UI pollster: 'Once in a lifetime' upset

UI pollster: 'Once in a lifetime' upset

At 4:20 p.m. today, Sheldon Jacobson will join Scott Beatty on WDWS 1400-AM

Sheldon Jacobson’s phone pinged constantly Wednesday morning, with reporters and anxious voters all asking the same question:

How did the polls get it so wrong?

The University of Illinois computer science professor, who runs an Election Analytics website with his students, didn’t blame the state polling data that they use for their analysis, which gave Hillary Clinton a 99 percent chance of winning.

Rather, he said, it was a “once in a lifetime” upset that defied the odds, based on a near-total swing in undecided voters to Donald Trump. That scenario was one of 21 laid out on the website, and it gave Trump a 77 percent chance of winning. But most election analysts didn’t think it was likely.

“We’ve thought about this a lot, believe me,” said Jacobson, who planned to post a summary of the election results on his website Wednesday.

The website presented 21 possible scenarios, based on polling data and two major factors: whether the presidential campaign was a two-way, three-way or four-way race; and which way undecided voters would swing (from heavily Republican to heavily Democratic).

What’s reported up front is the neutral one — a four-way race with undecideds split equally between Clinton and Trump — but the site explores other scenarios by changing those factors. Twenty favored Clinton; only one, with undecideds swinging strongly GOP, gave Trump a more than 50 percent chance to win.

The neutral analysis gave Trump less than a 1 percent chance, he said.

But as the election progressed, support for third-party candidates waned and turned it into a two-person race, Jacobson said. And, in day-after hindsight, it’s clear that undecided voters went overwhelmingly for Trump.

“That’s the only strong chance he had of winning,” he said. “Is this a surprise? Yes. Is it a shock? Not really.”

Analytics sites like his report the neutral scenario because they don’t want to make judgments about what undecided voters might do. Polls don’t provide that data, he said.

“You have so many scenarios,” he said. “It turned out the most correct one was the most extreme one.
“We didn’t see it coming,” he said.

Should polls do more to tap into the leanings of undecided voters? Jacobson said that would be “going beyond their skill set.” Most polls sample 500 to 1,000 people and try to extrapolate that to a larger populace.

“People are going to rail on the pollsters. I don’t think they’ve done anything wrong, or worse than in the past,” he said. “They’re always going to be a little off because it’s an inexact science.”

But you’d have to go back to Dewey vs. Truman in 1948 to find a comparable upset, he said. Polling was very different then, with only national voter surveys rather than state-by-state.

“This may be once in a lifetime,” he said.

Other websites, such as Nate Silver’s fivethirtyeight.com, had many scenarios but favored Clinton’s chances, too.

“There were just a handful of people who thought Trump had a chance, but they were outliers,” he said. “It’s not that anyone is wrong. We all had scenarios. We just didn’t realize that the most extreme scenario was going to occur.

“All we’re trying to do is give people a picture of the race, and they can make assessments.”

Trump’s winning march through nearly all of the battleground states was like “flipping a coin 10 times in a row and you get 10 heads,” he said, noting the narrow margin of victory in Michigan (less than 20,000 votes) and other swing states.

“It’s rare, but it happens. Each state was a tossup and it all went into his favor,” he said.

Jacobson’s inbox and voicemail were full of messages from unhappy voters.

“I think people want an easy answer to a complex problem, and there isn’t an easy answer,” he said. “When you put yourself out there and you have to make an assessment about the future, you’re going to be wrong sometimes.”

He likened it to a stock index fund, which has ups and downs but over the long term does “pretty well.” One stock doesn’t reflect its overall performance, he said.

“Sometimes we’re going to be really good, and sometimes we’re not going to be so good. In this case most of us were pretty bad,” he said. “We used the best tools and the best information available to us. This was an extreme outcome.”

Sections (2):News, Local
Tags (1):2016 election

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.
Joe American wrote on November 09, 2016 at 11:11 am

Famous last words of a dreadful loser.


It wasn't a "once in a lifetime" upset to many diehard Trump supporters, all of whom have known for months that the polls were skewed.  Many of them have been diligently watching the polls, and unlike Sheldon and the other now-obsolete pollsters, took into account what everyone knew as the "silent vote".


The little polling game is now irrelevant.  Time to move along.

787 wrote on November 09, 2016 at 12:11 pm

Giving Trump a 99% probability of losing was much more fun than giving a 77% probability of him winnning.

And if it discouraged some people from even going out to vote...  even better yet.

Hopefully, the News-Gazette will ignore this guy in the future.  He was about as wrong as he could get.  

A perfect example of why people need to ignore folks like Jacobson, and just go out and vote.

wayward wrote on November 09, 2016 at 12:11 pm
Mastadon-27 wrote on November 09, 2016 at 2:11 pm

One of the oldest truths in software programming GIGO (Garbage In - Garbage Out),  ROFL.

Cuthbert J. Twillie wrote on November 09, 2016 at 5:11 pm

Because in the rarified air of academia they really do not care what the common folk think.  They believe that they are way smarter than we are...................

MK wrote on November 09, 2016 at 7:11 pm

Prof. Jacobson -- Your comments about the swing states being a bunch of coin tosses seems to indicate you believe them to be independent events.  Is that so?  It seems to me that if one candidate outperforms in Michigan, that candidate is at least somewhat likely to outperform in Wisconsin.  These states have structural similarities and it strikes me as likely that those similarities would manifest in the vote.

On another note, it sounds like you had variants of the model about how the undecided (or third party/independent?) vote would break.  But why not bake that into the model?  You could assign probabilities to the various ways the undecided vote might break (hard Clinton, slight Clinton, even, slight Trump, hard Trump) and create a net likelihood of a candidate winning.  It seems to me your >99.5% probability of Clinton winning would have decreased with this and the previous paragraph.

I read all your analysis.  I think there is much to be gained from it.  But I'm a little concerned about comments from some of the aggregators who say the fault is not in their model, but in their inputs.  I could have read closer, but I only learned today that you only relied on one how-will-the-undecideds-break model for your prediction.  I don't get why you chose not to go deeper than that.  I think it's unfortunate, but I fear that people will drift away from objective (as much as we can...) analysis and towards demagoguery because the knob turners on the algorithms have not been sufficiently clear.

JamBam wrote on November 10, 2016 at 10:11 am

Pollsters have been oversampling Democrats for years in order to get targeted polling results meant to depress the support of the GOP candidate. It's been happening for years. It was alluded to in Hillary's emails.  Just 2 weeks before the election, ABC News had Hillary Clinton up 50-38.  

It was all junk.  Trump had all of the energy this election and he won a larger electoral college victory than any President since Ronald Reagan.  

It wasn't an upset for those actually waching each campaign.