among the demographics that typically make up a bigger share of midterm voters — older whites — presents Democrats with a challenge next year.