Generic Ballot Forecasting Model: Democrats Could Take Back Senate but Republicans Likely to Hold House With Reduced Majority
August 25th, 2016,
Since the conclusion of the Republican and Democratic national conventions last month, pundits, political reporters, and ordinary Americans have, for understandable reasons, been preoccupied with developments in the presidential campaign. And the contest between Hillary Clinton and Donald Trump has certainly provided plenty of material for serious political observers as well as late night comics. With the presidential contest getting so much coverage in the national media, however, much less attention has been devoted to the critical battle for control of the next Congress. Regardless of the outcome of the presidential election, whether Republicans or Democrats control the House and Senate will have enormous consequences for the direction of the country and the ability of the next president to carry out his or her agenda.
At present, Republicans hold a 247 to 186 seat majority in the House of Representatives (with vacancies in two formerly Democratic-held seats that the party will easily hold onto). All 435 House seats and 34 of the 100 Senate seats are up for election this year. In reality, however, only around 50 House seats and perhaps a dozen Senate seats are really in play — the rest are completely safe for one party or the other. Nevertheless, there are enough seats in play that there is some uncertainty about which party will end up in control of the House and a great deal of uncertainty about which party will end up in control of the Senate.
So what should we expect in the House and Senate elections this year? A simple forecasting model based on three predictors — the number of Republican seats at stake in the election, support for the two major parties on the “generic ballot” question in national polls, and whether it is a midterm election under a Democratic or Republican president — yields fairly accurate predictions of seat swing in the House and Senate. The number of Republican seats at stake is a measure of exposure to risk: the more seats Republicans have at stake, the more seats they are likely to lose. The generic ballot, based on polls asking voters whether they would prefer a Democratic or Republican candidate for the U.S. House without naming the actual candidates, is an indicator of the national political climate. Finally, the midterm election variable indicates whether an election is a Republican or Democratic midterm. There is a strong tendency for the party holding the White House to lose seats in midterm elections. However, 2016 is a presidential election year, so the midterm variable is not relevant.
Table 1: Results of regression analyses of House and Senate election outcomes, 1946-2014
Notes: Dependent variable is change in Republican seats; SEE = standard error of estimate. No generic ballot data available for 1948.
Sources: Data compiled by author
Prediction equations for 2016 are presented in Table 1. The estimated weights for the three predictors and intercepts (constants) are based on OLS regression analyses of data on U.S. House and Senate elections between 1946 and 2014. The results indicate that for both House and Senate elections, all three predictors have substantial and highly statistically significant effects. As expected, seat exposure has large and highly significant negative effects on seat swing in both House and Senate elections, the generic ballot variable has substantial and highly significant effects in both types of elections, and the midterm variable has large and highly significant effects in the expected direction in both types of elections. However, while all three predictors have strong and statistically significant effects for both types of elections, the model is much more accurate for House elections than for Senate elections, explaining about 82% of the variance in House seat swing compared with only about 63% of the variance in Senate seat swing. This is exactly what one would expect given the much smaller number of Senate seats at stake in each election and the larger proportion of competitive Senate races.
Two of the three predictors in this model are already set for the 2016 election. We know that Republicans will be defending 247 of 435 House seats and 24 of 34 Senate seats this year, and we know that 2016 is a presidential election year so the midterm variable does not favor either party. The only unknown for 2016 is the value of the generic ballot variable. Table 2 therefore presents conditional forecasts for House and Senate seat change based on values for the generic ballot variable ranging from a two-point Republican lead to a 14-point Democratic lead.
Table 2: Conditional forecasts of change in Republican seats in House and Senate
Sources: Table 1 and data compiled by author
The results in Table 2 indicate that for almost any conceivable values of the generic ballot variable, Democrats are likely to make gains in both the House and Senate. That is largely due to the fact that, as a result of their successes in the 2010 and 2014 midterm elections, Republicans are defending unusually large numbers of seats in both chambers this year. However, the results indicate that in order for Democrats to gain the minimum of four seats they need to regain control of the Senate (if there is a Democratic vice president to break a 50-50 tie), they probably would need a lead of at least two or three points on the generic ballot and to gain the minimum of 30 seats they need to regain control of the House, they probably would need a lead of at least 13 points on the generic ballot.
According to HuffPost Pollster, results of recent national polls give Democrats an average lead of five points on the generic ballot. If that lead were to hold up until the week after Labor Day, the traditional cutoff date for the generic ballot forecast, Democrats would be expected to gain about 16 seats in the House and about four seats in the Senate — not enough to flip control of the House but enough to flip control of the Senate if Clinton wins the presidential election.
Of course any forecasts based on a statistical model are subject to a margin of error. In this case, the results in Table 1 indicate that if Democrats maintain a five-point lead in the generic ballot, they would be very likely to pick up between six and 26 seats in the House and between two and six seats in the Senate. They would have about a 50% chance of regaining control of the Senate (if there is a Democratic vice president) but less than a 15% chance of regaining control of the House.
The Political Science Election Forecasts of the 2016 Presidential and Congressional Elections, Part 3
August 25th, 2016
|Dear Readers: This is the latest in a series of political science forecasts for the 2016 races for the White House and Congress. We’ll be featuring forecasts from nine different individuals and/or teams this year, which James E. Campbell is assembling as part of a project for PS: Political Science and Politics that we are also featuring in the Crystal Ball. These models are based on factors such as the state of the economy, polling, whether an incumbent president is running for reelection, and other indicators. They can often be a better predictor of the eventual results than polls alone, and many are finalized months before the election.
We are pleased to feature the work of the many top political scientists who have built these models, both in an attempt to predict the outcome of the election and, more importantly, to identify the factors that actually affect presidential and congressional races. Below, Campbell lays out the details and outlook of his congressional forecasting model. Additionally, we have updated our running tally of presidential and congressional forecasts to include a new version of the presidential forecast by Robert Erikson and Christopher Wlezien, as well as the forecasts for Alan Abramowitz’s congressional model that are discussed separately in this week’s Crystal Ball.
— The Editors
The Seats-in-Trouble House and Senate Election Forecasts
By James E. Campbell of the University at Buffalo, SUNY
The twin Seats-in-Trouble forecasting models of congressional elections predict the net two-party aggregate seat change for the Democratic Party in the House and Senate, respectively. These are essentially hybrid models combining the virtues of intensive district-by-district and state-by-state in-depth handicapping of congressional races with the value-added rigor of the careful construction of an aggregate index and the systematic statistical evaluation of its historical relationship to seat changes in past elections.
The models use the late-August competitiveness ratings of congressional races from the Cook Political Report. Based on an evaluation of the correspondence of these individual race ratings to their election outcomes in the past, indices of each party’s number of vulnerable seats were calculated. The aggregate difference between the parties was used as an index and then included in a regression analysis to estimate its relationship to actual net seat changes. The seats-in-trouble index might be regarded as an extension of Oppenheimer, Stimson, and Waterman’s exposure model (1986). Rather than the number of seats nominally in jeopardy, the seats-in-trouble index counts those seriously threatened.
Both models have proven quite accurate in real-time application. The House model has been used since the 2010 election (though simplified in 2014). It was quite accurate in predicting the 2010 Republican wave and off by only six seats in 2012 and three in 2014. First used in 2014, the Senate model’s forecast missed the 2014 Republican nine-seat gain by a single seat.
For House elections, the seats-in-trouble index is the difference between the number of current Democratic seats that are rated as only leaning Democratic or worse and the number of current Republican seats in similar jeopardy.
For Senate elections, the index is slightly more stringent. It is the net number of a party’s seats (Democrats minus Republicans) rated as toss-ups or worse. The indices use ratings available in late August of the election year.
Based on seven House Democratic seats being rated as only leaning to the Democrats, toss-ups, or tipped toward going to the Republicans, and 33 House Republican seats being rated as only leaning to the Republicans, toss-ups, or tipped toward going to the Democrats — a net of 26 more Republican than Democratic seats-in-trouble — the model predicts that Democrats will gain 32 House seats in November. This would bring the number of House Democrats up to 220 members, two seats more than required for a bare majority. The forecast was made on Aug. 18, 2016.
Based on one Senate Democratic seat being rated as a toss-up or tipped toward the Republicans and eight Senate Republican seats being rated as toss-ups or tipped toward the Democrats — a net of seven more Republican than Democratic seats-in-trouble — the model predicts that Democrats will gain seven Senate seats. This would bring the number of Senate Democrats (including two Independents who caucus with Democrats) up to 53 seats, a majority. The forecast was made on Aug. 19, 2016.
Based on the distribution of out-of-sample errors, the probability of a Democratic House majority (the model erring in favor of the Republicans by more than 2 seats) is 61%. The probability of a Democratic Senate majority is 88% with the election of a Democratic president (a 50/50 Senate split) and 73% with the election of a Republican president (a 51D/49R Senate split).
Table 1: Forecasts of the 2016 two-party presidential vote
Table 2: Forecasts of the 2016 U.S. House election
Table 3: Forecasts of the 2016 U.S. Senate election
August 25th, 2016,
With Hillary Clinton taking a large lead in the polls following the Democratic National Convention, journalists have begun to discuss the extent to which Democrats may be able to capitalize on these gains in down ballot races for the House and Senate. As has been the case in every recent election cycle, some journalists have even begun to write about whether or not 2016 could be a wave election.
Given the popularity of this concept among journalists and elections experts, being able to classify elections as either a “wave” or “not a wave” would appear useful. Unfortunately — and surprisingly given the widespread use of this term — there is not a precise definition of this concept. To try to correct this, I have developed my own definition that combines both scholarly rigor with the basic intuition of a wave election being a “big win” for one side at the expense of the other.
Specifically, I define a “wave election” to be a congressional election that (1) produces the potential for a political party to significantly affect the political status quo as (2) the result of a substantial increase in seats for that party.
What exactly do I mean by “significantly affect the political status quo?” Past wave elections — at least those almost everyone agrees are waves — have produced some sort of policy upheaval, either through the passage of consequential new policy (e.g. the Affordable Care Act after 2008), or by blocking policy (e.g. the virtual halt in legislative action on President Obama’s priorities after 2010). If the party gaining seats in the election either wins a majority or exceeds the average number of seats it held over the course of the previous decade, then the seat gain is considered substantial enough — at least potentially — to affect policy in a meaningful way.
As for the “substantial increase in seats for that party” portion of my definition, the seat gain for a party must exceed the average seat increase for the gaining party in the previous decade’s elections. Measuring seat gains based on recent elections accounts for differences in electoral volatility over time and allows this definition to be applied to any political era. Some political observers were shocked by the double-digit seat gains in the House in 2006, 2008, and 2010 elections after the electoral stability of the early 2000s, but the late 1800s frequently featured double-digit seat swings. Indeed, even with a smaller House of Representatives, all but three elections from 1862 to 1898 saw one of the major parties gain or lose at least 20 seats.
One additional condition for this portion of my definition should be added based upon an observation made by political scientist Angus Campbell in his 1960 article, “Surge and Decline: A Study of Electoral Change.” In this article, Campbell notes that a “surge” in congressional seats by the president’s party in presidential election years is often followed by a natural “decline” in midterm years. Similarly, political scientists Bruce Oppenheimer, James Stimson, and Richard Waterman note in their 1986 publication, “Interpreting U.S. Congressional Elections: The Exposure Thesis” that congressional parties tend to lose seats in Congress when they are “exposed” — that is, have more seats than is typical for the party in that time period.
To account for the fact that the seat swing in a congressional election is linked to the seat swing in the election that preceded it, my definition requires that the seat swing in a congressional election must exceed any gains for the other party in the previous election. For example, while the Democratic Party gained 26 seats in the 1982 midterm election, this seat gain did not exceed the 34 seats Republicans gained in the previous election — and thus 1982 is not classified as a wave election under my definition.
When determining whether a Senate election is a wave election, I make one minor tweak to account for the fact that Senate elections are staggered. When calculating whether a Senate election is a wave, I use an average of the seat gain in the previous election and the previous time that Senate class faced the voters (six years prior).
If — and only if — an election satisfies both parts of my definition does that election then get classified as a wave election.
So what does this definition mean for 2016? First, for the House of Representatives, Democrats must gain at least 28 seats if this election is to be classified as a Democratic wave in the House. To satisfy the first part of my definition pertaining to “potential to significantly affect the political status quo,” Democrats would have to reach at least 215 seats (a net gain of 27 seats) based upon their average level of seats following the previous decade of congressional elections. And to satisfy the second part of the definition, Democrats would need to gain 28 seats in order to exceed the average net partisan gain over the previous decade. As an election needs to meet both parts of my definition in order to be counted as a wave, Democrats would have to gain at least 28 seats in order for 2016 to be counted as a wave.
This seat gain would put Democrats two seats short of a majority, but would create difficult circumstances for Speaker Paul Ryan as near-unanimous cooperation in the Republican caucus would be necessary in order to pass legislation opposed by Democrats. Given the recalcitrance of members of the House Freedom Caucus on a number of votes in recent years (including the vote for Speaker of the House), Speaker Ryan would probably have to rely on some Democratic votes to pass legislation. It is also theoretically possible that Democrats would be able to find a few party-switchers before the House convenes in early January.
As for the Senate, Democrats would need to gain at least eight seats for this election to be classified as a Democratic wave in that chamber. The first part of my definition minimally requires Democrats to win the majority by gaining four seats (or five seats if Mike Pence becomes vice president), while the second part of my definition requires an eight-seat gain so as to exceed the average gain of the GOP in 2010 and 2014 where Republicans gained an average of 7.5 seats (six seats in 2010 and nine seats in 2014).
Although large Democratic seat gains may not materialize in the House or the Senate in 2016, this definition provides a consistent method for determining whether an election is a wave, yet one that is flexible enough to be applied to elections in different political eras. Using this definition, all those who follow elections can better understand the magnitude of electoral volatility in a specific election.
|Jacob Smith is a Ph.D. Candidate in the Political Science Department at the University of North Carolina-Chapel Hill, where he studies congressional elections, Congress, and public policy. He is especially interested in examining how political behavior and political institutions affect one another. He first constructed a definition for wave elections as part of his senior thesis at Kenyon College.|
1. For example, Republicans gained 13 seats in 2014, so a value of 13 is used in the average for this election.
2. The GOP would also have to gain 28 seats for this election to count as a Republican wave under my definition, which seems unlikely given current electoral conditions and the large House majority Republicans possess.
3. Again, as a near impossibility since November is almost certain to yield GOP losses in the Senate, Republicans would need to gain at least seven seats in order for 2016 to be considered a GOP Senate wave under my definition.