In a historic upset, Vermont Sen. Bernie Sanders snatched the Michigan Democratic primary from Hillary Clinton on Tuesday, despite weeks of polling that put Clinton’s lead anywhere from 10 to 30 percentage points or more.
The self-declared democratic socialist defeated the former secretary of state, 50 percent to 48 percent.
“Polling was just disaster on the Democratic side. Perhaps the worst that I’ve seen,” said Susan Demas, publisher of the Inside Michigan Politics newsletter.
Analysts say survey samples might have been weighted too heavily with older voters, and Tuesday’s unexpectedly large youth participation tipped the vote in Sanders’ favor. Other factors were the warm weather in Michigan, and perhaps people making up their minds late – persuaded by last-minute ads or the Sunday debate in Flint.
Whatever the reason, Sanders’ surprise victory is a reminder of the limits of predicting who’s most likely to vote in primaries.
“We were way off,” said pollster Bernie Porn, president of EPIC-MRA in Lansing. “I think what we and every other pollster did was underestimate the number of younger voters that were participating.”
Another factor that may have caused errors in polling was that statewide voter turnout was 34.5 percent, while participation in Detroit – Michigan’s largest city – was significantly less at 25.5 percent, Porn said.
His firm conducted a telephone-operator survey of 400 likely Democratic voters from Feb. 28 to March 1, showing Clinton held a commanding 25-point lead over Sanders.
RealClearPolitics, which averaged recent Michigan polls, predicted Clinton winning with a 21.4 percentage point spread. The statistical analysis website FiveThirtyEight forecast that Clinton had a 99 percent chance of beating Sanders in Michigan based on state polls, national polls and endorsements.
Both the Clinton and Sanders campaigns said they saw a tighter race. Former Michigan Gov. Jennifer Granholm, who heads a pro-Clinton super political action committee, said the contest would be closer than people expected.
“I think it’s competitive because they both have compelling messages,” Granholm told The Detroit News on Saturday.
Mitchell Research & Communications’ last poll had Clinton winning Michigan with 61 percent of the vote.
“We all blew it,” said Steve Mitchell, chairman of Mitchell Research & Communications. “Clearly, what we’re doing here in Michigan did not work. It’s always tough to figure out what happened when you have an event like this when no one is right.”
Mitchell projected turnout of 8.7 percent among voters ages 18 to 29, but exit polls of state voters showed nearly double of the electorate in that age range — 19 percent. Sanders won 81 percent of the age group.
Matt Grossman, director of the Institute for Public Policy and Social Research at Michigan State University, also attributed much of the polling gap to the turnout among voters ages 18 to 29, which Sanders wooed with university rallies around the state.
“It happened to be an election where age mattered a lot, and it was very hard to estimate the electorate,” Grossman said.
The sunny, warm weather – 70 degrees or higher in some places — encouraged more participation from voters who otherwise would have stayed home, Mitchell said.
Surveys such as Mitchell’s that used automated robo-calls to connect with likely voters excluded cellphones from their calling lists, as required by federal regulations.
As a result, those pollsters had to assume that young people answering land telephone lines were representative of those who only had a cellphone, and it wasn’t the case, Grossman said.
Pollsters who did include cellphones in their surveys may have used too strict of a screening process to focus on “likely voters,” he added.
Most national pollsters are moving to majority cellphone samples, but they can be more time-consuming and expensive than targeting land lines, he said. Obtaining accurate cellphone lists is difficult because subscribers change numbers more frequently.
MSU researchers interviewed several hundred Michigan residents between January 31 and March 3 as part of the ongoing State of the State Survey, which included a 50 percent cellphone sample and less stringent screening.
The results suggested a consistently close race between Clinton and Sanders since his momentum-building primary win Feb. 9 in New Hampshire.
“There’s a lot of talk about a last-minute shift on Monday, and I just think from our survey, there’s not a ton of evidence to show that,” Grossman said.
A Monmouth University poll conducted March 3 to 6, had Clinton finishing 13 percentage points ahead of Sanders, after surveying 302 likely Democratic voters in Michigan using telephone operators.
Patrick Murray, director of the Monmouth University Polling Institute, defended his sampling approach, saying it has worked elsewhere in both open and closed primaries. He couldn’t provide a generalized explanation of what happened in Michigan.
White voters in the poll were more evenly divided between Sanders and Clinton than exit polls reflected; however, Monmouth’s sample, which included young people with cellphones, was “pretty much on target,” Murray said.
“I have been doing a lot of polling in this primary, and this is the first time I have missed the Democratic winner. I was even the only one who correctly forecast Sanders’ win in Oklahoma,” Murray said by email.
“We have seen some instances where voters simply swerved at the last moment ... and this may be one of them.”
Anecdotal reports suggested that some Democrats believed Clinton had sewn up a victory, so they crossed over to spoil the Republican primary by voting for or against Trump, said Kyle Kondik, managing editor of the political newsletter Sabato’s Crystal Ball at the University of Virginia Center for Politics.
“But I don’t think that’s enough to turn a double-digit lead for Clinton into a narrow loss,” Kondik said.