Polling averages and forecast models are supposed to bring order to the chaos, put outlier polls in proper perspective and provide a sober, unbiased picture of the state of the presidential race.
So why are they all over the place in the final days, with some models asserting a Hillary Clinton victory is a near-certainty, and others giving Donald Trump a real chance at winning?
The dissonance has resulted in fundamental disagreements over how close the race really is in a presidential election that both parties have described, in no uncertain terms, as the most important election in our lifetimes.
The latest RealClearPolitics projections have the race teetering on a knife’s edge. But other, more complex forecast models — based on the same polls — give Clinton a 98 percent or 99 percent chance of defeating Trump next Tuesday.
As of noon Friday, the models and polling averages ranged from RealClearPolitics’ close-to-tied electoral map — if Florida were to flip back to Trump, he would be ahead in the Electoral College — to HuffPost Pollster and the Princeton Election Consortium, which each show Clinton’s chances close to 100 percent, and Daily Kos Elections’ model gives Clinton a 91 percent chance of winning.
In the middle are models from two high-profile data-journalism sites, FiveThirtyEight and The New York Times’ Upshot. But those two models aren’t in complete agreement, either. FiveThirtyEight’s “polls-only” model pegs Clinton’s chances at 67.8 percent as of late Friday morning, while the Upshot model is more bullish at 84 percent.
One explanation for the glaring disparity overall: Some models are quicker to adjust to the newest polls — which, over the past two weeks, have shown Trump surging — while others are more cautious about changing course. With the election only four days away, it’s sparked a dispute over which approach is more prudent.
The debate isn’t academic — especially for Democrats panicking in recent weeks that Trump’s chances of winning are improving as the election draws closer, and Republicans eager to cling to positive momentum in a race that appeared all but over weeks ago.
“If you have a model based on public polling, it ought to do a good job of reflecting the public polling as the data comes in, or it might be doing something wrong,” FiveThirtyEight editor in chief Nate Silver told POLITICO in an email. “Our forecast — with Clinton as about a two-to-one favorite — seems both empirically and intuitively consistent with public polling that mostly puts her ahead, but with a lot of volatility and a lot of disagreement and a ton of undecided voters (and a rather leaky Electoral College firewall for Clinton, such as in New Hampshire). Betting markets and financial markets — people putting money on the line — have also tracked 538’s forecasts much more closely than the other models.”
The HuffPost Pollster model, on the other hand, is far more confident in Clinton’s chances, giving her an 98.1 percent chance of winning as of Friday morning. It’s based on HuffPost Pollster’s polling charts — which, unlike RealClearPolitics, is not a simple average of the polls but is designed to make sense of seemingly contradictory results by discounting, though not dismissing, outlier polls.
“The advantage of that is that the model looks for the most likely polling estimate and doesn’t change drastically when there’s an outlier poll — whereas a straight average will move toward the outlier,” said Natalie Jackson, the senior polling editor at The Huffington Post. “The disadvantage is that it’s slower to pick up on changes in polls; it requires a few polls showing a change in the trend before it will reflect a new trend. That’s what we’re seeing now — some polls are less positive for Clinton, but others are basically in line with what the trends have been saying.
“Since the evidence is conflicting, the model sticks with past trends,” Jackson continued in an email. “It’s a trade-off in letting outliers and individual polls affect the aggregate, versus requiring more evidence to change a trend. Right now, we’re on the latter side. Of course, we’ll assess how that worked after the election.”
That approach — in addition to other decisions about which polls are included — has also led to a massive divergence between the nation’s top two polling aggregation sites: RealClearPolitics’ and HuffPost Pollster’s polling charts.
As of late Friday morning, Clinton led by 1.7 points in a head-to-head matchup with Trump nationally, according to the RealClearPolitics average. But Clinton’s lead in the HuffPost Pollster model was a far more robust 6.1 points.
There’s always been some differences in the forecast models, but the models have pulled apart rather than converge as the election has drawn closer — in large part because of Trump’s late surge.
On the day of the third Clinton-Trump debate on October 19, the FiveThirtyEight model had Clinton an 87.3-percent chance of winning, more comparable to the Upshot’s 92 percent, and Daily Kos Elections’ 95 percent.
But since Trump’s polling recovery over the past two weeks, Clinton has slid more in some models than others.
The FiveThirtyEight model showed Clinton’s win-probability slipping nearly 22 points from its October high. Silver said that better reflects the volatility in the race — both in the most recent swing in the polls toward Trump, along with the significant number of voters still undecided or lining up behind minor-party candidates.
“[E]verything depends on one’s assumptions, but I think that our assumptions — a Clinton lead, sure, but high uncertainty — has repeatedly been validated by the evidence we’ve seen over the course of the past several months,” Silver said. “The idea that she’s a prohibitive, 95 percent-plus favorite is hard to square with polling that has frequently shown 5- or 6-point swings within the span of a couple weeks, given that she only leads by 3 points or so now.”
But other models are far more stable, confident in the balance of public polling, which still shows Clinton ahead.
“[The HuffPost Pollster model is] designed to get more certain as we get closer to the election, so we’re seeing more certainty within the last few days,” Jackson said. “We have uncertainty built into the model for undecideds in the polls and the event that the polls could be wrong, but (clearly) not much. Basically, the approach is that we trust the polls, and if they go down we’re going down, too.”
Powered by WPeMatico