Why I think Nate Silver’s model underrates Clinton’s odds

Matthew Yglesias


7th November 2016

People want to know who will win the election on Tuesday and are also nervous about it. Indeed, some people seem to want to be nervous about it. A trainer at my gym managed to set off a mass panic by passing on the completely inaccurate news that a poll of students at historically black colleges in Virginia found Trump leading.

The truth, however, is that by almost all accounts Hillary Clinton is leading in the polls and has been leading for essentially the entire campaign. That’s why all major election forecasters say that she is more likely than not to win. So the search for doubt has settled on Nate Silver’s forecast, an outlier among poll aggregators, that pegs Clinton as “only” a 65 percent favorite rather than the 85 percent or more she is favored by in other systems.

But even if you buy Silver’s main modeling assumptions (and I largely do) there’s considerable evidence outside the realm of things captured by poll aggregators that leads me to believe that if the polls are wrong, they are more likely to be underestimating Clinton’s support than overstating it. Predicting the future is difficult. And if you are terrified by the prospect of a Trump presidency, even a 10 percent chance he’ll win is pretty scary.

So feel free to stay nervous. But if someone offers you odds on an election bet, you should probably take the Clinton side of the wager. Clinton is ahead in the polls A good place to start is that all the buzz about methodological disputes and polls narrowing seems to have confused at least some people about where the polls stand.

Clinton is leading. Final national polls: : Clinton +6Ipsos: Clinton + : Clinton + : Clinton +4Herald: Clinton +4Bloomberg: Clinton +3 — Matt McDermott (@mattmfm) November 7, 2016 Indeed, though the economics of competing poll aggregation platforms essentially require rival aggregators to fight with each other during the election season, even the most model is saying that Trump will probably lose. Lots more polls coming today but for now, our forecast is stable, with Clinton up 3 points and about a 2:1 favorite.

https: . pic. twitter.

— Nate Silver (@NateSilver538) November 7, 2016 All of this is just to set the table for the rest of the argument. When you hear people citing “rules” about how undecideds will break or counting on turnout to deliver desired outcomes, you should be suspicious. Back in 2004, lots of Democrats used these methods to convince themselves the polls were wrong and Kerry was probably going to win.

Then in 2012, lots of Republicans did the same to convince themselves Romney was probably going to win. I’m going to cite some similar factors, but I’m not pointing to them to say that the polls are wrong. I’m just saying that if they are wrong, it’s unlikely that they are wrong in a way that favors Trump.

The uncertainty is The main reason Silver is giving Trump better odds than other modelers is that Silver sees a race with high variability and a high degree of uncertainty. That uncertainty manifests itself largely in the high number of people telling pollsters they will vote for Gary Johnson, for Jill Stein, or that they are still undecided. His logic, which seems very correct to me, is that a lead is a lot less safe than a lead, even though they are both leads.

But while there really is a lot of uncertainty about where the voters will land, it’s wrong to think of that as symmetrical uncertainty. As Emma Roller wrote last month, when you talk to voters, few “genuinely [think] that Mr. Trump and Mrs.

Clinton were equally bad candidates, though some seem to truly believe that they are two sides of the same coin. ” For Stein voters this is a pretty matter of ideology. But Johnson’s own vice presidential candidate, former Massachusetts Gov.

Bill Weld, has also made his preference for Clinton clear. Johnson’s voters also skew very young, which is a terrible demographic for Republicans in general and for Trump in particular. In other words, of the four possible things a voter could do on Election Day — stick to his guns, defect to Clinton, stay home, or defect to Trump — the fourth option is a lot less likely than the other three.

If voters are fuzzing up the polls, in other words, they are probably doing so by understating Clinton’s true level of support — not overstating it. Barack Obama’s strong approval ratings over the past two months are another reason to believe that uncertainty in the election is not symmetrical. In many past races, the typical undecided voter will be torn between two candidates because they like them both.

This race features two unusually unpopular candidates, so most undecided voters are torn between two options they dislike. The fact that most people approve of the job President Obama is doing, and that his net margin on this score is a healthy 6 percentage points, however, gives us some clues to their overall thinking. The ultimate behavior of a voter who likes Obama but is undecided because he doesn’t like Clinton is pretty unpredictable.

But common sense says that of all the possibilities (stay home, vote Johnson, vote Stein, vote Clinton, vote Trump) casting a ballot for the most virulently candidate in the race is the outcome. Early voting has good news for Democrats Nevada politics expert Jon Ralston’s take on the early voting numbers out of Nevada is they put the state out of reach for Trump. The only real question is how deep the “blue wave” goes in terms of House and State legislature races.

Early voting numbers out of Florida are less overwhelmingly but still lean her way. It’s not a mistake that models don’t include things that aren’t polls. But one of the nice things about not being a professional forecaster is you are free to consider nonpoll information.

The geography of Clinton’s pattern of support gives her an Electoral College disadvantage, which is hurting her badly in the 538 model. But if you spot her Nevada’s six electoral votes, that disadvantage goes away. The Florida situation is a lot less clear.

But it’s an absolute state for Trump and not for Clinton. Early vote evidence saying she may have the edge there is an indication that the tightening of polls in some Midwestern states may be irrelevant. Last, but by no means least, by essentially all accounts Clinton has a significant edge in the “ground game,” whereby professionals use data to deploy volunteers to make sure supporters turn into voters.

Obama had the edge over Romney on this score in 2012, and Clinton is operating with significant continuity with what Obama began. But Clinton also has a significant cash advantage over Trump, which Obama lacked, and Trump is starting from scratch in terms of building an organization. Based on what we know so far, Trump seems to have a fairly amateurish organization that’s more geared to direct marketing (think selling hats) than mobilizing voters.

That’s particularly bad for Trump. Relative to Romney’s coalition, he is weaker with voters and stronger with voters who are less regular . There are some advantages to that swap, in that it hurts Trump the worst in states that don’t matter much.

(He’s going to get creamed by an absurd margin in Maryland, but who cares.) But it also means that building a quality field organization is more important than ever, and there’s no sign he’s done it. Bottom line: Clinton is probably going to win I’m not going to engage in the false precision of estimating how much I think these factors shift the odds.

The point is simply this: Clinton is clearly up in the polls and is more likely than not to win. The case against certainty is that the large number of and undecided voters mean there’s inherently more instability in the race. That is true, but the instability is .

The large clutch of voters out there not committed to either major party candidate is committed to not voting for Donald Trump. At her best moments of the campaign, Clinton has been able to consolidate this broad coalition into a landslide majority. Current polling does not indicate that a Clinton landslide is likely.

But a polling error that results in an unexpected landslide is a lot more likely than a polling error that results in a Trump win. Last but by no means least, historical evidence suggests that field operations and ground games do matter in close elections. Clinton’s ground game is both better and more extensive than Trump’s, which serves to partially counterattack his more geographically efficient base of support.

Trump is probably going to lose. Clinton is probably going to win. Nothing is certain in life until it happens, but her odds are a healthy margin better than .

Watch: The bad map we see every election .


Leave your comments, questions and feedback on this article below. You can also correct any listing errors or omissions.