From 538: ‘Shy’ Voters Probably Aren’t Why The Polls Missed Trump

I'll compile a list of articles highlighting what went wrong - and right - in the polls leading up to the 2016 presidential election. One prevailing theory was that some Trump voters did not want to tell a human on the telephone - a pollster - that they intended to vote for him.

The following analysis finds little support for that argument. Below is his first point - there was no evidence that his vote totals were higher in places where he was unpopular.

- Click here for the article.

. . . the “shy Trump” theory relies on the notion of social desirability bias — the idea that people are reluctant to reveal unpopular opinions. So if the theory is right, we would have expected to see Trump outperform his polls the most in places where he is least popular — and where the stigma against admitting support for Trump would presumably be greatest. (That stigma wouldn’t carry over to the voting booth itself, however, so it would suppress Trump’s polling numbers but not his actual results.) But actual election results indicate that the opposite happened: Trump outperformed his polls by the greatest margin in red states, where he was quite popular. The two states that had the largest polling error for Trump were Tennessee and South Dakota, where Trump won more than 60 percent of the vote. Meanwhile, Trump underperformed his polls in states where the stigma against him would seem to be strongest: deep-blue states like California, Hawaii, Massachusetts, New York and Washington.2We don’t have polling data for areas smaller than states, so it is possible that Trump outperformed his polls in the blue pockets of red states or underperformed them in red pockets of blue states. But there is no evidence to suggest that this happened. Overall, as my colleague Carl Bialik and I (as well as Andrew Gelman) have pointed out, there’s a very strong correlation between how Republican a state is and how much better Trump did than polling averages indicated he would.

The author offers the following graphic to support the point above. It also shows that states are increasingly polarized according to political party. What this means for the governability of the nation is worth discussing.

From Vox: Few predicted Trump had a good shot of winning. But political science models did.

Not everyone got the election wrong. Forecasts based on fundamentals were more accurate than those based on polls - which were actually accurate, as we'll see in other posts. 

- Click here for the article.

Why did so few people see Donald Trump’s win coming?

The polls got it wrong. The major election forecasting models got it wrong (though FiveThirtyEight’s Nate Silver deserved credit for being significantly less certain about it). The political professionals got it wrong. The pundits got it oh, so very wrong indeed.

Oddly enough, though, there were signs pointing to the fact that Trump had a better chance than people were giving him, and they were lurking in plain sight.

They were in well-known political science research on “fundamental-based” factors that has long been used to explain presidential elections.

In fact, of the major political science models that try to explain presidential elections, three predicted Trump would win and three others predicted only a very narrow Clinton victory.

From 538: How I Acted Like A Pundit And Screwed Up On Donald Trump - Trump’s nomination shows the need for a more rigorous approach.

In 2305 today we looked at Nate Silver's bad call during the primary. Along with most other political observers, he discounted Trump's viability as a candidate for the Republican nomination. Here he tries to figure out what he got wrong and why. He calls himself a "data journalist" which suggests that his approach is rigorous and needs to be adjusted if proven faulty. Here's what he came up with.

- Click here for the article.

. . . I’m going to proceed in five sections:

1. Our early forecasts of Trump’s nomination chances weren’t based on a statistical model, which may have been most of the problem.

2. Trump’s nomination is just one event, and that makes it hard to judge the accuracy of a probabilistic forecast.

3. The historical evidence clearly suggested that Trump was an underdog, but the sample size probably wasn’t large enough to assign him quite so low a probability of winning.

4. Trump’s nomination is potentially a point in favor of “polls-only” as opposed to “fundamentals” models.

5. There’s a danger in hindsight bias, and in over correcting after an unexpected event such as Trump’s nomination.

From the Washington Post: Election forecasters try to bring some order to a chaotic political year

For a handful of political scientists, presidential elections can be forecast with a properly designed algorithm. The debate is over what algorithm works best.

- Click here for the article.

One model, by Robert Erikson of Columbia University and Chris Wlezien of the University of Texas, points to Clinton winning with 52 percent of the two-party popular vote. (Actual vote percentages for Clinton and Trump will be lower because of the presence of Libertarian Gary Johnson and the Green Party’s Jill Stein on the ballot.) That model combines post-convention polls with the results from the index of leading economic indicators.

Michael Lewis-Beck of the University of Iowa and Charles Tien of Hunter College also see a Clinton victory, with just 51 percent of the two-party popular vote. Tien said that translates to a narrow electoral college majority for Clinton of 274 votes.

Andreas Graefe of LMU Munich and J. Scott Armstrong of the Wharton School at the University of Pennsylvania cite four different models, all of which point to a victory by Clinton larger than some of the other forecasts.

One outlier is Helmut Norpoth of Stony Brook University. His model takes into account sentiment for a change in parties, but most important, and unusual, is his reliance on performance by the major-party candidates during the early presidential primaries, in this case New Hampshire and South Carolina. On that basis, he predicted last spring that Trump would win the election and said the prediction came with an 87 percent certainty.

When I spoke with Norpoth a few days ago, he was admittedly nervous. “I do worry. . . . I’m clearly sort of the odd man out,” he said. But, he added, “it’s not a foregone conclusion that he’s [Trump] going down the tubes.”

Alan Abramowitz of Emory University uses what he calls a Time for Change Forecasting Model. His model does not rely on polling data but instead takes into account the incumbent president’s approval rating at midyear, the growth rate of real gross domestic product in the second quarter of the election year and whether the incumbent president’s party has held the White House for one term or more than a term.

On that basis, his model predicts a narrow victory for Trump. But Abramowitz also suggests that Trump could underperform. “A model like mine that relies entirely on fundamentals is likely to miss the result because Trump is such an atypical candidate,” he said.

Speaking of bad forecasts

Boing Boing points this one out.

50% Chance that Ohio Decides the Election

So says polling guru Nate Silver:

We are now running about 40,000 Electoral College simulations each day. In the simulations that we ran on Monday, the candidate who won Ohio won the election roughly 38,000 times, or in about 95 percent of the cases. (Mr. Romney won in about 1,400 simulations despite losing Ohio, while Mr. Obama did so roughly 550 times.)

Whether you call Ohio a “must-win” is a matter of semantics, but its essential role in the Electoral College should not be hard to grasp.

Economic trends, campaign advertising, the ground game

In response to a question about what matters between now and election day for Obama and Romney, smart guy John Sides lists the three factors above. Equally smart guy Jonathan Bernstein chimes in.

Commentators are suggesting that a very large percentage of the electorate has made a decision about who they will support and their positions will not change. Any shifts will happen among a very small number of people.

Predictive Models for Presidential Re-Elections

Good news for Obama, though there's never a guarantee. From Alan Abramowitz:

The 2012 presidential election is still more than 20 months away. While the early maneuvering for the Republican presidential nomination is already underway, the identity of President Obama’s GOP challenger won’t be known for more than a year. Economic trends will have a major impact on the President’s reelection chances and unpredictable events, such as the recent political turmoil in Egypt, could also affect the public’s evaluation of the President’s performance.

But even without knowing what condition the economy will be in, whether a major international crisis will erupt, or who will win the Republican nomination, one crucial determinant of the outcome of the 2012 presidential election is already known. Barack Obama will be seeking reelection as a first term incumbent and first term incumbents rarely lose.

62 + .62*256 -1.4*7.4 + 0.1*9.7 = 211.33 Democratic seats

That's the formula one polisci guy uses to predict the number of seats Democrats are likely to have in the House after the election is over. The argument is that much of the results of midterm elections are structural, that is, based on circumstances apart from any attitude one has about the president. The model predicts that given the nature of the last election and the current economic condition, Democrats should expect to lose 45 seats. Any less means they did well, any more means they did poorly.

Math is fun.

Two Forecasts

Charlie Cook.

Nate Silver.

Forecasting the 2010 Elections

On the economy and the vote.

Forecasts for 2010

From smart guy Nate Silver:

- Republicans have a 67% chance of taking the House.
- Republicans have a 25% chance of taking the Senate.
- Republicans are likely to win 30 Governors races.