2018 senate election predictions

2018 Senate Election Predictions

A Republican Majority

Our 2018 Senate Predictions show the Republicans have a 75.0% chance of retaining the majority in the Senate.

Democrats needs 51 Senate seats to hold a majority -- a tough ask given the Senate map (Democrats defend 10 seats in states Donald Trump won whereas Republicans defend a single seat -- Dean Heller's -- in a state Hillary Clinton took).

Republicans only need 50 seats to hold the Senate majority because Vice President Mike Pence casts the tie-breaking vote.  Though Republicans have a favorable map, Trump's unpopularity has hurt the party in generic congressional polls, a leading indicator of midterm fortune.  This makes it unlikely for them to pick up many (if any) seats.

If our House election forecasts prove correct and the Democrats take over the lower chamber, Congress will be split for the final two years of Trump's first term.

The Senate Map

PoliticalEdu's 2018 Senate election predictions have only one seat changing hands: Dean Heller's in Nevada.  One other, Arizona, is a tossup as incumbent senator Jeff Flake opted not to run for reelection.

Tennessee's open seat will be competitive if Democrats retain their huge advantage in the generic vote.

2018 senate election predictions



StateIncumbentIncumbent PartyLikelihood of a Democrat Winning
AZN/AGOP36%
CAFeinsteinDem99%
CTMurphyDem99%
DECarperDem100%
FLNelsonDem99%
HIHironoDem100%
INDonnellyDem98%
MEKingInd (Dem)99%
MDCardinDem100%
MAWarrenDem99%
MIStabenowDem98%
MNKlobucharDem99%
MSWickerGOP6%
MOMcCaskillDem99%
MTTesterDem97%
NEFischerGOP13%
NVHellerGOP67%
NJMenendezDem98%
NMHeinrichDem97%
NYGillibrandDem97%
NDHeitkampDem97%
OHBrownDem100%
PACaseyDem97%
RIWhitehouseDem97%
TNN/AGOP22%
TXCruzGOP2%
UTHatchGOP0%
VTSandersInd (Dem)98%
VAKaineDem97%
WACantwellDem98%
WVManchinDem98%
WIBaldwinDem100%
WYBarrassoGOP0%



Interactive Senate prediction map (hover for breakdown, scroll to zoom)

d3-cloropleth-map



THE MODEL

The 2018 Senate elections have a frightening map for Democrats.  Defending 25 seats, 10 of which come from states Donald Trump won in the 2016 election, and with few opportunities to gain seats – only Nevada and Arizona seem winnable – Democrats face a challenge simply to maintain the status quo.

Thankfully for the party, Trump’s unpopularity may well salvage Democrats in seats otherwise losable.  Even a strong swing year that makes the Democrat competitive in the House of Representatives will likely not tip the balance of power in the Senate, but will likely stave off a bad outcome in the Senate (a nightmare scenario would be dropping down to 39 seats and losing the ability to filibuster).

Democrats have a 9.12 percent chance of winning the Senate and are currently forecasted to hold all their seats and pick up Nevada from Dean Heller.  The image below shows the likelihood of a Democrat winning each state with a Senate election in 2018.  A deeper-blue implies a greater likelihood of victory; a deeper-red signifies the opposite.[1]

The model gives a strong advantage to incumbents, which explains why Democrats have a high likelihood of holding their seats while most Republican seats remain a deep-red.[2]  Similarly, a large lead for the Democrats in the generic congressional ballot,[3] a leading indicator for success in both the House and Senate,[4] further makes victory even in red states likely (and makes open-seat Tennessee almost competitive).



How the model works

The binomial logit model regresses Democratic Senate victories on the following variables:

  1. Forecasted congressional popular vote (drawn from adjusted generic congressional polls)
  2. The state’s Democratic vote share in the last presidential election
  3. A slight variation of the state’s partisan voter index
  4. The state’s Senate partisan voter index – this compares the Democratic vote share in the state’s preceding Senate election to the House popular vote of that same election. A state that tends to fare more Democratic in Senate elections than in the House boosts the Democrats’ chance of winning (eg, in West Virginia and North Dakota)
  5. The incumbent’s Senate partisan voter index, which compares how the incumbent performed relative to the Democratic share of the House popular vote for the year of his or her last election. Incumbents who underperform the national House vote are weaker and have a lower likelihood of winning
  6. How well the incumbent ideologically matches the state. This metric pits the incumbent’s DW-NOMINATE score against the average DW-NOMINATE score of the state’s House delegation.  The closer the ideological match, the higher the chance of reelection
  7. Whether the incumbent is Republican or Democrat
  8. Region
  9. Midterm mediated by the president’s party (the out-party tends to perform well in midterm elections)

The model has a Brier score, which measures error of probabilistic models, of 0.055.[5]



FORECAST METHODOLOGY

To generate probabilistic Senate outcomes, the model:

Forecast numbers from each new generic Congress poll

  1. Simulates each Senate election 30,000 times, varying at each iteration the forecasted generic vote tally based on the forecast’s errors
  2. Apply a uniform national error that adjusts the predicted likelihood by between 0 and 6 points
  3. Add a separate local error to each state for each iteration that changes the predicted likelihood by, again, 0 to 6 points

HANDLING UNCERTAINTY

Since one variable -- the popular House vote -- cannot be known ahead of time, the model must address the inherent uncertainty in the popular vote forecast (which can be best interpreted as a range of likely outcomes).  To do so, each simulation uses a different popular vote number drawn from a normal distribution centered at the mean forecasted and with a standard deviation of the popular vote forecast's RSME.  National and local errors also flatten the distribution curve and make the model a little less certain by moving any given initial prediction up to 12 points in either direct.



POTENTIAL SHORTCOMINGS

The model is rather aggressive and may overstate the incumbency advantage.  It is also agnostic to candidate characteristics and state-level polling.  A high-quality challenger may have a higher likelihood of winning than the model reflects (eg, Jason Kander in Missouri or a potential Rick Scott candidacy in Florida).  More data will be added to hopefully negate these two potential shortcomings.

[1] For interactive features and the ability to embed in a website, please see https://plot.ly/~pfp/0/_2018-senate-predictions-hover-for-breakdown/

[2] All seats are assumed to have an incumbent running unless that incumbent declares retirement, as in Tennessee.  Some incumbents face potentially challenging primaries, such as Jeff Flake (R-AZ).  The model assumes his victory in the primary, but, if he loses, will drop the incumbency factor from the predictions, likely moving Arizona into a swing state

[3] See http://politicaledu.org/generic-congressional-vote-forecast/ for an in-depth explanation of the number’s calculation

[4] Whereas 2018 House forecasts must balance the generic poll with geographic clustering and gerrymandering, such issues do not plague Senate models as states, of course, are not gerrymandered

[5] Brier scores range from a perfect 0 to a highly imperfect 1