Sabato's Crystal Ball

2018 Senate: The Democrats Are Very Exposed

But potentially counteracting forces -- partisan polarization & the negative midterm effect on the president’s party -- complicate the midterm picture

Kyle Kondik and Geoffrey Skelley, UVA Center for Politics December 8th, 2016


A potential silver lining for Democrats is that they head into the 2018 midterm as the party that does not hold the White House, and the “out” party typically makes gains down the ballot in midterms. But it will be difficult for Democrats to make Senate gains in 2018: Despite being in the minority, they face a near-historic level of exposure in the group of Senate seats being contested in two years, Senate Class 1.

It’s hard to overstate how disappointing 2016 was for Democrats in the Senate. Yes, the party did net an extra two seats by defeating Republican incumbents in Illinois and New Hampshire despite Hillary Clinton losing her bid for the presidency, so the next Senate will be 52-48 Republican. But given that for most of the cycle it looked like Clinton would win the White House and also deliver the Senate, the Democrats clearly did not realize their potential this year.

One of the big reasons why the Senate majority appeared in reach for the Democrats in 2016 was that the Republicans were, and still are, overexposed in Senate Class 3, the 34 seats that were up for election this past November. Republicans controlled 24 of 34 seats on a map where they had made substantial gains the last two times it had been contested, the GOP wave year of 2010 (when the Republicans gained six seats) and President George W. Bush’s reelection in 2004 (when they gained four). But the GOP largely held the line and now hold a 22-12 advantage in this Senate class, which won’t be up for election again until 2022, which could be President Donald Trump’s second midterm election (although that’s of course a very long way off).

Always looming over 2016, though, was the 2018 map. Including the two independents who caucus with the Democrats, the party holds 25 of the Class I Senate seats that are up for election in 2018, while the Republicans hold only eight. Again, a look back at the last few times this group of seats was contested explains the Democrats’ exposure. After Republicans netted eight seats on this map in the 1994 Republican Revolution (and party switches by Sens. Richard Shelby of Alabama and Ben Nighthorse Campbell of Colorado from Democrat to Republican would essentially make it 10 by the time of Campbell’s switch in March 1995), Democrats made big gains in Class 1 in both 2000 (four) and 2006 (six). Going into 2012, it appeared that Democrats would lose seats, but they upset expectations and instead gained two, which is why they are so overextended now.

Map 1: Senate seats up in 2018 midterm election (Senate Class 1)

Only three times before in the era of popularly-elected senators has a party begun a midterm cycle as exposed as the Democrats are in 2018, and only once since World War II. When we say “begun,” we mean the partisan makeup of the class up for election in December about two years prior to Election Day — where we are in the calendar right now.[*]

In December 1924, Republicans held 25 seats that were up in the 1926 midterm election, and that party makeup remained the same up to Election Day. In November 1926, the GOP lost six of the regularly-scheduled contests, as well as a concurrent special election in Massachusetts, with the Democrats defeating seven Republican incumbents (including an appointed one in the Bay State) to make their gains. The Democrats’ success in 1926 would eventually play a part in the record-high exposure for a party in any Senate cycle: In 1938, Democrats had to defend nearly 30 seats, the result of the party’s banner 1926 and 1932 cycles (they gained 11 total net seats in the latter). Following the appointment of a Democrat to fill a vacancy in a formerly GOP-held seat in South Dakota on Dec. 29, 1936, Democrats held 28 seats that were up in the 1938 midterm election. By the time Election Day 1938 rolled around, they were also defending a seat in Oregon, putting them at 29 seats out of the 32 that were regularly up. (Remember, this was before Alaska and Hawaii joined the Union, thus there were 96 senators and 32 in each of the Senate’s three classes.) The outcome was a Republican net gain of eight seats, with Democrats losing seven of the regularly-scheduled seats they held, as well as a concurrent special election in New Jersey. Five Democratic incumbents fell by the wayside in the 1938 general election. Republicans also picked up 81 seats in the House in their 1938 wave, a midterm shellacking that some argue effectively ended President Franklin Roosevelt’s New Deal.

The last time a party was as exposed as Democrats are in 2018 was in the 1970 cycle. At the end of 1968, 25 Democratic-held seats were up in the 1970 midterm. There are some similarities between the position of Democrats in 1970 and 2018. First, Class I Senate seats were up in 1970, just as they are in 2018. Second, a sizable number of Democratic-held 1970 Senate seats (13) were up in states that Republican Richard Nixon had just carried in the closely-contested 1968 presidential election, compared to the 10 Democrats are defending in 2018 in Trump states. Perhaps endangered Democrats up in 2018 can feel a little bolstered by the fact that Democrats only lost three net seats in the 1970 midterm despite having to defend numerous seats, many in states that backed the most recent GOP presidential nominee. Overall, 11 of the 12 Democratic incumbents running in states won by Nixon in 1968 won reelection in 1970 (though Harry Byrd Jr. of Virginia ran as an independent that cycle, eschewing his previous party label).

Democrats overcame their overextension in 1970 in part because Nixon held the presidency, while the overextended parties in 1926 and 1938 held the White House, perhaps contributing to their bigger losses. So as the Democrats assess their Senate odds in 2018, they can take some solace in the possibility that the midterm dynamic might help them protect their many vulnerable incumbents.

Table 1: U.S. Senate party exposure in midterm elections, at start of cycle and on Election Day

Notes: *“Start of cycle exposure” refers to the partisan makeup of the Senate class in question as of the December two years prior to Election Day (e.g. December 2016 for the midterm cycle 2018). ^The “start of cycle exposure” for 1914 began a few months before the ratification of the 17th Amendment, which established the popular election of U.S. senators, so 1914 data are based on the partisan makeup of Senate Class 3 as of the start of the 63rd Congress on March 4, 1913 rather than December 1912. #The appointment of Sen. Dean Barkley (I-MN) on Nov. 4, 2002 (the day before Election Day) to replace Sen. Paul Wellstone (D-MN), who died in the midst of his reelection campaign on Oct. 25, 2002, is not included. Changes in party exposure from the start of a cycle to Election Day reflect partisan changes due to vacancies and ensuing appointments and/or special elections, party switches, the addition of new states to the Union, etc.

Source: Crystal Ball research.

But comparisons between 1970 and 2018 should be made with caution. First, the political environment was far less polarized in 1970 than it is now. Take, for instance, the fact that about half of the states that held Senate races in the 1968 and 1972 presidential cycles had split presidential-Senate outcomes, the latter being Nixon’s “lonely landslide” win against George McGovern. Compare that to 2016, which saw 100% of states with Senate races vote straight-ticket for president and Senate (with Louisiana’s final outcome still to be determined in a runoff this Saturday). Granted, 2018 is a midterm year, but the point is that the high level of party polarization matters. In 2014, Republicans won all seven of the Senate contests that took place in states won by Mitt Romney in 2012. For Democrats in 2018, they only have one scheduled (i.e. Class I) target in a state won by Hillary Clinton — Sen. Dean Heller (R-NV), who may opt to run for governor instead — and only one other battleground state Senate seat to compete for — Arizona, where Sen. Jeff Flake (R-AZ) may face a notable primary challenge from Kelli Ward (R), a former state senator who lost to Sen. John McCain (R-AZ) in the GOP primary for Arizona’s other U.S. Senate seat in 2016.

Thus the 2018 midterm cycle features a clash of two patterns in American politics. On the one hand, the president’s party almost always suffers to some extent in midterm elections, though more consistently in the more-nationalized elections for the U.S. House than in the more-parochial elections for the U.S. Senate.

On the other hand, in this polarized era of American politics, the fact that Democrats are defending seats in some very Republican states suggests that the GOP should be in a good position to pick up seats despite the midterm environment.

[*] It’s hard to say how seriously they are under consideration, but Sens. Heidi Heitkamp (D-ND) and Joe Manchin (D-WV) both have been mentioned as possible Cabinet appointees in the incoming Trump Administration. If Heitkamp left her seat, North Dakota would hold a special election within 95 days of the vacancy, and we can only assume Republicans would be heavily favored (Trump won the state by nearly 36 points). If Manchin left his seat, incoming Gov. Jim Justice (D-WV) would presumably appoint a Democrat to replace him, but appointed senators often don’t have much of an advantage if they run for reelection, and Manchin may be the only Democrat capable of holding a Senate seat in the reddening Mountain State (Trump won it by almost 42 points, the biggest presidential margin of victory in the state’s history).

The States That Vote Most Often for Presidents

Kyle Kondik, Managing Editor, Sabato's Crystal Ball December 8th, 2016


Ohio has retained its title as the state that has most often voted with the presidential winner over the last 120 years, even though it took a strong turn toward the Republicans this year.

The Buckeye State voted for the winner for the 29th time in 31 elections this year, the best record of any state since 1896 (I explored the state’s presidential voting history in my book, The Bellwether). The full results from the states over that time period are shown in Map 1 and Table 1.

Map 1 and Table 1: State records of voting with presidential winner, 1896-2016

We chose 1896 as the year to begin this analysis because that election is often cited as a key realigning campaign where Republicans broke open a streak of highly competitive elections and became a majority party for most of the following three decades. If one started in 1916, looking at just the last 100 years of elections, Ohio and Nevada would be tied at 24/26, with Missouri and New Mexico at 23/26 (and Florida, the most competitive of the nation’s mega-states, at 22/26).

Notice that of the 10 states that have voted with the winners more than 80% of the time since 1896, Clinton carried eight of them — Ohio and Missouri were the exceptions. Many of these seeming bellwethers misfired in this election. But so did Ohio, given that Trump ran about five and a half points better in Ohio than he did nationally (51.7% in Ohio versus 46.2% nationally), meaning that Ohio wasn’t very close to the national average — the last time it voted so far from the national popular vote was 1932. So its reputation as a bellwether took a big hit in this election despite its vote for the winner.

The states that voted closest to the national average (Clinton currently leads the national popular vote by two points) were Minnesota (Clinton +1.5), Nevada (+2.4), and Maine (+2.7). The closest states that Trump won were Michigan (0.2), Pennsylvania (0.7), Wisconsin (0.7), and Florida (1.2). As Donald Trump prepares (presumably) for his reelection campaign, and Democrats begin thinking about where to battle him, these states — as opposed to historic bellwethers like Ohio, Missouri, Illinois, and New Mexico, all of which voted relatively far from the national average this year — probably merit the most attention.

In 2016’s Game of Musical Chairs, the Music Stopped at the Wrong Time for Clinton

Kyle Kondik and Geoffrey Skelley, Sabato's Crystal Ball November 30th, 2016


After the Bay of Pigs debacle, when U.S.-backed forces tried and spectacularly failed to topple Fidel Castro’s nascent communist regime in Cuba, President John F. Kennedy held a press conference and took blame for the failure. Speaking on April 21, 1961 — just a few months into his presidency — JFK memorably declared, “There’s an old saying that victory has a hundred fathers and defeat is an orphan,” meaning that when something goes right, many will want to take credit for it, but when something goes wrong, no one wants to take the blame.

We were thinking about the quote over the weekend after Castro’s death, particularly in relation to how it might be used to describe the 2016 election. That’s not just a reference to Trump advisers taking credit or Hillary Clinton’s campaigners shirking blame. Rather, we were thinking about how in a very close race, there can be many different factors that made the difference between one candidate’s victory and the other’s defeat. A tight election outcome might be said to have “100 fathers,” all of which may or may not have been decisive.

Indeed, the 2016 election was highly competitive. Trump won 306 electoral votes in an election where six states were decided by less than two points: Trump carried Michigan by 0.2 points, Wisconsin by 0.8, Pennsylvania by 1.1, and Florida by 1.2, while Clinton won New Hampshire by 0.4 and Minnesota by 1.5. Flipping Michigan, Pennsylvania, and Wisconsin would have given Clinton the Electoral College*, while a slightly better showing in Minnesota and New Hampshire would have given Trump a margin of more than 100 electoral votes.

After the election, there have been several explanations offered for Trump’s surprising victory. These include Trump’s digital campaign efforts, the Clinton campaign’s neglect of the Rust Belt battlegrounds, and the proliferation of “fake news” online — some of the hundred fathers of the 2016 presidential result. Every one of them might have had some bearing on the outcome.

While keeping in mind that all sorts of things potentially contributed to Trump’s victory, we have a theory that might help explain the essence of what happened.

Clinton seemed dependent on Trump’s many controversies throughout the campaign cycle. She enjoyed some of the best polling of the campaign during periods where Trump’s missteps dominated the news. These periods included the post-Democratic National Convention period, when Trump got into a spat with Khizr Khan, the Gold Star father who spoke against Trump at the DNC, and during the second weekend of October, when Trump’s offensive comments about women in the Access Hollywood tape surfaced. These stories seemed to suppress Trump’s vote, moving Trump voters from his column to either the third-party candidates or to undecided. Clinton generally kept her lead during the debate period (Sept. 26-Oct. 19), and post-debate polls indicated that the public thought Clinton got the better of Trump in all three debates.

But the last debate was nearly three weeks before Election Day, giving Trump time to recover and for new developments to alter the election’s state of play.

Sure enough, there was a major late development: FBI Director James Comey’s letter to Congress announcing that the bureau was looking into new emails potentially related to its investigation of Clinton’s use of a private email server while she served as secretary of state. Two days before the election, Comey followed up and said the emails did not change the FBI’s conclusion to not recommend charges against Clinton, but any damage to Clinton may have already been done — and the announcement itself underlined yet again Clinton’s email controversy.

The Comey letter dropped on Friday, Oct. 28 — nine days after the last debate, and 11 days before the election. By this time, Clinton’s six-point national lead on the day of the final debate had been cut to four, and it would fall to three in the RealClearPolitics average by Election Day. It now appears she’ll win the national popular vote by about two points, so the national polls taken together were close to the mark. Still, Trump managed to win by narrow margins in several key swing states.

Over the last two months of the campaign, the Crystal Ball kept an eye on the statistical relationship between the amount of attention Trump and Clinton received and the margin between them in the polls. We did this using a few different data points: Trump’s margin in the different iterations of the HuffPost Pollster and RealClearPolitics polling averages; how much more or less Gallup respondents said they had read, heard, or seen something about Trump versus Clinton in the last day or two; and how much more or less Google users searched for Trump compared to Clinton in the United States. While correlation between the polling margin and these two indicators of public attention does not necessarily imply causation, the notable correlations between some of these data suggest that there may have been something to the idea that greater exposure adversely affected a candidate’s poll numbers. After all, as Kristen Soltis Anderson and Patrick Ruffini of Echelon Insights recently noted in The Washington Post, there were more Google searches for “Donald Trump” in states that he lost on Nov. 8 than in states that he won.

In an Oct. 13 article, we found fairly strong negative correlations between Trump’s polling margin and the amount of attention he received over the last month of the campaign, according to Gallup and Google Trends. That is, when Trump got more relative exposure than Clinton, his poll standing tended to suffer, and vice versa.

What about the last month of the campaign up to Election Day? Shown in Tables 1 and 2, the results are mixed in comparison to the mid-October findings. The negative correlation between the Gallup data and the two polling averages actually became even stronger. However, the negative correlation between the Google data and the polls became weaker in the case of the RealClearPolitics data, but remained the same or became stronger in the case of HuffPost Pollster.

Table 1: Correlation between Trump’s polling margin and the difference between Trump and Clinton in Gallup’s “read, hear, or see anything about Clinton/Trump” question, as of Oct. 11 vs. Nov. 7

Table 2: Correlation between Trump’s polling margin and the Trump-Clinton difference in Google Trends’ search relevance, as of Oct. 11 vs. Nov. 7

Notes: 2-way polls are head-to-head averages for just Clinton versus Trump, 3-way averages include Gary Johnson, and 4-way averages include Johnson and Jill Stein.

Sources: Oct. 13 Crystal Ball article, Gallup, Google Trends, HuffPost Pollster (2-way, 3-way), RealClearPolitics, (2-way, 4-way)

The divergence between the Gallup correlations and the Google Trends correlations is due to differences regarding how much more Trump was on the public’s mind than Clinton in early to mid-October, following the release of the aforementioned Access Hollywood tape, and just after the third debate on Oct. 19, when Trump refused to say whether he would accept the election result. Google Trends found a much higher volume of Trump searches than Clinton searches in those time periods relative to how much more Gallup respondents said they had read, heard, or seen something about Trump than Clinton in those two periods.

If there was at least some negative causal relationship between attention and poll standing, it is notable that Trump received less attention in the final week and a half of the campaign than Clinton compared to the rest of the campaign. From the start of Gallup’s data (July 5) to Oct. 27 — just before Comey sent his letter to Congress — an average of 78.2% of respondents said “yes” regarding whether they had read, heard, or seen anything about Trump in the past couple of days, compared to Clinton’s 74.7%. From Oct. 28 to Nov. 7, the final day of Gallup’s data, the average percentages flipped to 80.7% for Clinton and 76.1% for Trump. In the same two timeframes, the Google Trends data showed a similar pattern. The average volume of Trump searches from July 5 to Oct. 27 was 28.1% to Clinton’s 21.0%; from Oct. 28 to Nov. 7, Clinton’s average search volume was 41.4% to Trump’s 36.3%. The Comey letter news coverage and Trump’s conspicuous discipline in avoiding controversial statements on the trail and on social media probably help explain this shift.

How much of this can be attributed directly to Comey’s letter and its subsequent coverage is debatable, particularly because there already were signs that Clinton’s lead was diminishing before the letter was released. There also were other unfortunate developments for Clinton during the last few weeks, including the drip-drip-drip of Russian-hacked emails released by WikiLeaks from Clinton campaign chairman John Podesta’s account, among others. In our view, there was never a particularly damning bombshell in the emails, but the steady stream generated daily negative coverage for the Clinton campaign.

The exit poll shows that a fair number of voters claimed to have decided in the final week or two of the campaign. If so, the focus on Clinton’s damaging storylines probably didn’t help her cause in winning over late-deciders. Nationally, the exit poll indicated that 63% of all voters were bothered “a lot” or “some” by Clinton’s use of private email, and Trump edged her 45%-42% among the 13% that decided their vote in the final week (and by a wider 48%-40% amongst those who decided in the final month, who made up a quarter of the electorate).

Crystal Ball senior columnist Alan Abramowitz examined the effect that these late deciders had on the four states that provided Trump his smallest margins of victory: Florida, Michigan, Pennsylvania, and Wisconsin. Those results are shown in Table 3.

Table 3: Late deciders in states Trump won by less than two percentage points

Note: “Late deciders” refers to those respondents in the exit polls who said they decided their vote choice in the week before the election.

Source: Exit polls, Dave Leip’s U.S. Atlas of Presidential Elections, Alan Abramowitz

Trump carried late deciders in all of these states — by notably larger margins than he did nationally — and the net effect that Abramowitz calculated from the exits was bigger than Trump’s margin of victory in all four states.

Now it’s possible that this was bound to happen anyway, without the Comey letter or the Podesta emails or any other late developments, in part because Clinton was representing the incumbent White House party after two terms. This election featured an unusually large number of voters who told pollsters they were either undecided or said they would support third-party candidates. In a prescient piece published on the eve of the election, Henry Olsen of the conservative Ethics and Public Policy Center correctly assumed that many of these undecided and third-party voters had profiles that made them likely Trump voters. Sure enough, most polls measured Clinton’s level of support well but understated Trump’s.

There were two competing changes in the electorate this year that, on balance, benefited Trump. Poll after poll indicated that Trump was going to do better with white voters that did not have a bachelor’s degree — certainly better than Mitt Romney did in 2012 — and that’s exactly what happened: Trump won non-college whites 66%-29%, surpassing Romney’s 61%-36% margin. Just as bad for Clinton, she did not do as well with college-educated whites as she needed to. After many pre-election polls suggested that Clinton could win white voters who had a college degree, Trump ended up carrying this group by three points — worse than Romney’s 14-point win, but not enough to push Clinton to an Electoral College majority. Meanwhile, Clinton’s performance with nonwhite voters was a little weaker than Obama’s in 2012 (with all of this, it’s important to remember that exit poll subgroup findings are not as precise as many think).

Had Clinton actually won college-educated whites nationally, she probably would have done better in some key places in critical states, such as the suburbs of Detroit, Milwaukee, and Philadelphia. Because of his special appeal and rural targeting, Trump was destined to significantly outrun Romney in the outlying, working-class parts of these states, and that much was obvious from pre-election polling. But Trump’s ability to hold the line in some large suburban counties was harder to predict in advance and perhaps was a late development influenced by the events we’ve discussed.

No one can be certain, of course, given the imperfect data available today. For example, isolating a single development such as the Comey letter to explain Trump’s win ignores many other factors in an extraordinary election. All of us in and around politics will be studying 2016 for a long time, hoping to unlock more of its tantalizing mysteries.

*There’s little or no reason to think that recounts being pursued by Green Party nominee Jill Stein will change the outcome in the decisive states, as friends of the Crystal Ball Jeffrey Blehar and Brandon Finnigan of DecisionDeskHQ recently argued.