Skip links

THE ELECTORAL BAROMETER VS. THE POLLS

A question that keeps arising in connection with this year’s presidential race is why, given the extremely favorable political environment for Democrats, Barack Obama has not been able to establish a clear lead over John McCain in national polls. Based on President Bush’s extraordinarily low approval ratings (28 percent in the most recent Gallup Poll), a deteriorating economy, and the fact that Republicans have controlled the White House for the past eight years, the political environment this year appears to be one of the worst for the party in power in the past sixty years.

The Electoral Barometer, a measure of national political conditions that combines these three factors, currently yields a reading of -62, which is similar to that received by Jimmy Carter prior to his decisive loss to Ronald Reagan. Yet according to the most recent realclearpolitics.com average, Barack Obama is only leading John McCain by about four percentage points in recent national polls.

Should we place more weight on polls showing a neck and neck race between Obama and McCain or indicators of the political environment that predict a decisive victory for the Democratic candidate?

In order to answer this question, I used data on the results of trial heat polls in every presidential election since 1948 that were generously provided by James E. Campbell. I conducted regression analyses to compare forecasts based on the Electoral Barometer with forecasts based on trial heats conducted at three different times during the campaign: late June, early September, and immediately before the November election.

Table 1 displays the results of the four regression analyses. The results show that the Electoral Barometer is a much better predictor of the outcomes of presidential elections than trial heat polls, including those conducted immediately before the election. The model based on the Electoral Barometer explains over 90 percent of the variance in the incumbent party’s share of the major party vote with a standard error of 1.7 percentage points.

In contrast the June trial heat model explains only 38 percent of the variance with a standard error of 4.4 percentage points, the early September trial heat model explains only 73 percent of the variance with a standard error of 2.9 percentage points, and the final pre-election trial heat model explains only 82 percent of the variance with a standard error of 2.4 percentage points.


Table 1. Explanatory Power of Electoral Barometer vs. Trial Heat Forecasting Models, 1948-2004

Electoral Barometer June Trial Heat September Trial Heat Final Trial Heat
Adjusted R2 .902 .382 .727 .815
Standard Error 1.75 4.39 2.92 2.40

Source: Trial heat data provided by James E. Campbell; other data compiled by author.


The data displayed in Table 1 indicate that the national political climate is a better predictor of the outcomes of presidential elections than trail heat polls, even those conducted immediately before the election. It is a much better predictor than polls conducted in June.

Another question that can be asked, however, is whether using the results of trial heat polls along with the Electoral Barometer provides a more accurate prediction than using the Electoral Barometer alone. In order to answer this question, I conducted regression analyses using both Electoral Barometer scores and trial heat results. The results are displayed in Table 2.


Table 2. Regression Analyses of Incumbent Party’s Share of Major Party Vote Using Electoral Barometer and Trial Heat Results, 1948-2004

Independent Variable June Trial Heat September Trial Heat Final Trial Heat
Barometer .941 .731 .641
Trial Heat .018 .277 .372
Adjusted R2 .894 .926 .942
Standard Error 1.82 1.53 1.35

Note: Entries shown are standardized regression coefficients.

Source: Trial heat data provided by James E. Campbell; other data compiled by author.


The data displayed in Table 2 show that the June trial heat results add nothing to the accuracy of the forecast based on the Electoral Barometer alone: the estimated coefficient of the June trial heat variable is negligible and the adjusted R2 of the model is slightly smaller than that of a model using the Electoral Barometer alone. Adding trial heat poll results from September or later to the Electoral Barometer model does produce a more accurate forecast than using the Electoral Barometer by itself, but the improvement in accuracy is fairly small.

For example, the model using early September trial heat results along with the Electoral Barometer has a standard error of 1.5 percentage points compared with a standard error of 1.7 percentage points for a model using the Electoral Barometer alone.

These results indicate that the Electoral Barometer is a much better predictor of the results of presidential elections than trial heat polls, especially those conducted relatively early in the election year. Based on this evidence, polls currently showing a very close race between Barack Obama and John McCain should be taken with a large grain of salt. The Electoral Barometer is currently pointing very strongly in favor of the Democratic candidate.


Dr. Alan Abramowitz is the Alben W. Barkley Professor of Political Science at Emory University, and the author of Voice of the People: Elections and Voting Behavior in the United States (2004, McGraw-Hill). He can be contacted via email at polsaa@emory.edu.