Sabato's Crystal Ball

Republican Senate Majority on the Way?

One model says yes

Harry Enten, Guest Columnist May 5th, 2011

Here at the Crystal Ball we want to encourage the work of promising young political scientists. This week, we are featuring the work of Harry Enten, a student at Dartmouth College who has an interesting take on next year’s Senate elections. Perhaps Harry’s prediction is a bit too optimistic for the Republicans, but this far out from the election it is hard to rule out his model’s prediction. One thing is clear, though: We expect that readers will be hearing a lot more from Harry as time goes by, and we thank him for his contribution. ~The Editors

If the recent budget debate has taught Americans anything, it is that the power of the gavel in Congress can be as powerful as the executive pen in the White House. In the blossoming 2012 campaign, we should, therefore, focus not only on the presidential election, but also the elections for Congress. Most analysts, myself included, believe that the Democrats face an uphill battle to re-capture the House. The main question is whether Republicans can pick up the necessary four (or three, if the Republicans win the presidency) seats to take control of the Senate. A model I have created suggests the GOP will not only gain the majority, but in fact control 54 seats in the next Senate.

Like predictions for the House of Representatives, forecasts for the Senate can be done one of two ways.

1. A seat-by-seat analysis. This method is most accurate in the closing days of an election, when polling data have been shown to be very accurate.

2. An aggregate forecast based on national variables. This method is less accurate, but can be employed without knowing certain characteristics about individual races. The model I present here is a national swing model developed with data from every presidential year Senate election since 1952.

The model predicts the percentage of seats up for reelection that the president’s party will win in the Senate. To do so, it utilizes five variables:

1. The percentage of seats up for reelection controlled by the president’s party. Senate elections, unlike those in the House, are held a third at a time. In 2010, Democrats only had to defend 19 of 37 (51%) seats and were able to maintain control of the Senate by winning 13 (35%). In 2012, Democrats control 23 (or 70%) of the seats up, while Republicans only need to defend 10 (30%). Even if the Democrats were able to win 58% (well more than half) of the Senate elections in 2012, they would still lose four seats and control of the Senate.

2. The number of military fatalities per million of the general population over the last four years in unprovoked foreign conflicts started by the president’s party or maintained for more than four years under the current president’s party. If President Obama had started a war with fatalities, we would expect his party to lose seats in the Senate. He has not started or continued for more than four years any war. This variable has no value for 2012.

3. Weighted quarter-on-quarter growth in real disposable income per capita (a variable originally described by Douglas Hibbs) over the current president’s term. Obama’s current projected real DPI growth (using Bureau of Economic Analysis data for past quarters and Wells Fargo’s forecast for the future) would rank 11th of 16 presidents since 1952. Not surprisingly, this slow growth does not bode well for the president’s party in the Senate.

4. An interaction variable between quarter-on-quarter real DPI growth and a simple dummy variable for whether the president’s party is in control of both houses of Congress. The president’s party does better when Congress is controlled by his party and quarter-on-quarter real DPI growth is positive, but does worse when growth is negative. Since Republicans control the House, this variable takes no value for 2012.

5. A Congress dummy variable for whether the president’s party is in control of both houses of Congress. This variable is only included for the model to be statistically correct (due to the interaction term), but it is not statistically significant to any great effect. It has no value for 2012.

Employing this data, we can estimate that at this point the model projects the Democrats will only win about 49%, or about 16, of the Senate seats up this cycle. If correct, this loss of seven seats would flip the Senate to the Republicans. The loss would parallel the Republican loss in 2008 of 8 seats.

But should the model’s results be believed?

On the one hand, the model does explain a relatively high percentage (87%) of the variation between Senate results since 1952. Within the model’s dataset, the margin of error with 95% confidence is only about three seats. If the model were instituted before the 2008 election, it would have correctly projected an eight-seat Republican loss. And although a seven-seat loss would be on the high end of the spectrum of the very early seat-by-seat analyses, it is not a great outlier.

On the other hand, the model only takes into account data from 15 elections. It is quite possible that the 2012 election ends up being unique in some unforeseen way. The Democrats may recruit unusually strong Senate candidates that stem the national tide. President Obama may win a large victory and have abnormally long coattails. The model cannot account for any of these possibilities.

Another possibility is that the real DPI forecast is too pessimistic. If, somehow, Obama experienced Reagan-like growth in the final seven quarters of his first term, the model would predict Democrats to win about 57%, or 19, of the Senate seats up this cycle. While that would still leave them short of a majority, they would be close enough to be well within the model’s in-dataset margin of error. For this reason alone, the model will need updating as we get actual (not forecasted) economic data from the Bureau of Economic Analysis.

At this point, this much is certain:

1. The Democrats are in great danger of losing their Senate majority in 2012. History is against them maintaining control of the Senate.

2. We will get a better gauge of the exact Senate picture when we know the candidates running in each race.

3. The true test of this Senate model’s predictive value will come in November 2012.

Harry Enten, a senior at Dartmouth College, is pursuing a degree in government with a concentration in statistics and elections. He has previously interned at the NBC Political Unit in Washington D.C. and Pollster.com. He blogs at Margin of Error and tweets @ForecasterEnten.