The Republican advantage in enthusiasm and intensity evident during the less-public phase of the 2010 campaign continues to pose serious obstacles to Democratic gains in Texas in the latest University of Texas/Texas Tribune poll. We timed our fall surveys to take snapshots of the last moment of the summer phase of the campaigns in early September and the period when Texans began to turn more attention to the election in October.
The October 2010 poll, as you now know, shows Rick Perry leading Bill White 50 percent to 40 percent among self-reporting registered voters, with Republican candidates in other statewide races enjoying double-digit leads and the results of the generic congressional ballot favoring the Republican Party by 18 points. (A summary of results can be downloaded here; as always, complete results including the data file will be available by the end of the week both here and at the UT government department’s Texas Politics website.)
The Perry-White result includes the responses to a follow-up question posed to those who responded “don’t know” about their preference for governor. This was a different approach from the September UT/Tribune poll, conducted at a time when we thought many voters had yet to turn their attention to the election. About 15 percent of the October sample said they were undecided in their initial response. When we pressed respondents to express a preference, equal numbers chose Perry and White, adding 5.4 percent to each candidate’s totals. Libertarian Kathie Glass gained an additional 2 percent from these “pushed” undecided respondents, and Deb Shafto, the Green candidate, gained another point. This left the undecided number at less than half a percentage point.
The enthusiasm gap
There has been a great deal of talk in this election cycle about the “enthusiasm gap.” The definition is straightforward: It is the tendency of one partisan side (in this case, the Republicans) to tell pollsters that they are more interested in this election and more likely to vote. We have certainly seen evidence of this in the UT/Texas Tribune polls this year. Across all polls, Republicans have been 9 points more likely than Democrats to say they are “extremely interested” in politics. The direction of the enthusiasm gap has reversed itself from 2006 and 2008, when Democratic voters — especially African-Americans and younger voters — were more likely to show interest and a determination to cast their ballots. The main question associated with this phenomenon in 2010, however, is whether it will manifest itself in a disproportionately pro-GOP electorate.
From the perspective of the pollster, the enthusiasm gap creates some difficulties. Initially, it tends to create response biases. That is, those who are more enthused are more likely to agree to talk to a pollster about politics. In 2010, it has been particularly difficult to get younger people and ethnic minorities to respond to our polls. But there are ways to deal with this issue, including multiple attempts to reach individuals from these groups selected for our samples and (of course) weighting these individuals' responses to representative proportions in the poll.
The other issue associated with the enthusiasm gap involves estimating the likely electorate. Like most polling organizations, we have been reporting our trial ballot results among registered voters throughout 2010. As the election nears, though, we wanted to estimate who is a likely voter within our poll and report the ballot numbers for that sub-sample. Almost all polling organizations make this switch in October, as voters focus in on the races and interest and engagement rises. ABC/Washington Post, CBS/New York Times, NBC/Wall Street Journal, CNN/Time, Fox News/Opinion Dynamics and Gallup have all reported their numbers for likely voters in their most recent polls. Predictably, since these likely voter models rely on self-reported interest and turnout intent, they have produced significantly more Republican-leaning results than we find among all registered voters. In the case of Gallup’s initial likely voter results, a single-digit Republican edge in the generic ballot for Congress swelled to a 19-point GOP advantage. Gallup did offer a less “restrictive” likely voter screen (their “high turnout” model), which showed a still huge 12-point Republican lead.
Who will vote?
In light of all this, we faced two important decisions: (1) how to estimate the likely electorate for 2010 and (2) whether and how to report our numbers. Let us dispense with the second question first: We decided to report our numbers among registered voters in our official reports and press releases for the sake of continuity. That is, we wanted people to be able to compare, for example, Perry’s lead over White over time and not cause confusion about any dynamics in the race with a switch in methodology. We also decided, however, to estimate a likely voter model and report the results here. We should emphasize that we are not “hiding” these numbers. We make clear that our top-line results are for all registered voters. Moreover, transparency is one of our primary goals. Perhaps most notably, the UT/TT data are available for download, so people can create their own turnout models.
So how to create a likely voter model? Campaign pollsters typically use a combination of past voting history — available off the registered voter list — and current interest and engagement. Those who have voted in the past, as well as those who are jazzed about voting this year, tend to get into the likely electorate.
Political scientists often use more elaborate, multivariate models. These take into account demographic and political characteristics, self-rated interest and self-rated past vote behavior and are specified based on examinations of past polls in which scholars have gone back after the election and checked to see which respondents actually voted.
Since we are not sampling off of the registered voter list and do not have reliable, validated turnout data from Texas, we rely on self-rated interest and turnout history. This is not ideal, but almost all of the major national polls use this method, and their results have tended to be very close to the actual results on Election Day. More particularly, our likely voters are those who:
1. Say they have voted in “every” election or "almost every" election over the past three to four years and say they are either "extremely" or “somewhat interested” in politics.
2. Say they have voted in past elections "some of the time" over the past three to four years and say they are “extremely interested” in politics.
How the screens affect the results
Given the undoubted inflation of both past turnout and current interest, this is a pretty generous screen, with 600 of 800 registered respondents qualifying as likely voters. And how does this affect the race for governor? Among all respondents in the sample, Perry leads White 50 percent to 40 percent, with just less than 1 percent undecided. Among likely voters, however, Perry leads White 55 percent to 35 percent, with 1 percent undecided. As you might expect from those numbers, those who are least likely to vote are also younger, more racially and ethnically diverse and more Democratic.
Other likely voter screens look even worse for the Democrats. What if we only took those 356 respondents who said they vote in every election? Perry leads White by 24 points (57 percent to 33 percent). The margins of error in these calculations increase as the voter screens reduce the sample sizes.
It is possible that Democrats will become more engaged and interested as Election Day approaches. This is certainly what President Barack Obama, Speaker Nancy Pelosi and Texas Democrats are counting on. And we will not be able to measure this (if it does occur), as this will be our final poll of the cycle. But in Texas, the enthusiasm gap in late October is very real, and it could be the difference (as Democratic pollster Peter Hart has said) between a Category 3 and a Category 5 hurricane.
This article originally appeared in The Texas Tribune at http://www.texastribune.org/2010/10/26/different-voter-models-different-poll-results/.