Who Will Win in November?

Using the Generic Vote Question to Forecast the Outcome of the 2002 Midterm Election

Alan I. Abramowitz

Alben W. Barkley Professor of Political Science
Emory University
Atlanta, Georgia 30322
Phone: 404-727-0108
Fax: 404-727-4586
E-mail: polsaa@emory.edu

         As we enter the final weeks of the 2002 election campaign, most expert observers expect little change in the makeup of the House of Representatives. In his most recent seat-by-seat analysis, Charlie Cook of The National Journal sees the Republicans holding a slight advantage entering the final month of the campaign.  According to Cook, 219 of the 435 seats in the House appear to be safe for the Republicans or leaning Republican while only 204 appear to be safe for the Democrats or leaning Democratic.  Similarly, Congressional Quarterly’s most recent analysis of the House elections shows Republicans favored or leading in 223 contests with Democrats favored or leading in only 203 contests. 

         But is it possible that these expert observers, as knowledgeable as they are about individual House races, are missing the big picture?  In some previous elections such as 1994, expert analysts have underestimated a national trend favoring one party or the other.  Could the experts be missing a national trend favoring one party or the other in 2002?  Is it possible that public concern about such issues as the war on terrorism or the faltering economy could lead to significant gains for either Republicans or Democrats on November 5th?

         Statistical forecasting models provide an alternative approach to seat-by-seat analyses for predicting the outcomes of House elections.  Until now, however, such forecasting models have had very limited success.  The most accurate such model, one developed by Gary Jacobson of the University of California at San Diego, utilizes the incumbent president’s approval rating, the change in real per capita income in the year preceding the election, and a measure of seat exposure (the number of seats held by the president’s party compared with its normal strength), along with an adjustment to incorporate an increase in Republican strength since 1990, to predict seat gains or losses by the president’s party.  However, despite its complexity, this model only explains about 70 percent of the variance in the outcomes of House elections since World War II. 

         This paper proposes and tests a new and relatively simple forecasting model for midterm House elections based on the results of the Gallup Poll “generic vote” question.  Prior to every midterm election since 1950 with the exception of 1986, the Gallup Poll has asked a national sample of likely voters which party they plan to vote for in the election to the House of Representatives from their district. 

         Even though the generic vote question does not provide respondents with the names of the Democratic and Republican candidates in their own district, it is a fairly accurate predictor of the number of seats won by the parties in midterm elections.  By regressing the number of seats won by the Democrats in midterm elections on the Democratic lead or deficit in the generic vote, we obtain the following result (with the estimated standard error shown in parentheses):

Predicted Dem Seats = 3.67(.550)*Generic + 226.1. 

         The generic vote variable by itself explains about 80 percent of the variance in the outcomes of midterm elections since 1950.  The standard error for this single variable model is only 12.3 seats—much smaller than that of any previous forecasting model.

The estimated coefficient of 3.67 for the generic vote variable means that every one percentage-point increase in the generic Democratic vote is worth an additional 3.67 Democratic seats in the House of Representatives. 

         While these results are impressive, we can improve the accuracy of the generic vote model by adding two additional predictors: the number of House seats held by the Democratic Party before the midterm election and the president’s party affiliation.  The number of House seats held by Democrats before the election measures the advantage of incumbency—the more seats with a Democratic incumbent, the more seats the Democrats should win.  The president’s party affiliation, a dummy variable taking on the value 1 if there is a Democratic president and 0 if there is a Republican president, should capture the impact of negative voting: in midterm elections, discontented voters are more likely to vote against the president’s party than contented voters are to vote for the president’s party.

         Regressing the number of seats won by the Democrats in midterm elections on the Democratic lead or deficit in the generic vote, the number of Democratic seats in the previous House, and a dummy variable for the party of the president, we obtain the following results (with the estimated standard error for each predictor shown in parentheses):

 Predicted Dem Seats = 3.00(.278)*Generic + .453(.070)*Previous Seats – 19.9(4.06)*Democratic President + 127.3.

                 Despite the small number of elections available for analysis, all of the estimated coefficients are highly statistically significant.  The model now explains an impressive 96 percent of the variance in the outcomes of midterm elections since 1950 and the standard error of the model is only 5.4 seats.  According to these results, every one percentage point increase in the generic Democratic vote is worth an additional three Democratic seats in the House of Representatives, every seat held by Democrats before the election is worth an additional .45 Democratic seats after the election, and the presence of a Democrat in the White House costs Democrats an average of just under 20 seats in a midterm election.

         Based on these results, what should we expect in the 2002 midterm election?  The following table displays the number of House seats that the model predicts will be won by Democratic candidates based on the Democratic lead or deficit in the generic vote in the final Gallup Poll before the election. 

         The results of the most recent Gallup Poll, based on interviews conducted from October 3-6, show Democrats leading Republicans in the generic vote by 48 percent to 47 percent, with five percent undecided.  Based on this one point lead, the three-variable generic vote model predicts that Democrats will win 226 seats in the 2002 midterm election while Republicans will win 209 seats.  This would be a pickup of 14 seats for the Democrats.  Even if the Democrats are even with Republicans or trailing by one point in the generic vote, the model predicts that they will regain control of the House, although by a very narrow margin.  Thus, in contrast to the expert observers, the generic vote forecasting model gives Democrats an excellent chance of regaining control of the House of Representatives on November 5th.