2018 U.S. Senate elections: the state of play after Labor Day

I have written in broad terms (here and here) about the 36 United States Senate (“Senate”) races which will determine whether Democrats or Republicans control the Senate after the November 2018 midterm elections[1]. Including Independent Senators Angus King of Maine and Bernie Sanders of Vermont, Democrats control 49 seats; Republicans hold the remaining 51 seats (now that Arizona Governor Doug Ducey has appointed former Republican Senator John Kyl to fill the seat held by the late John McCain). To regain control (assuming Vice President Mike Pence, a Republican, would break a 50-50 tie), Democrats need to win a net of two seats on November 6, 2018. And while that sounds relatively easy, bear in mind that 27 of these seats are currently held by Democrats (including King and Sanders), with 10 of those seats in states Republican presidential nominee Donald Trump won in 2016, many by large margins.

In those previous posts, I described the political context of each Senate race (a variation on what the indispensable FiveThirtyEight.com calls “the fundamentals”) using three measures:

  1. State partisan lean (using 3W-RDM[2]): what you would expect the Democratic candidate minus Republican candidate margin would be if the national vote were split evenly between the two parties.
  2. Democratic advantage in the total national vote for United States House of Representatives [“House”]. As of September 4, 2018, FiveThirtyEight.com projects Democrats to win the national House vote by 8.4 percentage points, according to their “Classic” model.
  3. Incumbency. I calculate that, all else being equal, Democratic and Republican Senate incumbents garner 8.3 and 7.5 additional percentage points, respectively.

When I wrote those original posts, the identities of the Democratic and Republican Senate candidates were not yet known, and the polling sat between non-existent and irrelevant. Now, however, it is the day after Labor Day, the traditional starting point for American political campaigns. The field of candidates is effectively set in each state, and a slew of polls (of wildly-varying quality) have been conducted.

My goal here is to compare my “fundamentals” characterization to the actual polling to assess how the Democratic quest to recapture the Senate stands with (at most) nine weeks left to campaign. For 10 especially key races, I calculated a “weighted-adjusted polling average” (WAPA; current margin by which the Democrat leads/trails the Republican), otherwise I used the RealClearPolitics (RCP) average.

Just bear with me while I explain how I calculated WAPA.

First, I collected every poll released in 2018 listed on each race’s Wikipedia page [an exception to my preference to steer clear of Wikipedia]. For each margin I added/subtracted the pollster’s average partisan “bias” (how much, on average, a pollster’s results favor Democrats and Republicans compared to all other pollsters) listed in FiveThirtyEight.com’s pollster ratings. I then weighted each bias-corrected margin by a) how long prior to November 6 it was released (midpoint of days poll conducted[3] divided by 309, the number of days between January 1 and November 6) and b) the letter-grade rating assigned that pollster by FiveThirtyEight.com (on a scale where A+=4.3, A=4.0, A-=3.7, etc.; weights were numeric equivalent divided by 4.3) If a pollster was unlisted, I did not adjust for bias and followed FiveThirtyEight.com and assigned them a C+ rating.

However, because the “polls” WAPA treated all polls from the same pollster as “statistically independent” data points, I calculated a second WAPA for the six races (AZ, FL, MO, NV, TN, TX) in which at least one pollster released multiple polls in 2018. This “pollster” WAPA is the rating-weighted average of the time-weighted, unbiased average margin for each pollster. The final WAPA was the average of the two; only in Arizona and Texas did WAPA differ by two or more percentage points.

For the record, I did not adjust for “likely” vs. “registered” voters in these analyses. While the former tend to be slightly more Republican historically, I have not seen evidence they differ much this year, nor would it materially affect my conclusions.

Table 1: Summary of 2018 Polling Data in 10 Key 2018 Senate Elections

State # Polls/ Pollsters Raw Margin Bias-Adjusted Margin Average Pollster Rating Adjusted

 Poll Average

Adjusted Pollster Average Final Ave
AZ 9/7 D+6.6 D+5.4/5.5 2.4/2.7 D+6.8 D+3.9 D+5.3
FL 29/15 D-0.1 D-0.7/-0.4 2.5/2.7 D-0.6 D-0.5 D-0.5
IN 3/3 D+3.0 D+1.3 1.7 D+4.7 D+4.7 D+4.7
MO 18/11 D-0.4 D-1.5/-0.7 2.4/2.6 D-1.0 D-0.7 D-0.9
MT 4/4 D+5.3 D+3.6 2.0 D+1.9 D+1.9 D+1.9
NV 8/5 D+2.9 D+1.2 2.4/2.5 D+1.5 D+0.8 D+1.1
ND 4/4 D-2.8 D-3.8 2.3 D-3.6 D-3.6 D-3.6
TN 14/10 D+0.7 D-0.04 2.6/2.7 D+1.5 D+0.6 D+1.0
TX 15/11 D-6.4 D-6.6 3.0/2.9 D-6.1 D-4.1 D-5.1
WV 10/10 D+8.2 D+7.2 2.6 D+6.9 D+6.9 D+6.9
AVE 11/8 D+1.7 D+0.6 2.4/2.5 D+1.2 D+1.0 D+1.1

Before I summarize WAPA and compare them to the “fundamentals” in these 10 Senate races, I have observations about the polling itself (Table 1). One, publicly-available polls of these 10 Senate races have had a modest Democratic bias of about one percentage point. The bias is worst (D+1.7) in Indiana, Montana and Nevada, primarily due to the inclusion of at least one poll conducted by SurveyMonkey/Axios, whose polls have a pro-Democratic bias of 4.9 percentage points! Two, the overall quality of polling in these races has been…meh. The average poll within each of these races has been conducted by a pollster with a C+/B- rating. Again, much of this mediocrity can be ascribed to SurveyMonkey/Axios, which FiveThirtyEight.com assigns a D- rating; removing their polls entirely bumps the average rating to a more respectable B/B-. Three, while the Florida, Missouri, Tennessee and Texas Senate races have been polled 14 or more times in 2018, the Indiana, Montana and North Dakota have only been polled 11 times in total (by C-level pollsters, on average)—with no pollster polling the race more than once; the latter is also true of West Virginia, though it has at least been polled 10 times. Finally, only 32 (28.1%) of these 114 polls were conducted in July and August. Granted, polling is even more exceptionally tricky in the summer, when potential respondents are on vacation and/or tuned out of political news. But I still would love to see higher-quality pollsters like Marist, Monmouth and Quinnipiac conduct more surveys of these races soon.


This analysis is divided into five parts:

  1. Safe seats
  2. Endangered Democratic incumbents
  3. Endangered Republican-held seats
  4. Once-endangered Democratic incumbents who appear safe
  5. Wildcards

Safe seats. The good news for Democrats is that 15 of the 27 Senate seats they are defending are safe: all 15 states were won in 2016 by Democratic presidential nominee Hillary Clinton, they average 14.8 percentage points more Democratic than the nation as a whole, and each features an incumbent running in what looks like an exceptional year for Democrats. And while I do not expect these 15 Democrats to win by an average of 14.8+8.4+8.3=31.5 percentage points, it would be a historic upset if any of these Senators lost.

And that includes two Senators, Amy Klobuchar of Minnesota and Tim Kaine of Virginia, whose states are “only” D+1.5. The “fundamentals” still say each should win by 18.2 percentage points; the RCP averages for these two races (D+22.0 and D+19.3, respectively) suggest they are beating expectations. Also, 2016 Libertarian presidential nominee Gary Johnson, the former two-term governor of New Mexico, recently entered that state’s Senate race between incumbent Democrat Martin Heinrich and Republican businessman Mick Rich as a Libertarian, introducing unexpected volatility. Still, in the five most recent polls, Heinrich leads his nearest opponent by about 14 percentage points, below the “expected” margin of 23.2 percentage points, but near-safe nonetheless.

Similarly, four currently Republican Senate seats—those currently held by Roger Wicker (MS), Deb Fischer (NE), Orrin Hatch (UT) and John Barrasso (WY) are certain to remain in Republican hands, even with Hatch’s retirement; 2012 Republican presidential nominee Mitt Romney is a near-lock to win that race.

Endangered Democratic incumbents. Six of the 10 Democratic incumbent Senators running in states Trump won appear vulnerable to varying degrees, according to Table 2. “WTD” is a weighted average of expected (0.33) and WAPA (0.67) margin, and it reflects a “best guess” (albeit with a wide “margin of error”) of what the final outcome will be on November 6; using no weights or weighting polls 3 times more than “fundamentals” does not substantively alter the conclusions.

With one exception, the polling is quite good for these six Democratic Senators, as they are beating expectations (mean=D-3.8) by 5.2 percentage points on average.

Table 2: Most-endangered 2018 Democratic Senate incumbents

State Democrat Republican Expect WAPA Diff WTD
ND Heitkamp Cramer D-12.7 D-3.6 D+9.1 D-6.6
WV Manchin Morrisey D-18.8 D+6.9 D+25.9 D-1.7
MO McCaskill Hawley D+0.8 D-0.9 D-1.5 D-0.3
MT Tester Rosendale D-1.9 D+1.9 D+3.8 D+0.6
FL Nelson Scott D+9.6 D-0.5 D-10.1 D+2.8
IN Donnelly Braun D+0.4 D+4.7 D+4.3 D+3.3
AVE     D-3.8 D+1.4 D+5.2 D-0.3

 First-term Democratic Senator Heidi Heitkamp, who won her first Senate race by 0.9 percentage points, currently appears headed for a mid-single-digits loss to North Dakota’s lone House member, Kevin Cramer. Granted, she is polling 9.1 percentage points better than expected, but that is not nearly enough to overcome North Dakota being R+29.4. Still, there are only three (excluding SurveyMonkey/Axios) polls of this race, and she beat expectations six years ago.


North Dakota Senator Heidi Heitkamp

The good news, however, is that the other five Democratic incumbents are currently no worse than toss-ups.

The most surprising race on this list is Florida; eight months ago, I would have expected Ohio’s Sherrod Brown to be far more endangered. Running as a Democratic incumbent in slightly-Republican-leaning Florida (R+3.4), Bill Nelson “should” be ahead by about 13.1 percentage points (almost exactly his average in three previous races). Republican Governor Rick Scott has actually taken a slight polling lead (R+0.5), though, shifting from about 0.5 percentage points behind to 1.1 percentage points ahead starting in mid-June. As Scott is currently a statewide officeholder, I allotted him half of the 7.5 percentage point Republican incumbency advantage. Nelson “should” still win by 9.6 percentage points—and the weighted average projects a low single digits win for him on November 6.

Claire McCaskill of Missouri is also (slightly) underperforming expectations. The fundamentals make McCaskill the slightest of favorites (D+0.6), but she is currently trailing Missouri’s Attorney General Josh Hawley, a Republican, by 0.9 percentage points. And that average masks the fact she has lost ground since early June (when embattled Republican Governor Eric Greitens resigned, freeing Hawley from deciding whether to indict him), dropping from nearly-tied (R+0.2) to 1.4 percentage points behind. Still, this race is nearly a perfect toss-up (R+0.3).

Joe Donnelly of Indiana, by contrast, has taken a small polling lead (D+4.7) over Republican former State House member Mike Braun in what “should” be a coin flip (D+0.4). Caveat emptor: only been three polls have been conducted of this race (including one by SurveyMonkey/Axios), two showing Donnelly behind by an average 1.5 percentage points while the other has Donnelly ahead by 12 points! Still, a low single digits Donnelly win seems in the cards right now.

Finally, Montana’s Jon Tester and West Virginia’s Joe Manchin are considered by prognosticators to be in the “Lean Democratic” category, and I agree, even as the fundamentals and WAPA suggest caution. Tester should be losing to Montana’s Auditor, Matt Rosendale, by 1.9 percentage points, while West Virginia’s Attorney General, Patrick Morrisey, should be walloping Manchin by 18.8 percentage points! However, Tester’s narrow polling lead (D+1.9—in only four subpar polls) has him eking out a very narrow win. Manchin has a larger polling lead—D+6.9—but that still would have him losing by around two percentage points. Regardless, both men are proven winners in their states, and I think will both win in the mid-single-digits.

Bottom line: Democrats are likely to lose one, maybe two, of these six seats—an improvement over the five or six Republicans envisioned flipping just two years ago.

Endangered Republican-held seats. Four Republican-held Senate seats are in varying degrees of danger of being captured by strong Democratic opponents, according to Table 3: while Democrats should be losing these four seats by an average of 7.6 percentage points, they are slightly ahead in the polls overall.

Table 3: Most-endangered 2018 Republican-held Senate seats

State Democrat Republican Expect WAPA Diff WTD
AZ Sinema McSally D-1.3 D+5.3 D+6.6 D+3.1
NV Rosen Heller D+2.9 D+1.1 D-1.8 D+1.7
TN Bredesen Blackburn D-17.4 D+1.0 D+18.4 D-5.1
TX O’Rourke Cruz D-14.4 D-5.1 D+9.3 D-8.2
AVE     D-7.6 D+0.6 D+8.2 D-2.1

The retirement of Arizona Senator Jeff Flake, facing a tough reelection against Democratic House member Kyrsten Sinema, turned a toss-up into the Democrats’ best chance to flip a Republican-held seat; the caveat is that Republicans nominated their strongest general election candidate: House member Martha McSally. The fundamentals say McSally should win by about 1.3 percentage points, but Sinema has opened up a 5.3 percentage point lead in the polls; that lead has actually widened slightly since mid-June. A low-single-digits win for Sinema seems a good bet at this point.

kysten sinema

Arizona Senate candidate Kyrsten Sinema

The only Republican Senator seeking reelection in a state won by Clinton in 2016 is Nevada’s Dean Heller. The fundamentals back this up, suggesting he should lose to Democratic House member Jacky Rosen by about three percentage points. However, Rosen’s polling lead is “only” 1.1 percentage points, and Heller won reelection to a full term by 1.2 percentage points in 2012, even as Democrats won the national House vote by 1.3 percentage points. That said, this should be a far better year for Democrats, and Rosen seems set to win by a hair under two percentage points.


Nevada Senate candidate Jacky Rosen

When Tennessee Senator Bob Corker announced his retirement, what looked a possible 25 percentage point loss for Democrats became a little less uphill, even if the fundamentals still had Democrats down 17.4 percentage points. And when former two-term Democrat Governor Phil Bredesen entered the race against Republican House member Marsha Blackburn, the early polls showed a lead for the Democrat (D+4.1 through May 2018). However, since June, Blackburn has opened the narrowest of leads (R+0.4); overall, Bredesen leads by about one percentage ahead. Right now, that does not appear to be enough of a lead to overcome Tennessee being R+25.8, and I would expect Blackburn to win in the mid-single-digits.

Finally, Texas Senator Ted Cruz should be ahead by nearly 14.4 percentage points in this R+15.3 state. However, Democratic House member Beto O’Rourke has dramatically reduced that gap with eye-popping fundraising and relentless campaigning. I have suggested Texas could soon be fertile ground for Democrats, and O’Rourke’s (relative) success appears to bear that out. Nonetheless, O’Rourke still trails Cruz by about five points, though this obscures that the gap has dropped from 8.4 percentage points through early July to 3.5 percentage points since then. Overall, I would expect Cruz to win in the high single digits—though Democrats could win a victory of sorts if they force Republicans to invest time and money helping Cruz rather than in vulnerable House races.

Bottom line: Democrats could easily win two of these four seats, with Tennessee and Texas tantalizingly just out of reach.

Once-endangered Democratic incumbents who appear safe. The other four Democratic incumbents representing states won by Trump in 2016, Wisconsin’s Tammy Baldwin, Brown, Pennsylvania’s Bob Casey and Michigan’s Debbie Stabenow, appear headed for an average win of 14.5 percentage points (using RCP averages), nearly their average “expected” win of 15.9 percentage points (Table 4). Baldwin, who will face State Senator Leah Vukmir, is polling under 10 percentage points (8.0).

Table 4: Once-endangered, now safe 2018 Democratic Senate seats

State Democrat Republican Expect RCP Diff WTD
WI Baldwin Vukmir D+17.4 D+8.0 D-9.4 D+11.1
OH Brown Renaccia D+10.9 D+15.4 D+4.5 D+13.9
PA Casey Barlettab D+16.3 D+14.8 D-1.5 D+15.3
MI Stabenow Jamesc D+18.9 D+17.3 D-1.6 D+17.8
AVE     D+15.9 D+13.9 D-2.0 D+14.5

            a House member James Renacci

              b House member Lou Barletta

              c Businessman John James

Bottom line: These four swing states—three of which Trump won very narrowly—appear to be swinging solidly back toward the Democrats.

Wildcards. Three Senate races—two featuring appointed incumbents—are not (yet) in danger of changing hands, but each could still be interesting.

The fundamentals suggest New Jersey Democratic Senator Bob Menendez—in line to chair the powerful Foreign Relations Committee should Democrats recapture the Senate—should be ahead by a whopping 28.7 percentage points. However, the most recent RCP average has Menendez defeating businessman Bob Hugin by “only” 8.3 percentage points (three recent polls show an even closer race). This is almost certainly because Menendez was tried in 2017 on corruption charges, only to have a hung jury. While U.S. District Court Judge Jose Linares ultimately dismissed all the charges, they appear to have cost Menendez more than 20 percentage points of “expected” support. That is still not enough to overcome New Jersey’s strong partisan lean (D+12.0), however, and I currently anticipate Menendez winning by 10-15 percentage points.

When Minnesota Democratic Senator Al Franken resigned following allegations by numerous women of unwanted kissing and touching, Governor Mark Dayton appointed Democratic Lieutenant Governor Tina Smith to fill the seat. Smith will face reelection this November against Republican State Senator Karin Housley. Because Smith has only served in the Senate since January I (arbitrarily) cut her incumbency “advantage” in half, though the fundamentals still have her winning by 14.1 percentage points. However, the RCP average “only” has Smith ahead by 8.4 percentage points; perhaps voters hold Franken’s misdeeds against her. Nonetheless, I currently anticipate Smith winning reelection by around 10 percentage points.

And…the special election in Mississippi could decide Senate control—in late November. When Republican Senator Thad Cochran resigned for health reasons, Mississippi Governor Phil Bryant appointed Mississippi’s Commissioner of Agriculture and Commerce Cindy Hyde-Smith, a Republican, to fill the seat, making her the first-ever female Senator from Mississippi. Hyde-Smith will run for reelection on November 6 in a non-partisan (i.e., no party labels appear on the ballot) “open primary” against two opponents: Republican State Senator Chris McDaniel and former Agriculture Secretary Mike Espy, a Democrat. If no candidate tops 50% of the vote, a runoff election between the top two finishers—which polls suggest would be Hyde-Smith (simple average=31.2%) and Espy (28.5%), with McDaniel (16.8%) trailing well behind—would be held on November 27, 2018. The fundamentals suggest that a quasi-incumbent Republican Senator should beat a Democrat in Mississippi by 14.1 percentage points, though Espy could certainly make that closer. Still, this race is most likely to be an asterisk—easily remaining Republican—rather than a game-changer.

Bottom line: These three Senate races are intriguing, but the partisan gravity of these states makes it very unlikely any will change partisan hands.

Conclusion. Out of 36 Senate races, only 17 (18, counting New Mexico) are even remotely interesting, and only 10 of them are more than slightly likely to change partisan hands, based on each race’s “fundamentals” (partisan lean, Democratic wave, incumbency) and the current weighted-adjusted polling averages. As of September 4, 2018, Democrats seem likely to lose a Senate seat in North Dakota; retain four vulnerable seats (barely) in Florida, Indiana, Montana and West Virginia; and face genuine uncertainty in Missouri. They also seem likely to flip seats in Arizona and Nevada while falling short in Tennessee and Texas.

Overall—and with all necessary warnings about polling accuracy, unforeseen events and margins of error—Democrats appear poised either to net one Senate seat or to break even, depending on what happens in Missouri. This is NOT a prediction, merely a “best guess” based on available evidence: a net Democratic loss of one or two seats (or a gain of two or three seats) is certainly possible and would still put Democrats in an excellent position to regain Senate control in 2020.

Until next time…

[1] In previous posts, Republican Senator Thad Cochran of Mississippi had not yet resigned from the Senate.

[2] A weighted average (2008=16.7%, 2012=33.3%,2016=50%) of the difference between two measures: the state-level and national margins between the Democratic and Republican percentages of the total state/national vote.

[3] If the mid-point fell between two days, I used the later day. For example, if a poll was conducted between May 24 and May 29, the mid-point would between May 26 and May 27, meaning I would May 27.

18 thoughts on “2018 U.S. Senate elections: the state of play after Labor Day

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s