How Much More Ticket-Splitting Do Governor Races Have?

The notion that voters are far more willing to cross party lines for statewide offices than for federal positions is one of the most-commonly held viewpoints in politics today. It’s a point reinforced by scores of data that we do have; there’s little chance that ultra-popular GOP governor Charlie Baker would be winning a Senate race in Massachusetts, just as there’s basically no chance that Andy Beshear would win a Senate race in Kentucky. When voting to send someone to Washington, voters are cognizant of the fact that they’re partially giving a vote for which party should control a federal chamber, which has far-reaching implications for national policy. But when voting for a statewide office, the narrative says, those ramifications significantly decline, because voters are not voting for a party as much as they are voting for an individual to oversee running critical functions of the state.

The data over the last two midterm cycles shows that despite the acceleration of partisanship in voting, this overall notion still holds true, especially when incumbency is thrown into the mix. As a crude first step, we can map out how states voted in midterm Senate and Governor elections and compare them to the state’s most recent presidential outcome. For 2014, we can see that although only three states voted for a different party for Senate than they did for President in 2012, twelve states voted differently for governor.

2018 saw less of a divergence between the two, but the number of split outcomes was still greater on the gubernatorial side, with eight races voting differently for governor than they did in the presidential election and seven races voting differently for Senate (and this is before considering Kentucky and Louisiana’s 2019 gubernatorial flips and the Alabama Senate’s 2017 flip, which would drive the gubernatorial number up to 10 and the Senate number up to 8).

But this is perhaps a crude approximation of a metric, as the set of contested states differs slightly between the senatorial and gubernatorial slates. Moreover, margins matter, and a binary representation of outcomes as presented in the maps above may not be the best way to gauge this. To get a better idea of whether split ticketing is more common in gubernatorial or senatorial races, we might want to try and quantify the actual magnitudes of the changes in partisan lean for states across offices.

We can measure the average environment-adjusted deviation of states from their past election’s presidential partisanship for each type of race in both the 2014 and 2018 cycles. The basic idea is that a state’s partisanship can be measured with respect to how it votes relative to the nation in any given cycle. The more this changes across cycles, the more prone to candidate quality effects a race is. A measurement like this shows that while partisanship is accelerating and electoral volatility is declining, gubernatorial races can be significantly more prone to variance than Senate races.

Average Deviation from Presidential Lean: 2014Average Deviation from Presidential Lean: 2018
The average deviation between presidential partisanship and elections for a given type of race in the 2014 and 2018 cycles. To allow for cross-cycle comparisons, we control for the environment by adjusting for the difference in national popular vote between the midterm and the presidential cycle. To control for third parties, we use only the two-party vote.

In the 2014 and 2018 cycles, gubernatorial races saw, on average, around four percentage points more volatility in terms of partisan lean than a senatorial race did. That might not seem like a tremendous amount, but it would be enough to see Democrat Josh Shapiro win in Pennsylvania even if Republican Mehmet Oz wins the concurrent Senate election, or Tony Evers win re-election in Wisconsin even as incumbent GOP senator Ron Johnson wins a third term. More concretely, it implies that cross-state variance is significantly higher in gubernatorial race results, meaning it’s not at all inconceivable for Republicans to win the gubernatorial race in New Mexico while losing in Pennsylvania, whereas the Senate analogue (Republicans winning Arizona but losing Ohio) would be borderline unimaginable to most in 2022.

We can now look at this through another (and perhaps the most detailed) angle by creating a model that tries to predict a race’s margin by the state’s presidential lean. Doing this shows something interesting: while the correlation between presidential partisanships and Senate margin increased markedly between 2014 and 2018, the correlation between partisanship and gubernatorial results actually did not.

R2 of 2012 Presidential Results on 2014 RacesR2 of 2016 Presidential Results on 2018 Races
The R2 values of a model predicting a cycle’s actual results from the previous election’s presidential lean. Higher numbers indicate a stronger correlation.

This might seem counterintuitive at first, especially given what we know about crossover voting declining across the board. But it is easily explained by one thing: incumbency. 2018 saw a number of extremely popular Republican incumbents sweep their opponents in blue states like Vermont, New Hampshire, Maryland, and Massachusetts, despite the environment being singularly unfavorable to Republicans on the whole.

To properly analyze anything, then, it is important to control for incumbency in our models. Doing so gives results that are now in line with what we know about crossover voting and its decline, and show that after accounting for incumbency, gubernatorial races have been hit with the effects of increasing polarization. But the difference between the types of races is still clear and stark, and the lower R2 values show that even after accounting for incumbency, gubernatorial races have so far been prone to significantly greater amounts of variance and candidate quality effects than senatorial races have been.

R2 of 2012 Presidential + Incumbency on 2014 racesR2 of 2016 Presidential + Incumbency on 2018 races
The R^2 of a model that predicts the cycle’s actual results from the previous election’s presidential lean and incumbency. Higher numbers indicate a stronger correlation.

It might not be a surprise that incumbency can be worth so much more in a gubernatorial election. Partisanship matters significantly less in a gubernatorial race, because voters are not choosing who controls Congress through their vote, and so cross-party gubernatorial wins are still not at all uncommon. More voters will consider local politics over national politics, which bodes well for governors who have high approval ratings, but are on the wrong side of statewide partisanship on the national level.

With that out of the way, elections involving incumbents become more of a referendum on their performance as governor. And as J. Miles Coleman of the Crystal Ball said in an interview with the Daily Caller, it is extremely difficult to ask voters to fire someone that they approve of. It’s why Phil Scott, Charlie Baker and Larry Hogan won re-election by double digits as Republicans in the blue wave of 2018, despite running in the cobalt-blue states of Vermont, Massachusetts, and Maryland.

But even then, only 59% of a gubernatorial election’s margin was explained by incumbency and presidential lean, indicating that factors like fundraising and candidate quality probably matter a fair bit more for statewide offices than they do for federal ones.

What might this mean for 2022? To begin with, the importance of candidate effects means that it is actually possible for Democrats to win the Arizona, Michigan, Wisconsin, and Pennsylvania gubernatorial elections even in a Republican-leaning year, due to a mix of incumbency and candidate quality. By the same train of logic, it is possible for Republicans to simultaneously win New Mexico and Maine, even under that very same environment that saw Democrats hold Arizona and the Rust Belt. It is even theoretically possible that Republicans take the Oregon governorship, currently rated by us as Likely Democratic, due to a combination of exiting governor Kate Brown’s unpopularity and moderate Betsy Johnson’s presence on the ballot as an independent. And it is not at all inconceivable that these scenarios all happen in the same universe where Republicans expand the Senate GOP caucus count from 50 to 53 seats in November. After all, the gubernatorial map has historically had far weaker cross-state correlations than the Senate map has had.

It is important to note that there is no guarantee that gubernatorial races do not stratify by partisanship as well. The strong correlation between the 2021 swings right in both Virginia and New Jersey point to partisanship becoming stronger and stronger across the board, and it is more likely than not that this correlation continues intensifying. But the difference between senatorial and gubernatorial outcomes is likely to remain, at least to some degree, especially if 2022 polling is predictive of anything.

I’m a software engineer and a computer scientist (UC Berkeley class of 2019 BA, class of 2020 MS) who has an interest in machine learning, politics, and electoral data. I’m a partner at Split Ticket, handle our Senate races, and make many kinds of electoral models.

Leave a Reply

%d bloggers like this: