01 November 20
It's Live on the HomePage Now:
Reader Supported News
What a Difference One Good Day Can Make
Raising RSN’s basic budget was a long hard struggle in October. On the very last day however we made a big comeback and avoided significant losses. There was no windfall, but we did make up a lot of ground in the last 24 hours.
It’s a reminder of what our readership is capable of when they are motivated. It’s also a reminder what would be possible if we were not waiting for the very last moment.
To all of those who sustain Reader Supported News a heartfelt thank you.
There is a big job and a big battle coming.
Courage.
Marc Ash
Founder, Reader Supported News
If you would prefer to send a check:
Reader Supported News
PO Box 2043
Citrus Hts
CA 95611
It's Live on the HomePage Now: FOCUS: Nate Silver | Trump Can Still Win, but the Polls Would Have to Be Off by Way More Than in 2016 No, I don’t know who’s going to win the election. According to our forecast, President Trump still has a chance at a second term: a 10 percent chance, to be more specific. But — even though we’ll still get a ton of polls on Sunday and Monday — I’ve seen enough based on the polls we got earlier this week to know that things aren’t likely to change all that much in our forecast between now and just after midnight on Tuesday, when we’ll freeze it. There just hasn’t been any real sign that the race is tightening. If anything, Joe Biden’s margins are expanding slightly in the Upper Midwest. And there isn’t any particular reason to expect the race to tighten when more than 90 million people have already voted and the most important news story — that the United States just set a record for the number of COVID-19 cases in a day — is a negative one for Trump. In fact, in many states, such as North Carolina, we’ve gotten what are likely to be the final polls of the state from most of the major polling firms. The one important exception is Pennsylvania, which some high-quality pollsters seem to have kept as the last state they’re planning to poll. And those polls could matter quite a bit. Pennsylvania is the most likely tipping-point state (it delivers the 270th electoral vote around 37 percent of the time in our forecast), so any deviation from Biden’s current 5.1-point lead in the polls there — say, if Biden climbs to a 6-point lead or falls to a 4-point lead — could make a fairly big difference in our forecast. But what we’ve seen so far in Pennsylvania doesn’t suggest much movement in the polls. We’ve gotten two live-caller polls since the debate: A Muhlenberg College poll published this morning had Biden up by 5 points, closely matching our average in the state. And a Quinnipiac University poll had Biden ahead by 7, which is not quite as good for Biden as it might seem — Quinnipiac has generally had friendly results for him this cycle and their previous poll of the state had him up by 7 as well. And if nothing changes at all in the polls, Biden’s chances of winning will nonetheless increase slightly by Tuesday morning in our forecast. That’s for two reasons:
That said, Biden’s current lead of 8 to 9 points nationally is quite large given our highly polarized political environment, so maybe a few of the remaining undecided voters will drift to Trump. Don’t be surprised if Biden drops to 86 percent — or jumps to 94 percent — in our final forecast. But I don’t think that Biden and Trump are likely to escape the current zone that they’re in. Here’s what I mean by that: I think of election forecast odds as basically falling within the following four zones.
The Gray Area is the closest zone, when it’s actually hard to tell who’s ahead, and different methods of averaging the polls or modeling out the results might give you different answers. The race isn’t in that zone this year: Biden is unambiguously ahead in the polls. The Normal-Polling-Error Zone is a place we talked about in 2016, when we told you that Trump was only a normal-sized polling error away from beating Hillary Clinton. What did that mean? It meant that if polls were off by about the amount they’ve been off in past elections — by around 3 points, on average — and the error favored Trump, then he’d probably win the Electoral College. And that’s basically what happened, although the polling was worse in some states than others. In probability terms, I think of the Normal-Polling-Error Zone as extending from the favorite having anything from around a 60 percent chance up to around an 84 percent chance of winning. That amounts to an error of no more than one standard deviation.1 Or to put it another way, it’s a zone where polling errors big enough for the underdog to win are going to occur quite routinely without anything particularly special having to happen. Polling is a challenging business, and while polls get the outcome right more often than not, nailing every election to within a point or two is hard. In probability terms, I think of the Normal-Polling-Error Zone as extending from the favorite having anything from around a 60 percent chance up to around an 84 percent chance of winning. That amounts to an error of no more than one standard deviation.1 Or to put it another way, it’s a zone where polling errors big enough for the underdog to win are going to occur quite routinely without anything particularly special having to happen. Polling is a challenging business, and while polls get the outcome right more often than not, nailing every election to within a point or two is hard. The Zone of Plausibility. This is where we are this year. I think of the Zone of Plausibility as extending out to reflect an error of up to two standard deviations — so, it’s a race where the favorite has somewhere from an 84 percent to 98 percent chance of winning. You wouldn’t consider the underdog winning in an election like this to be a routine occurrence. But, well, it’s plausible, and it isn’t that hard to find precedents for it. The polls were off by more than 7 points in 1980, for instance, underestimating Ronald Reagan’s margin of victory. (That would likely be enough for Trump to win in an election where he trails in the most likely tipping-point state, Pennsylvania, by 5 points.) Harry Truman beat the final Gallup poll in 1948 by 9 points in an upset victory. And the polls missed by 5 points in 1996, underestimating Bob Dole. Now, we can debate exactly how applicable those precedents are today. There’s much more polling this year than in 1980 or in 1996. And in 1948, it wasn’t “the polling” that was off since there was just one polling firm, Gallup — maybe if there had been a Quinnipiac poll or something back then, it would correctly have forecasted Truman’s victory. The point, though, again, is that a Trump win is plausible. And all the other models I’ve seen have Trump within the Zone of Plausibility too, although the Economist’s model, which has his chances at 4 percent, is pushing it a bit. At the same time, though, a 2016-style polling error wouldn’t be enough for Trump to win. In the chart below — taking a page from The Upshot, which has also been doing this — I’ve taken our final polling averages in 2016 and shown how they compared to the actual results. And then I’ve shown what the results would be based on this year’s polling average if the polls were exactly as wrong as in 2016 in exactly the same states. Takeaway? Joe Biden would win. In fact, he’d win 335 electoral votes, including those in Florida, Georgia and Arizona. A lot of these wins would be close — he’d win by around 2 points in Arizona and Wisconsin, by and less than 1 point in Florida, Georgia and Pennsylvania, so he’d have to sweat a bit, but he’d win. Meanwhile, although there are a lot of uncertainties this year that our model tries to account for — for instance, whether pollsters are correctly blending the early vote with the Election Day vote — there are also other things that could make a big polling error less likely. For instance, the polls have been very stable so far in the race, and the large number of people who have already voted makes a last-minute shift even less likely. There are also few undecided voters: Joe Biden is polling at above 50 percent in all states that Clinton won except Nevada, plus he clears that line Wisconsin, Michigan, Pennsylvania (albeit just barely; he’s at 50.1 percent there) and Nebraska’s 2nd Congressional District. Those are enough to give him 273 electoral votes. What’s beyond the Zone of Plausibility? Well, there are The Outer Reaches. But you don’t want to visit. It’s a cold, barren place, full of esoteric debates about whether probability distributions should have fat tails (ours do!) and how much you can distinguish, say, a 1-in-100 probability from a 1-in-1,000 one when you have only 15 or 20 elections in your sample to work from. But we’re not in the Outer Reaches, and we’re very unlikely to wind up there before Tuesday. A Trump win remains plausible. And note that, with his 10 percent chance, our model is specifically referring to a legitimate win; we do not account for what we call “extraconstitutional shenanigans,” by Trump or anyone else, such as trying to prevent mail ballots from being counted. Still, Trump isn’t in as strong a position as he was in 2016. As you can see in the table above, he’s polling worse than he was against Clinton in every single battleground state. Polls can be wrong — indeed, the whole point of our probabilistic forecast is to tell you the chances of that — but they’re more likely to be wrong when a candidate’s lead is narrower. As of right now, Biden’s lead is large enough that Trump’s chances of winning are 10 percent, considerably lower than the 35 percent chance he had at this point in 2016. |
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.