Skip links

Forecasting the Midterm Elections

COPYRIGHT 2009 SABATO’S CRYSTAL BALL


The 2010 midterm election is still 14 months away. Fourteen months is a lifetime in politics. We don’t know how many House and Senate incumbents from each party will be retiring, how many incumbents from each party will be facing serious challengers or what the national political climate will be like in the fall of 2010. Nevertheless, based on what we already know and the evidence from midterm elections over the past six decades we can make an educated guess about what is likely to happen in next year’s House and Senate elections. The results of a statistical analysis of congressional election results since World War II indicate that Republicans are almost certain to make at least modest gains in the House of Representatives and could pick up a few seats in the Senate. However, their chances of regaining control of either chamber appear to range from slim in the case of the House to none in the case of the Senate.

There are two basic approaches to predicting the results of congressional elections: the seat-by-seat approach and the statistical forecasting approach. The seat-by-seat approach involves conducting district-by-district and state-by-state analyses to identify individual House and Senate seats that are likely to switch parties based on factors such as retirements, incumbent popularity, challenger quality, fundraising and polling data. This is the approach used by such well-known election analysts as Stuart Rothenberg, Charlie Cook, and of course Larry Sabato and it can produce very accurate predictions when used by a skillful and experienced analyst. The Crystal Ball, in particular, has a remarkable track record for accuracy in both congressional and presidential elections. But the seat-by-seat approach relies heavily on information that is not available until the last few weeks of the campaign and the final predictions are generally released only a few days before the election.

An alternative approach to forecasting congressional elections involves predicting the overall seat swing based on a statistical analysis of variables known to affect the national outcome such as the type of election (midterm or presidential), the number of seats held by each party going into the election, the president’s approval rating and the generic congressional ballot. The statistical forecasting approach has also proven to be highly accurate and it can produce forecasts with a longer lead time than the seat-by-seat approach. In 2006, for example, a forecasting model that I developed predicted more than two months before Election Day that Democrats would gain 29 seats in the House of Representatives–a much larger pickup than most pundits and commentators were expecting at that time. The actual result was a 30 seat Democratic gain. A similar model that I developed for presidential elections has correctly predicted the winners of last five presidential contests more than two months before Election Day with an average margin of error of one percentage point.

In order to produce an accurate prediction, the congressional forecasting model requires information on the president’s approval rating and the generic congressional ballot from the summer of the election year. However, based on factors that are already known such as the type of election and number of seat held by Democrats and Republicans going into the election, we can calculate the range of likely outcomes. And based on the current values of the president’s approval rating and the generic ballot we can predict what the House and Senate results would be if the national political climate next summer is the same as it is now.

We already know that 2010 will be a midterm election year. That means that there is a very high probability that Democrats will lose seats in the House of Representatives and a better than even chance that they will lose seats in the Senate as well. The tendency of the president’s party to lose seats in Congress, especially in the House of Representatives, is one of the best known regularities in American politics. Table 1 presents data on the results of 16 midterm elections that have taken place since the end of World War II. The president’s party has lost House seats in 14 of these 16 elections and Senate seats in 12 of 16 with an average loss of just over 24 seats in the House and between 3 and 4 seats in the Senate.


Table 1.


While losses for the president’s party have been the norm in midterm elections, the data in Table 1 show that the pattern has been more consistent in House elections than in Senate elections and that the losses have been considerably larger in second or later midterms than in first midterms. In the first midterm election after a party takes over the White House, the average loss has been only 17 seats in the House and one seat in the Senate; in second or later midterms, on the other hand, the average loss has been close to 30 seats in the House and between 5 and 6 seats in the Senate. The good news for Democrats is that 2010 will be their first midterm after taking over the White House last year.

There are several theories that attempt to explain why the president’s party almost always loses seats in midterm elections. Surge and decline theory argues that midterm elections represent a return to normal voting patterns following presidential elections in which short-term forces can produce unusual gains for the winning candidate’s party. Negative voting theory argues that those who are dissatisfied with the status quo are more motivated to turn out and express their discontent in midterm elections than those who are satisfied. And balancing theory argues that, knowing that the president will be in office for the next two years, some voters seek to provide greater balance in government by electing members of the opposition party to Congress. All of these theories may be partially correct. Whatever the explanation, midterm elections are generally not kind to the president’s party.

The other key predictor that is already known is the number of seats held by each party going into the election. Democrats currently outnumber Republicans by 257-178 in the House of Representatives. They outnumber Republicans by 60-40 in the Senate but of the 36 seats up next year, 18 are held by Democrats and 18 by Republicans. These numbers are very important because they measure each party’s exposure to risk in an election: as we will see, the more seats a party has to defend, the more seats it tends to lose. This tendency is especially strong in the Senate so the fact that Democrats will be defending only half of the seats that are up next year should help to limit their losses. In 2012, however, Democrats will face a much more difficult situation because they hold 25 of the 34 seats that will be up for election that year.

Table 2 displays the results of regression analyses of the outcomes of House and Senate elections since World War II. The dependent variable in the House model is the change in the percentage of seats held by Republicans while the dependent variable in the Senate model is the change in the number of seats held by Republicans. The independent variables in the two models are identical–the percentage or number of seats held by the Republican Party going into the election, the incumbent president’s net approval rating (approval-disapproval), the Democratic lead or deficit in the generic ballot in the Gallup Poll in late August or early September, and two midterm election variables–one for first midterm elections and one for second midterm elections. Themidterm election variables are both coded to indicate whether a given election is a Republican midterm or a Democratic midterm.


Table 2.


The results in Table 2 indicate that these models perform quite well–the estimated coefficients for all of the independent variables are statistically significant and the House model explains over 80 percent of the variation in seat swing in the lower chamber while the Senate model explains two-thirds of the variation in seat swing in the upper chamber. It is not surprising that the House model explains a larger proportion of the variation in seat swing than the Senate model. Seat swings in House elections are based on 435 individual races while seat swings in Senate elections are generally based on only 33-35 individual races so the Senate results can be swayed more easily by idiosyncratic factors. Nevertheless, the estimated effects of the independent variables in the two models are fairly similar.

By far the biggest difference between the House and Senate results is in the estimates for the effects of the exposure to risk variable–the percentage or number of seats held by Republicans going into the election. The coefficient for the Senate variable is about twice the size of the coefficient for the House variable: because of the greater competitiveness of Senate elections, there is a greater risk of losing a Senate seat than a House seat with all other factors held constant.

The coefficients for the presidential approval and generic ballot variables indicate that the more popular the president and the better the president’s party performs on the generic ballot question, the fewer seats the president’s party can expect to lose in both the House and Senate. However, even after controlling for presidential approval and the generic ballot results, both midterm variables have significant negative effects. With all other factors held constant, the president’s party can expect to lose an additional 14 House seats and 2-3 Senate seats in a first midterm election and an additional 22 House seats and 3-4 Senate seats in a second midterm election.

Bottom Line

Based on the estimates in Table 2, the number of House and Senate seats currently held by Republicans, and the fact that 2010 will be a first midterm election under a Democratic president, we can calculate conditional forecasts of House and Senate seat swing depending on President Obama’s net approval rating and the Democratic margin or deficit on the generic ballot question. The results are displayed in Tables 3 and 4.


Table 3.



Table 4.


In order to become the majority party, Republicans would need to gain 40 seats in the House of Representatives and 11 seats in the Senate. The data in Tables 3 and 4 indicates that this is virtually inconceivable in the case of the Senate and very unlikely in the case of the House. However, the results also indicate the Republicans are almost certain to make at least modest gains in the House and have a good chance to make a small gain in the Senate.

Under what might be considered a worst case scenario for Democrats, if President Obama’s approval rating sinks into the low 40s next year, which would produce a net approval rating of around -10, and Republicans take a 5 point lead on the generic ballot, the GOP would still be expected to gain only 4 seats in the Senate. However, such a scenario would put Republicans in position to come very close to regaining control of the House with an expected pickup of 41 seats. On the other hand, if the President’s approval rating rebounds into the mid 60s, producing a net approval rating of around +30, and Democrats have a 10 point lead on the generic ballot, the GOP would be expected to lose one seat in the Senate and gain only 15 seats in the House. Based on the latest results (as of August 24) for the President’s net approval rating in the Gallup Poll (+16 percent) and the Democratic lead or deficit on the generic ballot (+6 percent), the predictions would be a Republican pickup of 1 seat in the Senate and 23 seats in the House.

Implications

Democrats are likely to lose at least 15 seats in the House of Representatives in 2010 and their losses could go as high as 30-40 seats. The Senate looks more promising for Democrats because there are as many Republican as Democratic seats up for election next year but a loss of 3-4 seats is entirely possible. Given the deep partisan divide in both chambers, diminished majorities will make it much more difficult for Democrats to pass any major legislation in the next Congress. If anything, Republican leaders emboldened by a successful election are likely to be even less interested in compromise with the White House and Democratic leaders than now. If Democrats can’t pass health care, carbon caps, and immigration reform in the current Congress, they probably won’t have another chance until at least 2013.


COPYRIGHT 2009 SABATO’S CRYSTAL BALL