Vinod B
Published in

Vinod B

Forecasting Who Will Win the Senate in 2020

The Answer is Key to the Future of American Democracy

(See here for the Presidential forecast)

Currently, Democrats are favored to win both the house and the presidency, making which party controls the senate of paramount importance.

If Democrats control the senate, then they stand a strong chance of controlling both legislative houses and the presidency, creating a unified government. That would allow them to pass potentially transformative legislation to tackle things like economic stimulus for COVID, climate change, and healthcare. This of course assumes they get rid of the filibuster or somehow get lucky in the election and manage to win 60 or more seats (which we will see is unlikely).

If, however, Republicans win the senate, Democrats will likely face a divided government with control of either the house and/or presidency only. With the current rates of polarization in congress that would almost surely lead to gridlock and no substantive legislation if Mitch McConnell repeats his standard tactics.

Winning the senate in 2020 is therefore essential for Democrats if they want to make progress on key tenets of their party platform. Without unified government there is little hope to tackle structural issues in American Democracy or long term problems like climate change.

Building a Senate Forecasting Model

To understand the current state of each race, we need to build a forecasting model for the senate in 2020. We will build what is known as a fundamental and polls hybrid model that combines the predictive power of fundamental factors like economic conditions and state demographics with senate polls.

The reason we want to combine the two is that historically polls based models are much more accurate in forecasting senate elections compared to fundamentals only models; however, not every state with a senate race always gets polled. Therefore, we need to supplement polling data with a fundamentals based prediction in some cases to ensure we have a prediction for all races.

To highlight the difference in performance, here is the yearly accuracy in each election since 2012 based on the number of senate races called correctly in a pure fundamentals based model vs. a fundamentals and polls hybrid (note we assign candidates like Bernie Sanders who caucus with Democrats as Democrats even though they are nominally independents):

In almost every election, a fundamental and polls hybrid is more accurate than a fundamentals only model, missing only 1–3 races each cycle, which is why we focus on the combination model.

Now we run through how each part of the model is constructed.

For the fundamentals model, we build a regression model that predicts the share of each party’s vote in a senate race in a given election year (for Democrats, Republicans, and a generic third party candidate) as a function of the following characteristics:

  • State the race is in
  • Indicator if the race is a special election or not (since these often have lower turnout and as a result can see unexpected results)
  • Indicator for the last party to hold the seat; scored as a +1 for a Democrat, 0 for a third party, and -1 for a Republican
  • Indicator for whether an incumbent is running in the race i.e. the person who last held the seat is running again; scored as +1 for a Democrat, 0 for a third party, and -1 for a Republican
  • Partisan voter index for the senate defined as the difference in the Democratic-Republican vote margin in the state compared to the national vote margin across races as a mix of the last two elections (75% and 25%)
  • Partisan voter index for president defined as the difference in the Democratic-Republican vote margin in the state compared to the national vote margin across states as a mix of the last two elections (75% and 25%)
  • State unemployment rate in the month prior to the election
  • State house price growth rate in the year prior to the election
  • National GDP growth rate in the year prior to the election
  • Fraction of white residents in a state as of the most recent census

The most important variables by far in this model are whether an incumbent is currently running in the race and which party last held the seat. This is because senate races see a huge bias towards incumbent candidates, which is likely due to their higher name recognition statewide relative to newer competitors.

This fundamentals based model allows us to forecast for each senate race in 2020 what each party’s vote share is expected to be based on historical senate races since 1970.

The polls portion of the model is trained on senate polling data since 1998, and we set up the polls to take the vote shares of the top three candidates in a state, which almost always map to a Democrat, Republican, and third party candidate (unless a party decides to not run a candidate in a given race, which has happened on rare occasions). We therefore ignore other candidates outside the top three such as in states with open primaries like Louisiana.

To get a prediction for each race we create a model to calculate a weighted average of all polls in a given state based on several factors:

  • Historical accuracy of the pollster
  • Historical bias towards a particular party (house effects)
  • Sample size in the poll (higher sample sizes get more weight)
  • Time till election day from when the poll was constructed (more recent polls get a higher weight)

We also adjust each poll using a linear time trend based on how polls in the state have trended over time. This allows us to act as if each poll was conducted closer to election day and ensure every poll is adjusted for overall trends in voter preferences based on more recent polling.

For each race we then calculate a final prediction for the Democratic, Republican, and third party vote shares by using a blend of the fundamentals based prediction and the final weighted average from all polls in a state. In cases where a state has no polls, we use the fundamentals based prediction only.

As a last step, we turn this single prediction into a probabilistic forecast by adding random errors to each state’s forecast. We estimate these random errors based on the historical accuracy of the model’s forecasts and allow for correlation across states as well as for national swings in voter sentiment. Repeating the addition of these random errors to our main predictions a large number of times (50,000) allows us to calculate the probability that Democrats will win each senate race and other statistics such as how likely each party is to get more than 50 senate seats.

The Current Forecast — Democrats are Ahead

The fundamental and polls hybrid model currently gives Democrats a strong chance of winning the senate with a 63% chance of winning more than 50 seats and therefore an outright majority.

There is a 10% chance that both parties will win 50 seats, leading to a tie, and a 30% chance that Republicans win more than 50 seats. If both parties gain 50 seats in the senate, ties are broken by the vice president, which is determined by who wins the presidential election. A fundamentals only forecast of the presidency gives Democrats a 75% chance of winning and Republicans a 25% chance. Therefore, Democrats are likely to control the senate roughly 71% of the time (63%+75%*10%).

An 80% confidence interval for the number of seats Democrats are likely to hold is 47 to 57 with a median of 52, suggesting they will likely just hold a majority with almost no chance of crossing the key 60 seat threshold necessary to avoid the filibuster. Indeed, in simulations Democrats win 60 or more seats only 3% of the time. This shows the importance of removing the filibuster if Democrats want to get anything major done.

And here are the predictions per state (bolded are races where neither party has a 60% or greater probability of winning):

Democrats are expected to win 17 of the 35 races up for election. With 33 guaranteed seats (and two held by independents Angus King and Bernie Sanders who previously caucused with the Democrats), that would leave them with the expected 52 seats and a slight majority in the senate (33+2+17).

The five closest states with a Democratic win probability between 40% and 60% are: Maine, Iowa, Montana, and the regular election in Georgia (Democrats hold a high chance of winning the special election).

Below we dive deeper into these races and discuss the individual candidates along with trends in recent polls to better understand the state of these key races. Other contests not discussed have a Democratic win probability below 40% and are likely out of reach or a probability above 60% and are therefore considered safer.

Maine: Susan Collins (Republican, Incumbent) vs. Sara Gideon (Democrat)

Our forecasting model gives Democrat’s a 57% to win this seat, which is somewhat surprising given that Susan Collins, the Republican candidate, is an incumbent who has held the seat since 1997 and won by a wide margin in 2014 against Democratic challenger Shenna Bellows by +37%.

However, Donald Trump is fairly unpopular in the state, having lost it in the 2016 election and looking unlikely to win it in 2020 according to current polls. This cascades down ballot to affect Susan Collins’ chances, especially given the importance of healthcare to Maine voters, Susan Collins’ inconsistent stance on the ACA, and Trump’s repeated attempts to repeal the ACA.

Under the current polling trends below, we see that Gideon became the favorite candidate towards the start of the year and has held the lead ever since, only seeing it grow. This explains why the seat currently leans Democratic despite the favorable fundamentals for Republicans due to the incumbency advantage.

Iowa: Joni Ernst (Republican, Incumbent) vs. Theresa Greenfield (Democrat)

Our forecasting model gives Democrat’s a 56% chance to win this seat, similar to the odds in Maine, and despite the fact that the Republican candidate here is also an incumbent.

Joni Ernst won this seat in 2014 by +8% against Democrat Bruce Baley, who was looking to replace the incumbent Democrat Tom Harkin when he decided to retire from the senate instead of seeking reelection.

While Trump won Iowa in 2016 by +10%, and the state overall has favorable demographics for Republicans with more white voters, less college educated voters, and a higher degree of evangelicals compared to the typical US state, Trump is not very popular in Iowa today. Biden narrowly leads in current polls.

Trump’s lack of support for the ACA and poor job handling COVID has made Ernst’s reelection campaign all the more difficult. Many Iowans think she has not done enough for the state in her time in congress, and she has higher unfavorable ratings than Theresa Greenfield. If Trump continues to turn off voters, especially those in the suburbs and women, it does not bode well for Joni Ernst’s campaign.

Polling trends show quite a close race. The most recent polls show Ernst with a slight lead while a larger number of slightly older polls show Greenfield with a narrow edge. All these factors align to make the Iowa senate seat one of the most hotly contested campaigns and the second most expensive race in 2020.

Montana: Steve Daines (Republican, Incumbent) vs. Steve Bullock (Democrat)

Montana is yet another race where a Democrat is running against an incumbent Republican but still stands a decent chance at winning with a 45% chance according to our forecasting model.

Steve Daines won the open seat in 2014 against Democratic challenger Amanda Curtis by +17%. Both candidates sought to replace Democrat John Walsh who was appointed to finish the final year of the previous term in 2014 when Democratic incumbent Max Baucus decided to retire early. John Walsh dropped out of the 2014 race due to plagiarism allegations, leading to Amanda Curtis taking up the challenger mantle.

Despite Daines’ strong victory in 2014, the race this year is proving to be much closer. Steve Bullock has higher name recognition than a typical senate challenger due to his failed presidential campaign in 2020 and also the fact that he is currently the governor of Montana. Steve Bullock also has high favorability ratings among voters for his job as governor.

Like elsewhere, Trump’s popularity has slipped in Montnna relative to 2016 when he won the state by +19%, which is hurting Daines’ down ballot, along with his anti ACA stance given the plethora of rural hospitals in the state.

Voters in Montana have also historically split tickets, voting Republican for President and Democratic for other offices. Bullock has brandished his bipartisan working credentials from his track record as governor where he passed numerous pieces of legislation alongside a Republican state senate, claiming he can fix Washington, which is always a strong platform to run on.

That image combined with his high favorable ratings in the state and stances on key issues like health care and climate change have given Bullock a better than expected chance. Any other democratic candidate would probably not be this close in the race.

Still, as the polling trends show Bullock’s support has been mostly flat over time while Daines has gained support. Key to Bullock’s victory will be turnout among his favorable demographics: young voters and Native Americans.

Georgia (regular): David Perdue (Republican, Incumbent) vs. Jon Ossoff (Democrat)

Georgia has two elections in this cycle with the regular election pitting incumbent Republican senator David Perdue against Democrat Jon Ossoff. Democrats have a 42% of winning this regular senate election according to our forecasting model.

David Perdue first won the seat in 2014 when incumbent Republican Saxby Chandliss decided to retire. Perdue won against Democrat Michelle Nunn by +7%.

Jon Ossoff has relatively high name recognition for a challenger in Georgia due to his notorious campaign for the the 2017 Georgia 6th district, which was the most expensive house election in history. Ossoff won the primary election but failed to get 50% of the vote and subsequently lost the runoff.

Georgia is an unusual state this year due to the national swing in the presidency towards Democrats. Biden is actually favored over Trump in some polls in Georgia, which hurts Perdue’s down ballot chances and makes the race narrower than it otherwise would be.

Ossoff has also been buoyed by a fundraising bump after Perdue (jokingly or not) failed to pronounce Kamala Harris’ name correctly and Ossoff used the opportunity to denounce Perdue as bigoted.

In terms of issues, like elsewhere, healthcare is a differentiating point. Perdue has also tried to pivot the election towards identity issues and paint Ossoff as an extreme socialist candidate who wants to defund the police.

Polling trends show the race is quite close though the trend for Perdue from recent polls looks better than for Ossoff, giving him a slight edge in our forecast.



Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Vinod Bakthavachalam

Vinod Bakthavachalam

I am interested in politics, economics, & policy. I work as a data scientist and am passionate about using technology to solve structural economic problems.