Election 2016: Forecasting the House


Surjit S. Bhalla, Abhinav Motheram and Rohini Sanyal

(November 8, 2016; 8 am EST)

The world media is currently abuzz with the forecasts for US Presidential Elections and the many opinion polls conducted by pollsters such as FiveThirtyEight and New York Times Upshot, have successfully managed to fuel the number of followers for the same. However, not many have explored the ambit of House Elections and the factors to forecast the distribution of parties. The US House of Representatives have a total of 435 seats, all of which are up for election. Currently, the House has a Republican majority but do the Democrats stand a chance at winning majority House?

A cursory glance at 2016 US House Election forecasts might not lead you to much information – not as much as for the Presidential race and Senate at least. The reasons for this are multifold but the prime reason is that non-partisan pollsters don’t poll house races often. Even among the few dozen competitive house races, the polls conducted during each cycle are often from partisan pollsters who tend to leak results and represent best case scenarios for the campaign they are affiliated to. Furthermore, factors such as incumbency, quality of candidates fielded against incumbents, the dollars they raise also make a huge difference that renders the House races somewhat difficult to model.

Nevertheless, several independent newsletters such as Cook Political Report and Sabato’s Crystal Ball give us an insight into the congressional elections. While Cook Political Report uses a Partisan Voting Index (PVI) with a 7-point scale rating, ranging from Solid Democrat to Solid Republican to show how strongly a congressional district or state leans towards either party compared to the nation as a whole; Sabato’s Crystal Ball tracks individual races by looking at state/district level polls in addition to other qualitative factors, some of which are mentioned above. The latest reports of both these newsletters suggest that the Democrats require not only all of the Toss Ups but a considerable portion of the Lean Republicans as well to win the House Elections. If we look at the trend lines tracking the changes in these ratings over a period of time, it would come as no surprise that more than two-thirds of these changes have gone towards Democrats in both reports suggesting an underlying movement towards Democrats.

Another way of modeling the House races uses a more aggregate and quantitative approach – examining historical election results in order to compare the change in the number of seats each election and using fundamentals such as GDP, president’s approval rating, incumbent party, etc., as input variables. Alan Abramovitz’s “Generic Ballot Forecasting Model” is one such example that is based on three predictors – the number of Republican Seats at stake in the election, the “general ballot” question in national polls and lastly, whether it’s a midterm election under a Democratic or Republican president. Although the model shows significant effects in both House and Senate elections, it explains about 82% of the variance for the former which is considerably better than the 63% for the latter. This happens due to the much smaller number of seats for Senate as compared to House. Given two of the three variables are already constant for 2016 election, the only dependent variable is the generic ballot question in the national polls. If we consider the Huffington Pollster average of +5 to the Democrats, the model gives the party less than 15% probability of regaining control of the House.

Another fundamental based model is the “Political Economy Model” developed by Michael Lewis-Beck and Charles Tien. Similar to the Alan Abramovitz model, this also takes three input variables – Political Popularity, Economic Conditions and Midterm Status – to forecast the Presidential race and the change in House & Senate seats. The model forecasts a tight presidential race as well as the senate but provides a House majority for the Republicans with 97% probability.

The key point to be noted here is that Democrats need to win 30 seats or more this November to regain control of the House. Unless there’s a landslide win for Democrats, most models (both qualitative and quantitative) suggest Democrats are likely to see an increase in the number of seats in the House, but not enough to regain the House itself. We see this kind of change of seats in mid-term elections in the years 2006, 2010 and 2014 where voters turnout against incumbent presidential party. However, one has to go back as far as 1964 to see an incumbent president’s party gaining as many House seats as Democrats need to secure a majority in 2016.

What if Trump turns out to be another Goldwater as the presidential polls suggest and Hillary Clinton rides roughshod over Trump? Here we highlight an important factor – the percentage of third party/undecided voters in the presidential elections. Clinton’s standing in FiveThirtyEight’s national polling average is at its highest this cycle with 46% and Trump’s at 41%. In other words, about 15% of the electorate isn’t yet committed to Clinton or Trump, as compared to just 5% who weren’t committed to President Obama or Mitt Romney at this point in 2012, which is quite a stark number. Even though it remains to be seen how the choices of these undecided voters translates into support for candidates in the ballot races, this should at least bring up the discussion of House races more widely.

Forecasting the election to the House of Representatives is truly tricky. Very few pollsters attempt this difficult task. We undertake it with some trepidation. Our model, with a 10 % victory margin for Clinton, results in “too close to call” result with a slight 8 seat edge to the Republicans i.e. 215 seats for the Democrats and 223 seats for the Republicans.

Table 1 reports various forecasts for Election 2016 and Table 2 reports on forecasts specifically for the House.

Table 1: Final Estimates of a Clinton Win!

Clinton / Democrats

Trump / Republicans




Real Clear Politics



Five Thirty Eight



Winning Probability



Huffington Post



Market 365



Five Thirty Eight



Electoral College Votes



Huffington Post 325 213
Five Thirty Eight 294 243
House Elections
Own 215 223
Real Clear Politics 201 234
Cook Political Report 188 247
Senate Elections
Own 51 49
Five Thirty Eight 50 50
Notes: For methods and details, please see articles contained in http://www.ssbhalla.org

 Table 2: Various Congressional Elections Forecasts

Sl. No Forecaster/Team Forecast Date



Projected Seats


    Rep Dem Solid Rep Lean Rep Toss Up Lean Dem Solid Dem Rep Dem Rem**
1 Cook Political Report 25-10-16 237 198 201 24 25 8 177 209 177 49
2 Centre for Politics 20-10-16 233 202 204 21 17 11 182 206 179 50
3 Real Clear Politics 27-10-16 236 200 191 35 21 18 171 205 176 54
4 Alan Abramovitz’s Generic Ballot Forecasting Model 25-08-16 231 204
5 Michael Lewis-Beck & Charles Tien’s Political Economy Model 29-07-16 244 191
Notes: ** Remainder seats

Click here to download this article in PDF

This Presidential Election was never close


The debates are over, and while the formal election date is November 8th, there are few undecided voters left. The latest RealClearPolitics estimate of the vote is a 6 percentage point lead in favor of Clinton, 48:42. Before the first presidential debate, it was a close tied race at 43:42; it was a close race as well before the revelations of lewd comments by Trump became known on Oct. 7

A detailed analysis of opinion polls, as well as historical demographic patterns of voting (propensities to vote according to  age, sex, race) indicate that this race has always been for Ms. Clinton to lose.  However, the media, and pollsters, have, until recently, seen the race as too close to call. Why this uncertainty about forecasting a landslide for Ms. Clinton? According to historical demographics, Clinton has been in a double-digit lead since end-July.

I believe there are two major reasons for this hesitancy.

First, the widespread fear that the anti-globalization Brexit virus is contagious and could spread across the Atlantic. The pollsters did not foresee this anger in England, and don’t want to make the same mistake in forecasting the scale of  Trump defeat. There is a major problem with this “correlation”. In a referendum, you vote for one issue; in elections, you vote for multiple issues embodied in each candidate. The probability of reaching the same “outcome” in a referendum as in a Presidential  election is very small.

A second factor is the widespread belief that there is a considerable amount of lying in the opinion polls i.e. the polls are understating Trump’s strength because many  voters are hesitant to admit that they would vote for him.. Many believe that missing out on this lying factor got pollsters, and others, mistaken about Brexit, and Trump in the primaries.

The lying factor can be tested. Opinion polls ask several questions relating to political views, in addition to a direct question about voting choices. The former, indirect questions, can be assessed for their revelation about “true” choices. Towards this end, forty-six questions (e.g. favorability of candidate, ability to handle an international crisis, trustworthiness etc.)  from five different pollsters were pooled. [see Table – All polls before the final debate on October 20]. The striking result – the opinion questions reveal the same result as the polling questions i.e. lying may not be a factor in this election.  Clinton is ahead by 9 percentage points (ppt) for the larger question set, and ahead by 11 ppt for a set of five specific opinions.

A noteworthy feature about Trump’s popularity is that no matter what the criteria or question, he finds it very difficult to cross 42 percent of the vote. Thus, the evidence overwhelmingly converges to this simple reality – Trump has a core support of only 30 to 35 % of the electorate.

A large fraction of this core support is expected to come from the white, less educated population. The latest data on educational attainment (US Census Bureau) indicates that there are 80 million white individuals (out of 190 million) above the age of 18 and with zero college education.  Assuming a voter turnout of 60 percent, this yields 48 million “likely” voters for Trump. Assume 80 percent of these potentially disgruntled 48 million white voters are certain to vote for Trump (the reader can insert her own preference).  Hence Trump receives 38 million votes from this demographic, or a 29 million voter advantage.

However, this vote advantage is “cancelled” because of the expected, overwhelming support for Clinton from the 40 million strong non-white voter population (see Bhalla for a detailed analysis) What one is then left with then are approximately 50 million well-educated white voters.

Election 2016 is about how this educated 50 million strong voting group votes (comprising of 35 million with a college degree).  If this group splits 65:35 in favor of Ms. Clinton, Clinton will have 15 million extra votes. For a voting population size of 138 million, that is an 11 percent margin.

What does history tell us about double digit Presidential win

There are only 13 Presidential elections since 1828 where a candidate has received less than 41 percent of the vote. In five elections, third-party candidacy was a major factor affecting vote shares of the two major parties. Since 1948, there have been three elections without significant third party presence. This list has many honorable politicians and none said that he may not accept the result of the election. The three honorable losers: Goldwater, 1964, 38.7 %; McGovern, 1972, 38.2 %; Mondale, 1984, 40.8 %.

A Trump vote of 41 % will reach the maximum of the above three honorable losers, and is close to the upper bound that Trump is likely to obtain. It would also make Trump “equal” to that of a kinder, gentler Mondale. The third party candidates are unlikely to reach double digits, say 8 %. This leaves the prediction of 51 % for Clinton.


Surjit S Bhalla is Contributing Editor, Indian Express and some of his election research can be found on ssbhalla.org. He tweets @surjitbhalla


US Election 2016 – Not Even Close
Clinton Trump
(%) (%)
Weighted average of 46 responses to questions from 5 different pollsters* 48.5 37.8
Favoribility of candidate 47 40
Judgement to serve as President 53 37
Foreign Policy 55 37
Social Security & Medicare 50 42
Immigration 50 44
Average above five questions 51 40
Average of 48 polls**,  Sep 18-Sep 25, 2016 43.1 41.7
Average of 54 polls**,  Oct 11-Oct 18, 2016 44.9 40
Indian Express, Aug. 6, 2016˖ 50 41
Indian Express, Sep 27, 2016** 51 39
* The five pollsters considered were Fox News, Reuters/Ipsos, Washington Post, NBC & YouGov
** Polls using Likely Voter samples
˖Bhalla, Surjit S., Hillary Clinton by a landslide, Indian Express, 6 Aug, 2016
˖˖Bhalla, Surjit S., A double-digit win for Hillary Clinton, Indian Express, 27 Sep, 2016


Article Written on October 24, 2016 & Posted on November 04, 2016

click here to download this article in pdf