'Polling is as much an art as a science' — Professor explains why polls got it wrong again

In spite of changes after 2016, Trump overperformed
Posted at 4:36 PM, Nov 13, 2020
and last updated 2020-11-13 19:08:32-05

CLEVELAND — Aaron Montgomery is honest about who he is: "I'm a total numbers guy. I like getting my hands dirty with data. I'll take any excuse I can get to do it."

Montgomery, an associate professor at Baldwin Wallace, got his excuse this year participating in the Baldwin Wallace University - Great Lakes Poll.

"I mean, when you're looking at a big spreadsheet with, you know, thousands of responses and you've got to try to figure out how to slice them and you've got to write the code to do whatever, I really get a lot of unironic joy from that," he said about why he likes participating in the poll.

As one of the math minds behind the poll, "I prefer proofs, but sometimes in the real world we have to settle for guesses."

The poll, like several others looking at the potential moves of voters during the election, got the numbers wrong.

He got a first-hand look at what happened with the polling margins this year.

"One of the secrets about the polling industry is that polling is as much an art as a science," Montgomery said. "And there is some mathematics involved, but there's also some best guesswork."

That combination can be tricky to get right.

"I think we were more on track than we were four years ago," said Tom Sutton, News 5's political analyst and the director of the Community Research Institute at the university. "And the fact that the results that we had in the four states for Biden in that 50-49% range were very similar to what he got in the actual vote count."

This year, the final Great Lakes poll had president-elect Joe Biden trailing President Donald Trump by two points. Instead, Trump carried Ohio by eight points - 53% to Biden's 45%. The poll also surveyed voters in Wisconsin, Pennsylvania and Michigan.

"So we had very, very accurately guessed the vote share for Biden," Montgomery said about the overall performance of the Great Lakes Poll. "Where we missed was for President Trump."

Trump outperformed the projected numbers.

"The undecided numbers were very high," Sutton said. "They were in the 15 percent range even a week before the election. And a lot of those voters then apparently voted for Trump."

After the 2016 election, the industry made changes to the poll algorithms.

"For instance, the weighting by education was something a lot of pollsters were not doing previously," Montgomery said. "That is something that we and most other pollsters are doing at this point."

Both Montgomery and Sutton said there were several factors that lead to issues in polling across the country.

There are a group of people opting out of participating in polls. The plethora of spam calls to cell phones led to an increase in calls from pollsters going unanswered. The pandemic shifted when people were at home to take calls and the historically high voter turnout rate all led to final numbers falling outside the poll's margin of error.

"But we'll certainly be looking at how we've been doing our sampling and making changes to that to try to more accurately reflect what we see in Ohio," Sutton said.

Both men cautioned it is too early for a full diagnosis of what happened with polls this year. Votes are still being counted and data from exit polls are still coming in. Experts are skeptical of the exit poll data this year because the pandemic changed the way traditional exit polling is done.

But, the issues with polling in the last few years have not diminished the importance of their results.

"For a candidate in the campaign, that information is critical to see what kinds of strategies they need to use," Sutton said. "If they're ahead, if they're behind, whether they do a debate or not, whether they do negative ads or not. That's all based on the polling. So we're going to continue to have polling because it's the best way to capture public opinion."

Across the country, the polling numbers were off. In Florida, some polls had Biden winning the state by five points. Although that state is still certifying results as of the publication of this article, Trump is ahead in the vote count in Florida.

"But we saw a similar pattern where the Trump votes clearly were understated in terms of the poll compared to what actually happened with a turnout," Sutton said about poll results across the country.

The Great Lakes Poll wasn't the only poll that had issues in Ohio. The final Quinnipiac University poll published on Nov. 2 had Biden up 11 percentage points over Trump nationally and four percentage points in Ohio.

"Clearly there's something that we didn't catch in the surveys that we did," Sutton said.

Sutton acknowledges there are problems with current polling but "the alternative will figure out who you are and what you think based on all of your shopping habits, based on your predilections, based on what you buy on Amazon. Do you really want that to be the alternative? Because, folks, that's what already happens."

For Montgomery, the polling numbers will always bear out.

"What else is there if not polling?" he asked. "You can count rally sizes, I suppose, and you can count yard signs," he said. "But, I don't think any of that has any kind of lasting predictive power."