Before Tuesday many pundits wrote Gov. Sam Brownback’s political obituary prematurely, looking at polling data that put Brownback and U.S. Sen. Pat Roberts on the verge of defeat.
The polls were wrong. Dead wrong.
Roberts trounced independent Greg Orman by 10 percentage points in the general election, and Brownback was able to put up a respectable four-point victory against Democrat Paul Davis, who had led in almost every poll since July.
The political forecasting site FiveThirtyEight had predicted both Davis and Orman would win based on polling data. The New York Times’ Upshot also incorrectly predicted that Orman held a slight edge in what was supposed to be a close race for U.S. Senate.
Chapman Rackaway, a professor of political science at Fort Hays State University, said that as elections near, a phenomenon called “herding” happens with polling firms.
Rackaway pointed to Fox News. Fox’s polls showed Davis and Brownback tied in early October, but then later in the month showed Davis with a comfortable six-point lead.
“As it gets close to the election it’s almost like groupthink kicks in and a herd mentality. They want to be close to each other, so if they’re wrong, they’re no more wrong than everybody else,” Rackaway said. “Turns out Fox should have stayed the course because their earlier polls were probably more accurate. So that’s why you saw this clustering and that’s why you saw Davis’ numbers particularly jump up a little bit.”
Rackaway should have stayed the course, too. He had predicted a Brownback victory all year, but on the Sunday before the election he told the Kansas City Star he thought Davis would prevail, basing his new opinion on polling data. He has joked that he should still get credit for being right 95 percent of the time.
Insight Kansas, a political science site that Rackaway contributes to, gave Davis a three-point edge going into Election Day based on aggregated polling data. Its data showed Orman with a lead of less than a point on Roberts.
Rackaway blamed Orman’s underperformance on the lack of a get-out-the-vote machine.
“He had a crew. He had a staff. But he didn’t have the infrastructure of a full state political party organization like Democrats did and Republicans did,” he said. “Orman didn’t have enough resources, people, what have you, to say here are the people we need to get out and vote for us on Election Day.”
The polls also failed to accurately predict the behavior of specific groups of voters. A poll from Survey USA and KSN-TV that came out the week before the election showed voters over the age of 65 favoring Davis over Brownback by a 10-point margin.
But according to exit poll data, collected by the National Election Pool, this group of voters came out in favor of Brownback more so than any other group by a margin of 56 percent to 42 percent.
Brownback’s campaign manager, Mark Dugan, had repeatedly disputed the validity of polls, which showed the governor trailing, and on Wednesday he felt vindicated.
Dugan said polls were consistently underestimating the number of Republican voters that would show up to cast ballots.
“We had a very sophisticated and complex voter modeling program that we invested a lot of time and energy in. And we spent a good portion of our resources focusing on getting our voters to the polls,” Dugan said. “So we very intentionally knew this would be about who showed up to the polls.”
Dugan accused newspapers and “the state’s partisan academics” of exaggerating Davis’ support.
The Secretary of State’s Office offered an unofficial estimate that 864,109 people voted, which would be about 49.5 percent of the state’s registered voters, but could not provide a breakdown by political party on Wednesday.
Voters are becoming harder for pollsters to reach because of mobile phones and caller ID, said John Hancock, a Missouri GOP consultant.
“I think polling has become a very challenging enterprise,” he said.
If those who participate in the polls don’t match the perceived demographics of likely voters in a region, some polling companies will weight – or adjust – the results to account for the underrepresented groups.
“It’s tricky, and my suspicion is all of these pollsters are unable to produce results they don’t have to tinker with to try to weight them,” he said. “When you start doing that, it’s much more art than it is science, and if you make one wrong assumption, you’re increasing you error rates. If you make three wrong assumptions you’re not right at all.”
Patrick Miller, an associate professor of political science at the University of Kansas who studies polling, said that polls always underestimate one party. He said in 2012 polls underestimated Democratic voters in the presidential election. This year they underestimated the Republican vote in Kansas and nationwide.
“When you do have a wave, the polls, public polls at least, tend to underestimate the magnitude of that wave,” Miller said. “Just like today we’re looking at the polls and saying, ‘Well, why weren’t they Republican enough?’ Two years ago we were looking at them and saying, ‘Why weren’t they Democratic enough?’ ”
He said that according to responses to exit poll questions, the pre-election polls accurately gauged voters’ feelings about issues like Obamacare and same-sex marriage, but what they failed to do was accurately predict how Republicans would vote. A higher percentage of Republicans stayed with the incumbents in Kansas than expected, he said.
Miller noted that based on polling data, Brownback is the governor with the lowest approval rating to be re-elected in any state since 1990.
Dugan rejected this claim, saying it was based on polls that were already proven to be faulty on Election Day.
Contributing: Lindsay Wise of the McClatchy Washington Bureau