Internal Polls Flopped in Michigan and Hawaii. How Will They Do in Minnesota and Wisconsin? 

In a mid-July column, I detailed how internal polls, combined with scarce independent polling, warped perceptions of Graham Platner’s electoral resilience in Maine’s Senate race before he bowed out. I further raised an eyebrow at the flurry of internal polls showing fat leads for Abdul El-Sayed over Representative Haley Stevens in Michigan’s Democratic Senate primary, when other springtime polls depicted a much closer race.  

Prompted in part by the internal polls, the former Wayne County health department chief and left-wing insurgent was crowned the frontrunner by mainstream media outlets. Several independent polls very late in the campaign echoed El-Sayed’s internals, firming up the media narrative. Two days before Election Day, a rumor shot through the X social media platform that Stevens’ own pollster told the candidate she was down 14 points and had “no path to victory.” The pollster denied it, but the rumor further fed a perception that the cake was baked. 

Yet my earlier skepticism proved warranted. El-Sayed beat Stevens, but by a single percentage point, not the double digits forecast by the surveys affiliated with his campaign.  

The colossal miss raises the question of whether El-Sayed’s internal poll flex led to a case of “herding,” the polling phenomenon once described by Nate Silver as “file-drawering (not publishing) results that don’t match the consensus or torturing their turnout models until they do.” Following the surprise Michigan result, The New York Times noted the challenge for pollsters to anticipate the demographic mix of primary election turnout: “Conducting surveys in partisan primaries has long presented a challenge to pollsters, as it requires them to make assumptions about who will actually turn out to vote. Without mandatory party registration in Michigan, pollsters must guess who will show up on Election Day.” Fearful of becoming an embarrassing outlier, “independent” pollsters had an incentive to make like-minded guesses. 

One pollster who worked for a pro-El-Sayed political action committee, Zenith Research’s Adam Carlson, published a post-mortem in which he acknowledged his turnout model “undershot” demographic groups favorable to Stevens: senior, low-income, and rural voters. One possible explanation Carlson offered is that his company uses “text-to-web” technology to reach out to potential respondents, as opposed to more expensive “live calling,” which may have tilted the respondent pool towards younger votes more partial to El-Sayed. 

Less noticed over the weekend was Hawaii’s First Congressional District Democratic primary, which pitted moderate incumbent Ed Case against progressive challenger State Senator Jarrett Keohokalole. No independent pollster surveyed the race. Then on August 6, two days before the primary, two sets of Keohokalole’s internal polls were leaked to Jewish Insider: one from early July with Case up 8 and one in late July with Keohokalole seizing a 7-point lead. The attempt to show late momentum proved futile. Case won by 21 points.  

Perhaps the only immediate victims of the Keohokalole poll push were bettors on political prediction markets, as in the final hours of the campaign, for no obvious reason but this tiny wisp of poll data, traders on the Kalshi platform suddenly deemed Keohokalole the favorite and pushed his win-probability to 70 percent. But as I noted in last month’s column, we all have reason to worry about being manipulated by internal polls: “the decline of professional polling from media outlets leaves us more vulnerable to spin polls from campaigns. That vulnerability was cannily seized by the Platner campaign, to the detriment of Maine Democrats who perhaps wanted to believe a little too much what the numbers seemed to say.” In the case of Michigan, false perceptions that El-Sayed was a lock risked misleading eligible voters that they need not bother to vote.  

Today’s primaries will be held in six states, but Democratic eyes are most fixated on the Wisconsin gubernatorial primary and Minnesota’s Senate primary. The Minnesota contest, pitting the state’s progressive Lieutenant Governor against the moderate congresswoman Angie Craig, will provide another test for internal polls, though we don’t have the makings of a herding problem. Only two pollsters have sampled the race in the past two months: Public Policy Polling, working on behalf of the Democratic Lieutenant Governors Association, which has a political action committee backing Flanagan, and SurveyUSA, which polls for several local TV stations. Public Policy Polling claims Flanagan has widened her lead from 7 to 14 over the course of July. SurveyUSA found a dead heat in mid-July but now sees Craig opening up a slight 2-point edge. At the very least, SurveyUSA has not been influenced by the Flanagan internals. 

The Wisconsin gubernatorial election has been one of the most challenging races to poll because who is actually running keeps changing. Milwaukee County Executive David Crowley suspended his campaign in early July and endorsed Lieutenant Governor Sara Rodriguez. One week later, Rodriguez suspended her campaign after firing her campaign manager over accusations of mismanaging finances. Crowley then re-entered the race with the endorsement of incumbent Governor Tony Evers. At the end of July, former Lieutenant Governor Mandela Barnes dropped out, concluding he had no path to victory.  

Meanwhile Assemblywoman Francesca Hong, a member of the Democratic Socialists of America, has been the steady plurality poll leader of a fractured field, despite a string of awkward media interviews focused on her past controversial social media posts. Notably, unlike all the other progressive candidates mentioned above, the Hong campaign has not published any internal polls and can’t be accused of trying to use polls to shape a media narrative.  

The one remaining candidate publishing internal polls is Crowley. Using Public Policy Polling, Crowley has published two sets of internals since re-entering the race. Most unusual for internal polls, they show the candidate who paid for them losing by nearly 20 points. 

So why is Crowley making them public? Because the campaign’s most recent poll re-asked respondents who they support after hearing unspecified information about Crowley and Hong. And then the Public Policy Polling crew further tweaked the generational demographics of the voter pool to reflect the 2022 electorate, increasing the share of older voters. After all that extra work, the race became a dead heat.  

Is the Crowley campaign using a convoluted internal poll to make a strained case they have a shot tonight? Or is it plausible that the poll reflects belatedly materializing anxiety about Hong yet to be detected by other polls, and—as with Michigan—bad turnout assumptions made by other pollsters?  

Keep in mind that we have very little independent polling in Wisconsin. Marquette Law School has long produced highly respected polling in the Badger State, and its last poll had Hong leading by 25 points. But that poll was sampled in late July before Barnes dropped out and hasn’t gone back into the field since. If the electorate soured on Hong in the final days, as Governor Evers has been sharply questioning her electability while campaigning with Crowley, Marquette’s pollsters wouldn’t know.  

The only other independent poll comes from the upstart text-to-web operation State Navigate, which produced a poll last week showing Hong up by 22 points. But cognizant of the polling miss in Michigan, the State Navigate team cautioned, “the error in Michigan seems to be due to multiple factors, including, but potentially not limited to: El-Sayed voters being more eager to answer polls, non-response biases among lower-income voters, underweighting voters above the age of 65, and inaccurately capturing rural voters. But to be clear, it is impossible to weight this survey (or any of the recent ones) out of the possibility that Hong’s voters are just more eager to answer surveys.” 

Might Crowley’s final internal poll end up being more accurate than independent polling? We’ll find out tonight. 

The post Internal Polls Flopped in Michigan and Hawaii. How Will They Do in Minnesota and Wisconsin?  appeared first on Washington Monthly.