Election Update: Barnes Withdraws from Wisconsin Governor’s Race
Today, former Lieutenant Governor Mandela Barnes withdrew from the Democratic gubernatorial primary in Wisconsin. In our survey, which was sponsored by UpNorthNews and Main Street Alliance, we asked voters who had not yet voted for their second choices; if we allocate Barnes, Rodriguez, and Hughes voters who had not yet cast a ballot to the candidates they said were their second choice, Hong’s lead barely changes. In that scenario, she carries 47 percent of the vote, far ahead of Crowley at 19 percent, Roys at 6, and Brennan at 5. The three candidates who dropped out still have some vote share (mostly from people who already cast their ballots): Rodriguez has 3 percent, Barnes has 2 percent, and Hughes has less than 1 percent. Seventeen percent remain undecided.
Of course, it’s likely that these numbers will have shifted since our survey was conducted. Voters have been able to vote early in-person since Monday, so many more voters are locked into their vote choice than when our survey closed on Sunday morning. In any case, with Barnes’ vote share splitting between the other candidates, Hong’s lead is still likely to be insurmountable; the real question is whether she can secure a majority of the vote.
Given Barnes’ dropout and our desire to be the most accurate pollster of this contest, State Navigate will be re-entering the field next week with a short recontact survey to see how vote choices have changed in the primary among our respondents who had yet to cast a ballot, so stay tuned.
A Note on Our Polling Methodology
It’s been a while since we’ve written a full article detailing our polling methodology. While our Executive Director, Chaz Nuttycombe, has often Tweeted out who we’ve been attempting to poll when our polls are in the field, and we’ve occasionally livemapped raw responses in real time, some still have questions about our polling methods, including about the nuts and bolts of how we polled Wisconsin. Here are some answers:
State Navigate uses a polling method called text-to-web to reach potential voters. We use i360 to obtain lists of registered voters. Sometimes, for smaller area polls, like a poll of a closed Republican primary in a state house district, we send text messages to i360’s entire universe of registered Republicans in that house district with a cell phone number that is supposedly reliably linked to the voter’s record on the voter file. For larger area polls, we usually pull a random sample of reliable cell phone numbers for the election universe we want to poll on Day 1, and then supplement and recontact undersampled groups on later poll days.
Our Wisconsin poll wasn’t our first polling rodeo, but it did raise some unique challenges that we hadn’t met before. Wisconsin has open primaries and no party registration, so we had to figure out a way to recruit mainly likely Democratic voters into the sample. Initially, we thought we’d be able to do that by mainly pulling records of people who had previously voted in at least one Democratic primary over the past 4 even-year cycles and new registrants. We fairly quickly learned that Wisconsin doesn’t even record which voters voted in which party’s primary, and confirmed this with the Wisconsin Elections Commission. We scrambled for another solution.
On Day 1, we used a voter file to target mainly voters that i360 had modeled as Liberal and Swing voters who had voted in at least 1 prior primary, with a supplement of voters who registered after Election Day 2024. But because it was an open primary, we also targeted a small sample of Conservative voters with prior vote history to see how many of those voters might cross over and vote in the Democratic primary.
We had our texting vendor, Alliance Forge, send 30,000 texts on Day 1: 10,000 to modeled Liberals with primary voting history, 10,000 to new registrants, and 5,000 each to modeled Swing and Conservative voters with primary voting history. All recipients received the following message: “Hey [First Name]! Please take State Navigate’s 2 minute Wisconsin Democratic Primary survey! [URL] Stop = end.” These texts were sent from a Milwaukee-area 414 area code to Eastern Wisconsin and out-of-state cell phone numbers, and a Madison-area 608 area code to Western and Northern Wisconsin cell phone numbers.
Even though the text specifically said Wisconsin Democratic Primary, we also asked all voters who had already voted or said they definitely wouldn’t vote in the primary if they voted in or would vote in another party’s primary. (We include voters who say they definitely will not vote in our sample at a low weight because research shows some people who say they definitely won’t vote ultimately do.) We terminated the survey for those who said they had voted or would vote in another party’s primary.
We typically track the raw results and responses in real time. About 27,300 of the 30,000 Day 1 messages went through to valid cell phone numbers. 391 voters completed the poll within 24 hours, for a 1.43% same-day response rate. We also tracked response rate and responses among relevant demographic groups, including 18-29 year-old-voters, in real time.
Usually, we have serious problems getting responses from younger voters and have sent Zoomers a specially crafted all-lowercase text, which has slightly increased response rates in our prior polls. There was no need to do that here – 18-29 voters were responding at similar or even higher rates than the rest of the sample. Earlier this year, voters in Southwest Virginia were answering our Virginia Constitutional Amendment poll at higher-than-expected rates, and that turned out to be a sign of higher enthusiasm for the No vote there. Here, most younger voters are voting for Francesca Hong, and their relatively higher response rate in our survey may turn out to be a signal of higher-than-usual turnout in the primary.
We almost always have some loose sense of what our weights will be up front, but after the Virginia experience, we are willing to adjust expected weights up or down slightly based on the responses we receive. We were having some issues getting responses from the other usual suspects: minority voters, non-college voters and, to a lesser extent, female voters. When some demographic groups are underrepresented in our surveys, we send a follow-up text the next day to members of those groups, asking them to help make our poll representative.
On Day 2, we sent about 16,600 recontact texts, mainly to those underrepresented groups, and 30,000 new texts – 20,000 to modeled Liberal voters with primary voting history, 5,000 to modeled Moderate voters with primary voting history and 5,000 to Liberal non-white voters with primary voting history. We didn’t add new Conservative voters because it became quickly clear they weren’t responding at the same rate, largely because they indicated they were unlikely to vote in the Democratic primary, and those that did respond were largely undecided, anyway.
Our problem reaching female voters was mostly solved after Day 2, so on Day 3, we only sent about 6,600 recontact texts to minority and non-college voters, and new texts to 25,000 new potential voters. 10,000 of the 25,000 were randomly pulled from non-white Liberals with or without vote history who registered before November 2024, 5,000 randomly pulled from all Liberals with the same criteria, 5,000 randomly pulled from all Moderates with the same criteria, and 5,000 randomly pulled from non-college Liberals with primary voting history.
We also sent 375 texts to those who had started our poll on Day 1 or Day 2 but not finished it, provided they had not opted out of future texts or answered that they were voting in another party’s primary, asking them to please complete the survey. 20 of the 373 who received texts that went through took us upon that offer – 5.4%.
Overall, we tried to send 108,616 texts to 85,000 Wisconsin cell phones. About 98,400 reached a recipient. 1,274 people provided valid responses (completed the survey and did not respond multiple times), before culling a small number (20) who appeared to be lying about their demographic information or voting plans, for about a 1.3% response rate for all texts and 1.5% response rate to all phones reached after recontacts.
