Hillary Clinton’s loss in 2016 was at least partly attributable to the millions of Barack Obama voters who shifted their support to Donald Trump. Four years later, it is now Trump who must worry about losing some of his supporters in November: As many as one in 10 Trump voters is considering voting for somebody else in 2020, according to our analysis of data from the 2019 Cooperative Congressional Election Study survey. 

These voters are individuals who neither strongly approve nor strongly disapprove of the job Trump is doing as president. If Democrats want to win over these voters, they’ll have to choose their message wisely.  

Why should campaigns focus on persuading these potential swing voters rather than turning out nonvoters? Swing voters might be rare but they do exist, and persuading them can be particularly influential for two reasons. First, switching a person from opposing your side to supporting it has a two-vote effect on the margin: It subtracts one vote from your opponent and adds one vote to your tally. Mobilizing an additional supporter to come out to vote has only half that impact.

Second, mobilizing people to vote when they haven’t done so before can be challenging. Even if Democrats can mobilize more nonvoters to get to the polls, it is not clear that this will help them in swing states like Wisconsin and Michigan, where many nonvoters fit the profile of Trump’s base.  

The Trump voters who might change their minds are mostly under 40

The dataset we used, a large academic election survey that YouGov has conducted online every year since 2006, allows us to better understand the share of 2016 Trump voters who are up for grabs. In November 2019, the survey interviewed 18,000 American adults who had been interviewed in 2016 about their vote, asking them how they planned to vote in 2020. 

They found that most of Trump’s voters plan to stick with him: Ninety percent of those who voted for Trump in 2016 say they plan to vote for him again. But 10 percent seem to be up for grabs. Four percent are already planning to vote for the Democratic candidate, and another 6 percent say that they are still undecided. By contrast, 94 percent of Clinton voters are already committed to the Democrats: just 2 percent of Clinton’s 2016 voters are planning to vote for Trump in 2020, with another 4 percent undecided. 

The largest share of Trump’s support in 2016 came from baby boomers, and only 7 percent of them are considering abandoning him in 2020. Trump’s older supporters are quite loyal to him. Where Trump is losing the most support is among Americans younger than 40 who voted for him in 2016.

Younger voters are less certain, and as many as 20 percent of Trump’s younger voters are considering abandoning him: 9 percent of the millennial and Gen Z voters who supported Trump in 2016 are now planning to vote Democratic in 2020, and another 11 percent say they are not sure how they will vote.

While the persuadable Trump voters are mostly distinguished by their age, they are also more likely to have college degrees and are more likely to be women when compared to the Trump base. They are also much more likely to identify as politically moderate. 

Considering that Trump only received about one-third of the vote among millennial and Gen Z voters in 2016, the fact that these voters are at least considering voting for a Democrat in 2020 is important. This erosion of support among younger Republicans continues a trend that has been taking hold since 2016, when many young Mitt Romney voters decided not to vote for Trump. This low support could sink even lower, chipping at the narrow margins we typically see in elections. 

Potential swing voters are less conservative than Trump’s base

But just because these Trump voters say they are persuadable now does not mean that Democrats will necessarily win them over. After all, the fact that these voters supported Trump in the first place means they are likely more ideologically conservative than the typical Democratic voter. 

The graph below shows how the 2016 Trump voters who are now unsure of how they’ll vote in 2020 compare to the Trump base (those who already say they’ll back him again in 2020), the Democratic base (Clinton voters who already plan to vote for the Democrat in November), and Trump voters who have already decided to vote for the Democrat. 

On several issues, the persuadable Trump voters do appear to have views that fit well with the Democratic agenda. For example, about two-thirds of persuadable Trump voters want a ban on discrimination based on sexual identity and oppose Trump’s decision to leave the Paris climate agreement, putting them at odds with the voters in Trump’s base. 

On health care, persuadable Trump voters are to the left of Trump’s base but less liberal than some of the most progressive Democrats in the field. Fewer than half support Sen. Bernie Sanders’s single-payer Medicare-for-all approach, which would ban private health insurance, but more than 70 percent support a public option for health insurance that would still allow people the option to keep their private insurance.

On immigration, persuadable Trump voters appear to support the trade-off that used to be the base of comprehensive immigration reform proposals: increasing border security while giving undocumented immigrants already in the US a path to citizenship. 

Breaking with Trump’s dedicated voters, this group overwhelmingly supports granting legal status to undocumented immigrants who have held jobs and paid taxes for at least three years. But they’re closer to the viewpoint of the Trump base when it comes to border security, with just 13 percent opposing an increase in the number of patrols on the US-Mexico border. A slight majority — 55 percent — of the Democratic base would oppose increasing border security. 

Finally, the persuadable Trump voters are not quite as convinced as the Democratic base when it comes to increasing the minimum wage to $15 per hour. While this policy is supported by the vast majority of core Democratic voters, only about 50 percent of the persuadable Trump voters are on board. 

Of course, the 2020 election can turn on any number of factors, including the likability of the candidates, the potential impact (economic or otherwise) of the coronavirus, and how these persuadable voters evaluate Trump’s first term. But to maximize their chances of winning over the persuadable Trump voters, Democrats should focus on issues that appeal to their base like climate change and LGBTQ rights, while favoring more moderate solutions on issues like health care and immigration.

There may be other paths to defeating Trump, including mobilizing progressives who don’t usually vote. But if the strategy is to win over persuadable Trump voters, then a moderate nominee like Joe Biden likely has a better chance at appealing to this small but pivotal group — who are not only important in 2020 but will also shape election outcomes for many years to come. 

Brian Schaffner is the Newhouse Professor of Civic Studies at Tufts University. Laurel Bliss is a research associate at Tufts.

The expansion of early voting has not only made it more convenient for Americans to cast their ballots, but it has also become a tempting source of data to try to predict what might happen once the dust has settled on election night. In 2016, our analysis of the early vote was fairly predictive of turnout differences that helped define the election outcome —namely, while Latino participation surged, so too did turnout among white voters, while black voters turned out at a lower rate than they had four years earlier.

This year, our reading of the early vote suggests a somewhat similar story: Minority turnout will be higher than usual for a midterm election, but the same is true for white turnout. That means 2018 will see a continuation of the trend toward a more racially diverse electorate, but the pace of that trend is not sufficient to power a “blue wave” on its own.

As is well-documented at this point, early voting is up substantially from the last midterm election, but making sense of those trends requires careful attention to context. More and more Americans who decide to vote end up casting their ballot early, so we would expect that the number of early voters will rise even if overall turnout remains the same. A greater number of early voters in 2018 than the previous midterm election (2014) does not, on its own, suggest anything about which party is going to do better this year.

Looking at the demographics of who votes early — such as race/ethnicity — might be a better marker, given the close connection between race and party. If black and Latino early voting is increasing more than white early voting, for example, that might suggest higher overall turnout for those groups and that Democratic candidates will do well on Election Day too.

To draw conclusions about the demographics of who is voting early, we take advantage of official records of who voted as provided by the voter file firm Catalist. As in previous years, Catalist provides detailed information about who has already voted in the November 2018 election and a comparison to the same point in time in 2012, 2014, and 2016. We combine Catalist’s probabilistic estimates of the race/ethnicity of these early voters with state-level figures on the population eligible to vote, using data from the Census Bureau’s American Communities Survey (ACS) and Population Estimates Program (PEP) to project the citizen voting-age population (CVAP) in November 2012, 2014, 2016, and 2018.

The first graphic shows the percentage of black, Latinos, and white people who have voted early so far in 2018 compared to what percentage had voted early at this same stage in previous election cycles. As we can see, the percentage who have already cast their ballots is up in every state, and for all groups, compared to 2014. Yet early turnout is only rarely higher than 2012, and never exceeds the rate for 2016.

2018 may feature higher turnout than a typical midterm election, but turnout will almost certainly not be as high as it was in recent presidential elections. Furthermore, there are no clear patterns suggesting especially large gains by any racial/ethnic group, compared to the others. The lines in the graphic generally move in parallel across the election cycles.

That said, there are some exceptions to this general rule. One that might work in favor of Democrats is in Georgia. While early voting has dropped among all three groups compared to this point in 2016, the drop among white voters is greater than it is for black and Latino voters. The relatively strong turnout patterns for people of color in Georgia may bode well for the gubernatorial campaign of Stacey Abrams; yet even here, we do not see higher rates of early voting for black people than for white people, like we did in 2012.

In Texas, Latino early voting is only down slightly from where it was at this time in 2016, whereas early voting rates are down more among white and black voters. Beto O’Rourke’s chances of pulling an upset over Sen. Ted Cruz will certainly depend on getting higher-than-expected turnout from both Latinos and African Americans, so this is mixed news for that contest. A growing Latino population helps O’Rourke, but if Latino turnout remains low, that population growth won’t be enough to turn Texas purple.

 Of course, one thing campaigns will want to know is how many of these voters are new to the electorate in 2018 and how many are people who vote anyway but are perhaps voting early this year rather than waiting until Election Day. Our second chart delves into this question by looking at the percentage of each group’s early voters who are voting for the first time in 2018.

This chart helps illustrate one of the patterns we highlighted in 2016 — Latino turnout surged in that election, and much of that surge was evident in the early vote. For example, more than 15 percent of Latinos who voted early in Florida and Texas in 2016 were first-time voters.

Once again, a higher share of Latino early voters in 2018 are first-time voters compared to black and white voters. In Texas, Nevada, Georgia, and Arizona, we see that the share of first-time voters who are Latino is much higher than in 2014, but not as high as 2016. Again, mixed news but no clear evidence that the early voting we see for minorities guarantees a “blue wave.”

Our final chart shows each group’s share of the early vote compared to their size among all citizen voting-age residents. This metric indicates whether a group is “overperforming” or “underperforming” in early voting relative to what would happen if all groups voted at the same rate. In each state, whites make up a much larger share of early voters than they do of the population as a whole. But in some states, that gap is bigger than others. Georgia and North Carolina are both states where participation approaches (though does not achieve) equality. In other states, the gaps are much larger, particularly when we compare white and Latino voters.

Overall, expectations of a radically transformed electorate in 2018 are almost certainly overstated, at least when it comes to the racial composition of those who will vote. There are some key states (like Georgia and Texas) where turnout among racial minorities is relatively strong, particularly for a midterm election. But there are others (like California and Nevada) where — relative to white participation — minority turnout looks much like it did in 2014, a wave year for Republicans.

There may still be a “blue wave” on the horizon, but it will have to come via an electorate that will not be dramatically more diverse than it was when Donald Trump was elected.

Bernard L. Fraga is an assistant professor of political science at Indiana University. Find him on Twitter @blfraga.

Brian F. Schaffner is the Newhouse professor of civic studies at Tufts University. Find him on Twitter @b_schaffner.

As he promoted his new book, Sen. Jeff Flake (R-AZ) made news last week by calling for more civility from politicians and especially President Trump. He said one of the major concerns with the declining civility in Washington is the example it sets for the public in general. As he put it on Face the Nation Sunday, “We have to model behavior that we’d be proud that our kids are watching.”

But to what extent does the tone in Washington actually appear to affect how the public discusses politics? In a new project, we have been working to track and understand incivility by examining the extent to which users post offensive comments on Reddit. Reddit has received a great deal of attention from political analysts in recent years due to the fact that it has a robust offering of political discussion boards. In fact, Reddit has become an important enough player in online political discussions that President Obama and a host of 2016 presidential candidates took to the forum to engage with its users.

We examined every political comment posted on Reddit from January 2015 to January 2017, more than 39 million comments in total. Because it would be impossible to code millions of comments by hand, we instead developed an algorithm to accurately identify “offensive” comments. Among other things, the algorithm tagged posts as offensive when they used profanity, compared someone to Hitler or a pedophile, engaged in name calling, and so on.

We tested the accuracy of our algorithm on a data set of 14,500 tweets that were coded by humans, and our automated approach correctly classified 90 percent of the tweets based on whether or not human coders would have found them offensive. In testing, we have found that when the classifier does make errors, it is more likely to underestimate the presence of offensive speech. Thus, our results can be considered conservative estimates of the presence of offensive speech on Reddit.

Posts became more offensive during the general election campaign

The figure below shows the percentage of Reddit comments that were classified as “offensive” during each week of 2015 and 2016. The blue dots represent political comments posted to Reddit, while the red dots are from a sample of nonpolitical comments. The first thing to note is that roughly 8 percent of nonpolitical comments are classified as offensive, and that this level of offensiveness is remarkably consistent over the entire period that we examine. 

By contrast, the offensiveness of political comments posted to Reddit varied much more over the two-year period. Interestingly, political comments became somewhat less offensive from early 2015 through the presidential primary period of 2016. It seems the focus on the primary competitions softened the tone of discourse, since intraparty battles are generally less hostile than cross-partisan ones. The one major departure from this trend happened during the week when Trump announced his candidacy for president in a speech that featured a highly offensive categorization of Mexican immigrants.

While political comments on Reddit were no more offensive than nonpolitical Reddit comments during most of the presidential primary period, this clearly changed once the general election campaign kicked off in full. Indeed, we see that the share of offensive comments spiked during the party conventions and continued to increase as the general election campaign was in full swing from August through November 2016.  In early October, we see a large single-week spike in offensive comments; this coincides with the week during which the Access Hollywood tapes were released. During the week of and following Election Day, we see similar levels of offensiveness — during these two weeks, nearly one of every 10 political comments posted on Reddit was classified as offensive.

Offensive posts are more popular

Perhaps just as striking as the increase in offensiveness during the general election campaign is the fact that in the two months following the election, offensive speech on political Reddit remained quite high. In fact, after the election, political comments on Reddit were about 20 percent more likely to be offensive than those posted about other topics.

Why are offensive comments so common on Reddit? One reason suggested by our data is that such comments are typically more popular among the Reddit community. To gauge popularity, we calculated the net score for each comment, which is simply the number of “upvotes” that a comment received minus the number of “downvotes.” Offensive comments posted to Reddit receive an average score of 8.9 versus 6.7 for non-offensive comments. Thus, the Reddit community provides substantial positive reinforcement when its users post offensive content.

How important are these patterns we’ve uncovered? After all, Reddit is a forum that attracts just a small proportion of the population who are not very representative of Americans in general. According to a Pew survey, just 4 percent of Americans use Reddit. Reddit users are more likely to be younger, male, and more liberal in their political leanings compared to the average American.

Nevertheless, even with these caveats, our findings have important implications for our politics. Research has shown that contentious political discussions tend to alienate moderates from the political process and make it so that people view opposing viewpoints as illegitimate. Millions of Americans are exposed to political discussions on Reddit every week, and while these individuals are not representative of ordinary Americans, many are political moderates. In fact, the same Pew survey found that people who report learning about political news on Reddit are just as likely to identify as political moderates as the general population.

What’s most telling is that our research shows that the tone of discussions on Reddit seems to respond to how our politicians are behaving. Thus, Flake appears to be correct when he worries about the effect that incivility in Washington may be having on political discourse more generally.

Rishab Nithyanand is a postdoctoral researcher in the College of Information and Computer Science at the University of Massachusetts Amherst. He obtained his PhD from Stony Brook University in 2017.

Brian Schaffner is a professor of political science at the University of Massachusetts Amherst. Find him on Twitter @b_schaffner.

Phillipa Gill is an assistant professor of computer science at the University of Massachusetts Amherst. Find her on Twitter @phillipa_gill.

This post is part of Mischiefs of Faction, an independent political science blog featuring reflections on the party system. See more Mischiefs of Faction posts here.

In the wake of Donald Trump’s victory over Hillary Clinton last week, many are still struggling to understand what happened. Explanations of Trump’s support have largely been driven by exit polls, which often identify “working-class whites” or “non-college-educated whites” as the significant swing group that help propel Trump to victory. But while the exit polls can help us identify which groups were crucial, they are much less helpful for understanding why these groups voted the way they did.

This is largely because exit polls fail to ask questions that measure the key concepts that may have been in play in this election. As a result, we are left with competing narratives, with some reports suggesting that economic insecurity was the decisive factor in this election, and others highlighting the role of racism or sexism in driving voters toward Trump. The truth, however, is that there is no single cause of Trump’s success among whites. All three factors played an important role.

To demonstrate that this is the case, I draw on a nationally representative survey designed by Matthew MacWilliams, Tatishe Nteta, and myself and fielded by YouGov two weeks before the election (October 26 to 31). Fortunately, that survey includes the kinds of questions needed to measure the competing narratives about what drove the white vote.

To capture economic security, I use a question that asked individuals how satisfied they were with their own economic situation. For sexism, I used an item taken from a well-established battery of questions designed to measure the concept. The question asks respondents whether they agree or disagree that “women seek to gain power by getting control over men.” And to capture racism, I use an item taken from a battery that measures the extent to which an individual acknowledges the existence of racism. This particular item asks respondents to indicate whether they agree or disagree that “[w]hite people in the US have certain advantages because of the color of their skin.”

The first chart below shows how whites responded to these questions in our survey. Note that approximately one in four white Americans were “not at all satisfied” with their personal economic situation. Such economic dissatisfaction is presumed by many to be the chief explanation for the Trump vote. But there is also a significant percentage of white Americans who demonstrated a denial of racism on our survey – more than 40 percent disagree that white people have advantages in the US because of their skin color. About 40 percent of white Americans were also either in agreement or neutral in rating the statement that “women seek to gain power by getting control over men.”

To get a sense of the relative importance of each of these factors in explaining the vote for Trump, I estimated a statistical model of these white respondents’ vote choices. In the model, I included standard demographic and political control variables, such as gender, age, education, income, ideology, and party identification. I then added all three questions described above into the model at the same time. This allows us to see, for example, whether views about racism are still important when one accounts for a respondent’s reported levels of economic dissatisfaction.

To show the comparative importance of each of these three measures, I plot the relationships between the items and predicted support for Trump in the graph below. These graphs show how changing an average white respondent’s answer on each of these questions would affect his probability of voting for Trump rather than Clinton.

It is clear from this graph that there is no simple single explanation for why whites supported Trump over Clinton. Indeed, each of these three explanations appears to be important in explaining the vote. Dissatisfaction with one’s own economic situation leads to a sizable increase in the probability of supporting Trump, but so too does increasing levels of denial that racism exists in America.

Sexism is also a powerful force, as those who are more in agreement that women seek to gain control over men were far more likely to support Trump. Notably, the impact of sexism appears to be unique to the 2016 election, as I have shown in an analysis of other polling data.

Ultimately, the competing narratives about why Trump performed so well among whites are not competing at all; they are complementary. To truly understand Trump’s success means acknowledging that economic insecurity was part of the story, but so too were racism and sexism. And in truth, it likely took all three factors to allow him to edge past Clinton for a narrow victory.

Brian Schaffner is a professor in the department of political science at the University of Massachusetts Amherst and a faculty associate at the Institute for Quantitative Social Science at Harvard University. His research focuses on public opinion, campaigns and elections, political parties, and legislative politics. He is the co-author of the book Campaign Finance and Political Polarization: When Purists Prevail, co-editor of the book Winning with Words: The Origins & Impact of Political Framing, co-author of Understanding Political Science Research Methods: The Challenge of Inference, and author of Politics, Parties and Elections in America (seventh edition). His research has appeared in more than 30 journal articles and has received more than $2 million in external funding.

Schaffner is also the founding director of the UMass Poll and a co-PI for the Cooperative Congressional Election Study.