Now that the dust has settled and all the votes have finally been counted, it’s time to look back at the 2018 election polling: what it told us, what it didn’t, and what we can learn for future elections.
In my poll-analysis columns leading up to the election — see here, here, here, and here — I focused on the statewide races (senators and governors) and how we might expect them to break in the Democrats’ direction if 2018 was going to follow the pattern of wave elections over the past two decades.
Looking at both the pre-election indicators (presidential approval, generic ballot) and the Democrats’ 41-seat gain in the House, this was undoubtedly, in broad outlines, a wave election. As Sean Trende notes, using a metric that evaluates the overall strength of the parties across the Senate, House, governors’ mansions, and state legislatures, the overall results were fairly middling for a wave election compared to past midterm wipeouts. And the Senate and governors’ races were a part of why.
Let’s start with the Senate. As I explained in prior columns, one way of looking at the RealClearPolitics polling averages is to count up the fraction of voters who show up as undecided in the averages, and ask what percentage would need to break towards one party or the other to get them to 50 percent of the two-party vote. It’s hard to tell how much of the last-minute movement is actually undecided voters deciding, and how much is polling error, specifically errors in projecting who would show up (most of public polling these days is about projecting turnout).
On the other hand, if your goal is to test how well public polls project election outcomes, it’s almost irrelevant whether polls were “right” or “wrong” in what they were reporting, if outcomes diverge from them predictably. It’s like having a watch that runs two minutes slow: If you know it is always going to be off by the same amount in the same direction, it’s not a problem, just something to adjust for.
So, how did the “undecided vote” break at the end towards Republicans?
In the six races at the top of the chart — Tennessee, Missouri, Michigan, Indiana, Florida, and West Virginia — Republicans outperformed the polling averages strongly enough that we can safely conclude that the polls were just off the mark. (Obviously, if over 100 percent of undecideds break one way, that suggests that there were more people undecided than that, or at least fewer supporting the other candidate than the polls estimated.) In Florida, the most intensively polled race down the stretch, the polls nonetheless got the outcome wrong in both the Senate and governor’s races. Given past polling misses in Michigan and Indiana, that should perhaps not surprise us. In Tennessee, even post-election polls have found that voters like Democrat Phil Bredesen, a former two-term governor, quite a lot, so it’s possible that his support in polls was just soft, and voters when asked to choose stuck with their partisan preference.
At the opposite end, the Nevada, Arizona, Texas, and New Jersey races stand out as ones that underestimated the Democrats. In Texas in particular, where the polling had been remarkably consistent leading up to Election Day, that was clearly a matter of a surge in Democratic turnout. In Arizona, where the final results only shifted to the Democrats as votes were counted days after the election, it looks more like underestimating the early mail-in vote for Sinema. Nevada, by contrast, is more or less the slow watch: Democrats almost always do about two points better than they poll there, because some segments of the Nevada electorate are just very hard to reach on the telephone.
Now, let’s step back to the mid-September polling, given my initial thesis (backed up strongly by the outcomes in 2002, 2004, 2006, 2008, 2010, 2012, and 2014) that in a wave election, the races will shift consistently across the board to the wave party, so that the wave party will predictably win all the races where it leads in mid-September, and pick off a few more:
That’s not how it turned out. Only three races ended up with a different outcome than the mid-September polls had predicted, and all three were basically dead heats at the time: Arizona flipped to the Democrats, and Tennessee and Indiana flipped to the Republicans. Combining the Tennessee blowout (Marsha Blackburn won by double digits), Mike Braun’s solid five-point win, and Republicans busting open small leads in Missouri and North Dakota and narrowing the races in West Virginia and Montana (and even ending a lot closer than projected in Ohio, Michigan, and Pennsylvania), it’s almost impossible to treat this as behaving like a traditional Senate wave. Democrats did win more easily than projected in Nevada (which was neck-and-neck well into October), New Jersey, Wisconsin, and Minnesota, and the apparent break towards Corey Stewart in the Virginia race is partly attributable to there being very little polling after Labor Day, but there was only one race besides Arizona where Democrats pulled ahead after trailing in mid-September — Florida — and they ended up losing.
Does this mean the wave theory of Senate elections is dead? The playing field was, very disproportionately, unfavorable to Democrats, putting them on defense in a lot of deep-red territory and offering them few easy pickups, but for these purposes, that normally shouldn’t matter: The playing field is already baked into the polls in mid-September. But maybe it mattered this year in two ways. First, a large, polarizing event in late September (the Kavanaugh hearings) mattered more in places like Tennessee and Indiana than it would have in a more evenly divided state like Arizona, weighing down Democratic efforts to defy partisan gravity.
Second, the pattern of states that defied the wave is suggestive: It matches up with states that had a lot of white-working-class Trump supporters in 2016. By contrast, the states where the wave broke more predictably had a lot of non-white voters of the sort who supported Obama but did not regularly show up at the polls in previous midterms: Texas, Arizona, Nevada. I spoke after the election to Brent Buchanan of Cygnal, a Republican pollster that ran public polls in Florida, Georgia, Indiana, Missouri, Ohio, and Tennessee and had an excellent record. His view was that turnout looked much more like a presidential race this year, helping Democrats a lot in places like Georgia, but helping Republicans more in states like Ohio and Montana. Cygnal, unlike the public poll averages, projected (correctly) that Rick Scott and Ron DeSantis would pull out their races in Florida, which Buchanan attributes to the public pollsters overestimating Hispanic turnout as a proportion of the massive turnout in those races. On the whole, however, Buchanan sees 2018 as a good year for pollsters, one that should restore some trust in their usefulness – and he told me that it’s important for voters skeptical of pollsters to take the time to answer their questions honestly, given how much influence polls have in shaping policy and campaigns.
Let’s turn to the governor’s races:
The playing field was almost as hostile for Republicans in the governor’s races as it was for Democrats in the Senate, given the number of two-term Republican governors retiring and the number of deep-blue states Republicans entered the cycle holding. Here, compared with the Senate, we see a lot more races where less than 40 percent of the undecided vote broke to the Republican. And yet the polling variance again included six races that were way off, four of them in the direction of underestimating the Republican, and the final RCP averages missed the outcomes in four races, three of them won by Republicans (Florida, Ohio, Iowa) and one by the Democrat (Nevada again). Races that broke hard to the Democrats included Maryland and Massachusetts, where Republican incumbents were running away with their reelections anyway.
Scott Walker, unsurprisingly, finished more strongly than the polls predicted, though ultimately not strongly enough to win a third term. If you look at the history of the Wisconsin governor’s races going back to the last of Tommy Thompson’s four consecutive victories as the state’s previous Republican governor, you can see that while Walker’s majority support eroded this time, it was a narrow loss from what was never a very large majority:
Look closer at the vote totals:
Walker actually got a few more votes in the 2012 recall contest than in any of his other three elections, in part because some voters who may have preferred a Democrat nonetheless responded to his argument that a midterm recall was an improper and abnormal attack on an election outcome. But on the whole, his support was more or less the same from 2012 through this year; Wisconsin Democrats just got more people to turn out this time. If those had been voters on fire to get rid of Walker, they’d have shown up in earlier years; the more obvious explanation is that he was a victim of the national wave conditions.
Here’s how the governor’s races broke after mid September (and bear in mind, the trend of races’ breaking consistently with a national wave is one I derived from past Senate races; governor races are different).
Doug Ducey blew open the Arizona governor race, using his popularity as an incumbent to dodge the wave that ended up tipping the Senate race away from Martha McSally. By contrast, the nation’s most unpopular governor, Dan Malloy in Connecticut, made his Democratic successor’s bid a lot closer at the end, whereas the Democratic incumbent next door, Gina Raimondo, took control down the stretch.
Yet again, we see here an overall pattern of more races breaking towards the Democrats, yet two of the three races that got away from the mid-September leader (Florida and Oklahoma, the other being Kansas) went to the Republicans, as did two of the three races that had been tied in mid-September (Georgia and Iowa, the exception being Maine). Some of that was just bad early polling combined with the lower public profile of gubernatorial elections: the RCP poll averages had both candidates under 40 percent in Rhode Island, Iowa, Maine, Kansas, Nevada, and Oklahoma in mid September.
Two of the most interesting contests were in Florida and Georgia, where long Republican winning streaks were extended in racially charged races, confounding pundits who penned glowing tributes to Andrew Gillum and Stacey Abrams. First, Florida:
In percentage terms, the large majorities rolled up by Jeb Bush started to erode when Charlie Crist was elected against the Democratic wave in 2006 and have evaporated since then; Rick Scott and now Ron DeSantis won the last three races by razor-thin margins. But DeSantis’s win was quite different in raw numbers of votes cast:
Turnout in Florida was simply off the charts. The combination of Gillum and the blue-wave conditions drove Democratic votes to new heights, but the Florida Republicans still held them off. DeSantis’s vote totals look much more like those of presidential-election-year candidates (all totals in millions):
The state’s dramatic population growth makes it hard to usefully compare Florida elections over the long run, but you can see here that the 2018 vote totals look a lot more like presidential turnout in Florida than they do like a normal midterm. Now, the trend in Georgia:
We see a similar pattern here to what we saw in Florida, a state that Republicans conquered in 2002 but have been losing ground in recently. Now, the vote totals:
This, too, is almost presidential-level turnout; not quite, as each of the last three Republican presidential candidates cleared 2 million votes. There’s been a huge surge in new voter registrations over the past few years in Georgia, and it’s a disproportionately younger, non-white electorate — a trouble sign if Republicans don’t have a good plan to appeal to those voters. But as in Florida, even in a hostile environment and against a turnout surge, Brian Kemp held on.
In the end, the statewide races showed that Republicans could break the historic trends that typically attend wave elections, even amidst heavy losses in the House. But in contrast to 2016, when many down-ticket Republicans ran ahead of Trump by winning in areas where he was doing badly (mainly suburbia), they did so by circling the wagons around the Trump base areas. It will take more election cycles to tell whether that type of strategy is an effective defense at breaking waves.