For the February 2018 edition of the University of Texas/Texas Tribune Poll released today, we decided to sample primary voters differently than in previous attempts. This time, estimates for our primary election trial ballots were based on a sample of likely voters with a verified record of primary vote history, as opposed to relying on a subset of voters who express requisite levels of political interest and participation.
While it may be tempting to think that this change in process is a result of some notable state-level polling errors in the 2016 election, the impetus for this change actually began back in 2014. During that election cycle, our estimates for the Republican gubernatorial and Senate primaries turned out pretty well, but that’s where the plaudits ended.
Since then, we’ve been thinking, obsessing even, over how we might approach the next off-year primary Election given our experience in 2014, but also in light of the broader discussion about the volatility in polling, and the ever-increasing difficulties of reaching a representative sample.
The simplest version of our prior method to identify likely primary voters hinged on self-reported attitudes — in particular, on their stated interest in politics and public affairs in conjunction with their stated vote history over the last few years. In some cases, we included one or more questions intending to assess their likelihood to vote in the upcoming election as well. In some cases, this approach works relatively well. While people tend to overstate their interest in politics, their past voting history and their likelihood to vote, in general elections, the difference between people who we included in the “likely voter” pool and those people who actually turned out to vote (a smaller group by all measures) have never been so different as to terribly bias the estimates of those trial ballots. As the actual electorate gets smaller, and more distinct, this is less likely to be the case. A primary electorate fits this description, as we’ve recently written, and a primary electorate in an off-year election is even smaller, and more distinct still.
In the February 2018 poll released today, we rely on past vote history to determine whether or not we should consider someone to be a likely voter, and in particular, past primary voting history. For a respondent’s opinion to be considered in our primary trial ballot estimates, he or she had to have participated in a Texas party primary in 2012, 2014, or 2016. The benefit to this approach is that we’re not relying on what people tell us, but on what they have actually done. The major limitation to this approach is that if there is a major upsurge in participation that brings new people into the process who haven’t regularly participated in primaries, then we might miss some of these new voters. At this point, we’ve decided that past behavior is likely to be a more reliable guide, and, for what it’s worth, that mirrors the approach taken by campaigns when they conduct surveys for their own internal research.
The reality is that the simple description above actually removes one element of the complexity of this process. The UT/TT poll is a survey of registered voters in Texas, and we’ve been collecting estimates on the attitudes of registered voters since 2009, resulting in over 30 publicly available surveys over that time span. The initial process of the current poll was no different. As in the past, we collected 1,200 responses from registered voters on all of the attitudes we usually measure: presidential job approval, job approval for statewide elected leaders, assessments of the economy in Texas and the country as a whole, and attitudes on topics like immigration and Hurricane Harvey relief. We’re able to report the current attitudes and any changes that have taken place over time for registered voters, as we have been doing. But among these registered voters, only a fraction have the vote history that we’ve described above — unsurprising given the extremely low levels of primary participation in Texas. More importantly, that’s not enough primary voters to make reliable estimates for the trial ballots.
In order to overcome that challenge, we augmented our initial sample in the new poll with an additional 424 confirmed primary voters, known in polling lingo as an ‘oversample.’ We’re reporting survey results on the primary trial ballots — the so-called horse races — from this sample alone. We’re reporting results for registered voters for the remainder of the survey, as we’ve always done. As always, readers will find the detailed description of the survey methodology in the documents released by the Texas Tribune and in the polling section of the Texas Politics Project website.