Uh Oh: LA Times (Which Has Trump Leading) Projects 352 Electoral Votes For Clinton

Posted: Nov 07, 2016 5:15 PM

Since September, the LA Times poll has had Donald Trump ahead by anywhere from one to five points. It did have Hillary Clinton leading by four points in late July-early August. It also had both Clinton and Trump tied. Still, the last poll they conducted had Trump ahead by five points, so what is their Electoral College projection? A massive landslide for Hillary Clinton, with 352 electoral votes:

Our projection would give Clinton 352 electoral votes, while Trump would end up with 186. That would put Clinton's electoral majority midway between President Obama's 2008 win and his 2012 reelection.

Of these last picks, Ohio and Arizona were the hardest. Polls have been close in both states.

Ohio does currently seem like a jump ball, but we lean toward Clinton winning there because of the strength of her get-out-the-vote operation.

In Arizona, we're expecting that the surge in Latino votes that has been visible next door in Nevada will put Clinton over the top. Polls that show Trump winning may be underestimating the size of the Latino turnout.

Holy shrimp! So, what’s going on here? Well, Dan Schnur, who conducted the poll, explained the discrepancy; they might have oversampled Republicans, specifically Romney supporters from 2012. That’s how they formulated their samples:

Whereas most polls simply ask voters to choose between alternatives, the Daybreak poll attempts to determine the intensity of voter preferences by asking how committed a respondent is to his or her candidate (on a scale of 1 to 100).

Few voters shift their support on an absolute basis – from total and complete certainty for one candidate to equally unequivocal certainty for the other. Most change their minds much more incrementally, suggesting that it’s reasonable to augment traditional all-or-nothing surveys by gauging commitment over time. The polling inaccuracies we’ve seen in so many recent elections — both in this country and internationally — substantiate the need for alternative methodologies.

In measuring voter intensity, the Daybreak poll’s results do not contradict the consensus that Hillary Clinton has consistently attracted more supporters than Donald Trump. It simply shows that Trump’s backers are more fervent — and therefore more likely to actually vote.

So far so good. But to achieve a representative sample of Republicans and Democrats, the Daybreak poll creators used a potentially unreliable proxy, asking possible respondents whether they had voted for Barack Obama or Mitt Romney in 2012. Since voters often don’t like to admit that they picked the losing candidate, it’s likely that Romney backers were oversampled. This almost certainly exaggerated Trump’s level of support.

Over at Hot Air, Allahpundit pretty much said that when it comes to polling this cycle, we’re literally in the unknown:

There’s a raft of final national polls out this morning from other outfits (although we’re still waiting on the final data from the big one, the ABC/WaPo tracker) that more or less support the LA Times’s electoral map — but with a major caveat. Fox News’s final survey has Clinton up four, 48/44; CBS’s final poll also has her up four, 45/41; Bloomberg’s last one has her up three, 44/41; and the last poll from NBC/SurveyMonkey has her up six, 47/41. ABC/WaPo did have a new tracking poll out this morning, but with another to follow this afternoon: Clinton led this a.m. by four points as well, 47/43. Those results are strikingly consistent, which might lead you to believe that the race has settled and the actual state of play is now showing up across various different surveys. Could be, but read this 2014 Nate Silver post about a bad habit pollsters tend to fall into at the very end of an election known as “herding.” By pure chance, thanks to margins of error, there should be some natural variation in the results among a group of new polls. If X is leading Y by three points in reality and the MOE is three points, you’d expect one poll to have the race tied, another to have X ahead by six, and the others to fall at different points in that range. If instead you’ve got (nearly) everyone showing a three- or four-point lead, that suggests that the pollsters might be afraid of publishing an outlier so close to election day, knowing they’ll be judged harshly for having “missed” the result, and are tweaking their data by playing with turnout models in order to get a number closer to what everyone else has. If that’s what’s happening here, then we’re not really getting independent data. And that calls into question how accurate any individual result really is.


YouGov published an insightful post about that same topic, “nonresponse bias,” a few days ago. There’s a school of thought out there that this has been a four- or five-point race for months and that many of the swings we’ve seen in the polls are due less to voters switching their votes than simply refusing to respond to pollsters during periods when the news isn’t good for their candidate. In other words, Trump’s actual support didn’t dip much after the “Access Hollywood” tape but the number of Republicans willing to take a call from a pollster during that period might have, creating the impression that he had lost votes in surveys. If Clinton wins by four tomorrow night, there’ll be much more written about that. Was this actually a stable race all along?

Well, we'll all know by tomorrow night who is correct.