by Alan Abramowitz, Alben W. Barkley Professor of Political Science, Emory University
It’s déjà vu all over again. The 2006 midterm elections are only seven weeks away and the Gallup Poll likely voter screen is back to its old tricks. Last week Gallup released its first poll of the 2006 campaign using its likely voter screen and the results showed that Democrats and Republicans were tied on the generic ballot question. That’s the question that asks respondents whether they would vote for the Democratic or Republican candidate for the House of Representatives in their district if they election were being held today. The question doesn’t provide respondents with the names of the candidates—hence the term “generic ballot.” Nevertheless the generic ballot question has proven to be a fairly good predictor of the results of congressional elections once the sample is pared down to those likely to vote. But that’s not easy to do, especially more than seven weeks before Election Day.
At first glance, it might appear surprising that the new Gallup Poll found a tie on the generic ballot. After all, one week earlier another Gallup Poll produced a 12 point Democratic advantage on this question, similar to that found in most other national polls in recent weeks. But, as Charles Franklin and Mark Blumenthal recently explained, the shift from a 12 point Democratic lead to a tie wasn’t really as dramatic as it seemed. It was almost entirely a result of the introduction of the likely voter screen. Among all registered voters, the new Gallup Poll found a 9 point Democratic lead on the generic ballot.
Almost all polling organizations use some sort of likely voter screen to measure voting intentions shortly before an election. This is especially critical in a midterm election in which, historically, only about 40 percent of eligible voters go to the polls. However, the Gallup likely voter screen is more complicated than most—it involves a series of seven questions including one that measures current political interest. As a result, while it appears to work quite well in the last day or two before an election, it can produce highly volatile results when applied several weeks before an election. During the 2000 election campaign, for example, Gallup’s presidential preference results sometimes gyrated wildly over the course of only a few days—on October 4th, Gallup showed an 11 point lead for Al Gore. Three days later, on October 7th, they showed an 8 point lead for George Bush. However, much of this apparent volatility was caused by the likely voter screen rather than actual shifts in candidate preference among voters.
Some of the volatility in Gallup’s results in 2000 may have been caused by their use of a tracking poll with a 3-day rolling sample. Tracking polls tend to be more volatile than standard polls because they allow less time for callbacks. During the 2004 campaign, however, Gallup’s likely voter screen sometimes produced large swings in candidate support over relatively short time periods. For example, between October 10th and October 16th, Gallup reported a shift from a 2 point Kerry lead to an 8 point Bush lead. In contrast, the CBS/New York Times Poll found a 1 point Bush lead on October 11th and a 1 point Bush lead on October 17th and a Time Magazine Poll found a 1 point Bush lead on October 7th and a tie on October 15th.
Why does Gallup’s likely voter screen produce greater volatility in candidate support than likely voter screens used by other polls? The answer is not clear but a reasonable guess would be that it has to do with the use of a question asking about a respondent’s current level of political interest. Political interest may vary considerably depending on what types of news stories are making the headlines. When the President delivers a major address to the nation for example, supporters of his party are likely to be more interested in news about the speech than supporters of the opposing party. In the final days before an election, interest in the campaign and knowledge about where one’s polling place is located are probably good predictors of voter turnout. Seven weeks before an election, however, these questions may not predict turnout very well.
The 9 point gap between registered and likely voters in Gallup’s recent generic ballot results appears overly large by historical standards. Don’t be surprised to see the gap close by the end of the campaign. In the meantime, however, don’t be surprised to see big fluctuations in the results of Gallup’s generic ballot question among likely voters.