1. Survey sampling and weighting

Baselice & Associates (Baselice) conducted the tenth wave of the Texas Voter Poll from August 22 to 26, 2026, completing 1,369 interviews with Texas registered voters and reporting results on a weighted base of 1,001 interviews. The report refers to these respondents as Texas voters or registered voters.13 Interviewers screened for registered Texas voters and concluded the interview for anyone who works in the news media, in market research, in elected office, or on a political campaign.  

Baselice drew the sample from two sources. Of the 1,369 respondents, 675 came from a registered-voter file and were reached by cell phone (458 interviews), text message (184) or landline (33). Those records include each voter’s history of turnout and 86 percent of the 675 had voted in at least one November general election since 2020. The remaining 694 interviews came from an online panel (for which no primary election turnout records are available). Any analysis that uses turnout history therefore covers the 675 file-matched interviews only.  

Every respondent self-reported a county of residence, and respondents in the six largest counties also gave a ZIP code which Baselice used to manage interview quotas. All interviews were conducted in English. Median interview length was about 22 minutes by phone, 13 minutes by text message and 11 minutes on the online panel.  

Baselice weights the sample within twelve cells, formed by crossing the two interview sources with six media-market regions. Within each cell the firm rakes14 the sample to population targets on gender, party, race and age (and education for the online panel only). Baselice reports a margin of error of plus or minus 3.1 percentage points for the weighted base of 1,001 interviews. Statistical weighting makes some interviews count for more than others. Examining the plausibility of the weights we find that the weights here stay in a narrow, plausible range common for polls this size. That is, about nine in ten interviews count no more than about twice as much as the average interview, and only one counts more than four times as much. When statistical weighting is used, counting interviews unequally means the estimate leans more on some respondents than on others, and their answers have more leverage (move the average estimate more), but in this survey the sampling design helped ensure that weighting did not greatly magnify a small number of voices. Because party is one of the weighting targets, the weighted size of any group defined partly by party, including the Unaffiliated and Undecided voters moves with that weighting choice. 

Baselice randomized (or rotated) the answer choices on several questions, so that the order in which respondents heard them would not shape the results (to protect against the influence that the ‘ordering’ of questions a respondents encounters could have in priming or influencing their later answers). For instance, the 13 candidate-position statements were asked in random order, the two party-image questions rotated their first two choices, the vote-intention question rotated top-to-bottom against bottom-to-top, and the race question randomized its first five choices. Baselice asked two questions in two different wordings, one to each random half of the sample, these were the question on prioritizing water projects (672 and 697 interviews) and the question on whether Texas is acting now to remain among the best states. Respondents who answered “unsure” when asked which electricity and grid issue matters most were not asked the follow-up question about their second choice, so we treat those 81 interviews as missing on the follow-up rather than as a substantive answer. 

We report percentages to one decimal place except when comparing to external surveys or published sources with other reporting formats. Combined categories, such as “extremely” plus “very” concerned or “much” plus “somewhat” more likely, are computed from the unrounded weighted shares, so a combined figure can differ by a tenth of a point from the sum of the rounded values shown in a chart. 

Throughout this report we call a difference a difference only when it is larger than the variation the sample itself can produce. All margins use a simple-random-sample calculation at the 95 percent confidence level, the same basis as the plus or minus 3.1-point figure Baselice reports for the full sample, without an adjustment for the design effect of weighting.  

2. Identifying Unaffiliated and Undecided voters

We wanted to know how many Texas voters had not settled on a choice for the November election when we interviewed them in late August 2026, and describe those voters and their views on salient policy issues. The report calls them Unaffiliated and Undecided voters, and in other contexts, voters like these are called swing, target, or persuadable voters.   

How we define an Unaffiliated and Undecided voter 

The definition has two parts, and a voter must meet both. We combine two pieces of information: the party grouping Baselice builds from each respondent’s voting history and party image, and one survey question that asks how settled or decided the respondent’s November choice is. 

1. The party grouping separates committed partisans from everyone else. A committed partisan, in this context, is a voter who has tended to vote for one party’s candidates in past November elections, holds a positive image of that party and a negative image of the other. We call a voter who does not fit that description unaffiliated. Roughly half of Texas voters (49.9 percent of the weighted sample) are unaffiliated in this sense; the group includes soft Republicans, soft Democrats, self-described Independents and voters who named no party. 
2. The decision question offers five answers: fully decided, mostly decided, mostly undecided, fully undecided, and unsure. We label a voter undecided when the answer is mostly undecided or fully undecided. 

We count a respondent as Unaffiliated and Undecided when both of the following are true: the respondent is unaffiliated, falling outside the two committed partisan groups, and answered either mostly undecided or fully undecided. We do not count committed partisans, whatever they said about their decision, and we do not count unaffiliated voters who said they were fully or mostly decided. On this definition, 201 of the 1,369 respondents are Unaffiliated and Undecided, 16.7 percent of the weighted sample. 

We exclude from this measure 83 respondents, 5.9 percent of the weighted sample, who are unaffiliated and answered unsure about their decision. An unsure answer fits a decided voter who preferred not to say as well as an undecided voter who could not say, so it does not tell us which group the respondent belongs in. 

Examining the quality of our Unaffiliated and Undecided categorization  

For the 675 respondents drawn from the registered-voter file (includes the phone and SMS text respondents sub-group only), we can triangulate our measure with external administrative voting recordings by comparing what they told us with what they have done previously. We used records for whether each voter turned out in each November general election since 2020 and which party’s primary, if any, they voted in during 2020, 2022, 2024 and 2026. Those records do not depend on anything the respondent said to us which makes them an external check on our measure.   

Among respondents who have voted in two or more primaries since 2020, the primary voting records show the Unaffiliated and Undecided group do not differ from decided unaffiliated voters in how many general elections they have voted in (though they do vote in primaries at a lower rate than the rest of the respondents). Switching between the two parties’ primaries is rare among all voters; yet, the switching that does distinguish this Unaffiliated and Undecided group happens between the primary and the general election. Of the Unaffiliated and Undecided voters who voted in a party’s primary in March 2026, more than half (57%) now say they will vote a mix across parties (or are unsure), compared with 20.5% of decided unaffiliated voters and 3.8% of core partisans who voted in a March primary (Fisher’s exact test for the two unaffiliated groups, p < 0.001).  

We also checked whether Unaffiliated and Undecided voters plan to vote for the party they say they usually support. The survey asked which party’s candidates the respondent has tended to vote for in past November elections, and then separately how the respondent intends to vote this November. Among Unaffiliated and Undecided voters who named a party on the first question, half do not plan to vote that party’s ticket in November; nearly all of them say they will vote a mix across parties or are unsure. Among decided unaffiliated voters who named a party, 21% do not plan to vote that party’s ticket, and among core partisans about 7% do not (Fisher’s exact test, undecided versus decided unaffiliated voters, p < 0.001). Taken together, these checks help show that our two-part Unaffiliated and Undecided voters definition separates a group whose plans for November have come loose from both their past primary ballots and the party they say they usually support, at rates well above those of decided unaffiliated voters and core partisans. 

3. Sorting voters into urban, suburban and rural groups

To compare voters in Texas’s largest metropolitan areas with those in mid-size and smaller metropolitan areas and those outside any metropolitan area, we sort all 254 Texas counties into three groups and assign each respondent the group of the county they gave as their residence. 

The three categories follow the 2023 National Center for Health Statistics Urban-Rural Classification Scheme for Cou15nties, which the Centers for Disease Control and Prevention and the Substance Abuse and Mental Health Services Administration also follow: 

• Large metropolitan: every county in a metropolitan area of one million residents or more. 34 counties, 916 interviews, 68.1 percent of the weighted sample. 

• Small and medium metropolitan: every county in a metropolitan area of fewer than one million residents. 52 counties, 300 interviews, 21.1 percent. 

• Rural: every county outside any metropolitan area. 168 counties, 150 interviews, 10.5 percent.