How Winners Convince Themselves They're Going to Lose

From my presidential modeler, a quick peak into how each well-polled state is looking now, relative to the 2016 outcomes.

We're 10 days out. By my estimate, Biden is up by over 9 points nationally, and by 6.2 in the current "tipping point" state (WI, in case you're curious).  There are over 54 million early votes already cast, about 40% of the total from 2016 (and already far more than the early votes that year). Democrats have a very heavy lead in early votes cast, not only in absolute numbers (nearly a 2:1 ratio), but more importantly, in their return rate.  In places like CO and OR (both important tests because they exclusively vote by mail), Dems are 36% and 46%, respectively, more likely to have returned their ballots already than their Republican counterparts.

In other words, things objectively look pretty damn good. And yet... Muh 2016.

People are still so shellshocked from the last time around that they are simply having a tough time wrapping their heads around the facts that things are fundamentally quite different this time around.  A lot of this is well trod ground, but there's at least three points worth hammering into the ground:

1. There are far fewer undecideds or 3rd party this time around, which means that last minute decisions or 3rd party protest votes aren't going to make the difference.

From my trend tool. For most of the race, there were ~20% undecided or 3rd party voters in 2016, which means that even when/where Hillary had the lead, she did so with ~43-44% of the vote.  This time, that number has been around 6%, and Biden routinely breaks 50% in most national and battleground polls.

2. Biden's lead is larger and far more stable.

Same tool as above. Since June, Biden's national lead has hovered between 8 and 11 points.  By contrast, Hillary's 2016 lead was only in that range during the early summer (convention bounce) and earlier.

3. The national lead reflects a very real, and consistent shift from 2016.

The dotted line represents y=x – if this year's polling simply matched the 2016 results.  The solid line shows a 6.4 point shift to the Dems (almost exactly matching the WI tipping point estimate).  With few exceptions, wherever there's polling, we see more or less the same shift.  

Put all together, this paints a picture where Biden is on track to cruise to victory.  My own model puts the odds at 97.5% if the election were held tomorrow (and since it's basically being held today, that seems reasonable).  538 is a little more conservative (I've critiqued their model previously), giving him only an 87% chance.  Meanwhile, The Economist splits the difference, at 91%.  The point is that, looking at this in terms of numbers, and with the hindsight of mistakes made in 2016 (pollsters, for their part, now generally include education level in their modeling), Trump faces very long odds for reelection.

But if all of that is true, why do all of my friends (and certainly my enemies) insist that Trump is probably going to win?

I'm no psychologist, but here's how they game it out...

Let's start with the states that are really, truly in play.  Actually, let's start with every state currently polling within 10 points – many of which are clearly and truly not in play, just to illustrate how far Biden is ahead:

From my 2020 modeler.  This requires a zoom-in, and even so, as you can see, it's pretty noisy.

To give you a sense of what an absurd error 10 points would be, we'd include MO, KS, IN, MT, AK, SC in that batch, along with CO, NH and... PA, MN, MI.  In other words, just as a start, I've given you a list of 6 red states that it's ridiculous to think will go for Biden, but by similar margins, we've got a bunch of important possibilities for Trump that Biden is leading by 8-9 points.

But just for fun, let's get rid of those "absurd" states like SC and focus only on ways that Trump can win. We're left with 14 states grouped into roughly 4 categories:

1. CO, NH, MN, NV

aka Clinton states where Biden is only leading by single digits

Most of these have been sparsely polled, but the closest, NV is still a lead of 7.2% by my estimate.  It's also only 6 EV, so unless Biden loses NV as a harbinger of other losses, it probably won't make the difference. Clinton states+PA,WI,MI minus NV, for instance, leaves Biden at 272, and the presidency.  NH is worth 4, so that alone wouldn't do it either.  MN could, but again, the lead is around 7.4, more than Trump's lead in MT or SC.  Short version: it's unlikely that one of these will flip without something else having gone wrong.

2. MI, PA, WI

aka Clinton's failed midwest firewall.  

I get that every poll that shows Biden only winning by 3 in one of these (or partisan propaganda polls from Rasmussen or Trafalgar actually showing Trump winning) cause you deep anxiety.  But looking at the recent high quality polling (I use B or better according to 538), here are the margins:

MI: 8.6
PA: 7.7
WI: 6.2

(If I don't say so, assume the number means that Biden is leading)

It's not outside the realm of possibility that Biden loses one of these, but it's unlikely.  It requires a systematic polling error along with a specific polling miss. Yes, that happened in 2016 in WI, in particular.  It's unlikely to happen this time around for the reasons I mentioned above.

3. FL, AZ, NC

aka The heartbreakers.

I am old enough to remember 2000.  Also, 2016.  Also, 2018.  It's not difficult to look at Florida elections through the lens of close races where somehow Republicans just manage to eke out a victory.  I'm not accusing them of cheating, but ... well, actually, I am accusing them of cheating.  And I think that's the point.  Here are the current margins, per my model:

FL:  4.8
AZ: 4.6
NC: 2.9

North Carolina is genuinely a nail biter, but the others aren't. It would actually be a polling miss to assume that Trump is going to win either.  By contrast, consider 2018 FL Senate polling, which gave Nelson a 2.4 point lead, but he lost by 0.2 (a 2.6 point error, which wouldn't be enough in any of these cases).  Or 2016, where (per the realclearpolitics average, which differs in methodology from mine), Trump had a 0.4 point polling lead, and then won by 1.2.  

Incidentally, while RCP shows a much tighter race, only a 1.5 point lead for Biden in FL (including as they do, obvious partisans like Raz and Trafalgar), they still have Biden leading.  Indeed, all of RCP's numbers are tighter than mine. Their national race is only 7.9 points, for instance.  This is pretty clearly a function of their inclusion and equal weighting of low quality polls.  By contrast, 538 has a national average of 9.7 (a little higher than mine), and, for what it's worth 3.3, 2.2, and 2.9 points in the 3 states, respectively.

And I get it, those numbers seem close.  But even if there's a bigger national polling miss in 2020 than 2016 and it's still in Trump's favor (a combination of which is only about 16% likely), Biden would still have about a 50-50 shot in each of those states.  The odds of losing all of them is quite low.  Also, and I can't stress this enough, he wouldn't actually need them.

4. TX, OH, IA, GA

aka The reaches.

There are people who will walk around simply insisting that this list is out of reach and that it's absurd to think that Biden can win it.  But here's the thing.  Not only is it possible for him to win them, most of them are essentially coin flips. My estimates (with 538's in parentheses):

IA: +2.0 (+0.3)
GA: +1.3 (+1.0)
OH: even (-1.2)
TX: -2.4 (-0.5)

Unlike with the heartbreakers, 538 and I have a very similar read on the situation. Unless there's a big polling error, we're likely to win 1 or 2 of these, making almost everything else moot (especially if we win more than just IA).

So what's the takeaway?

Look, you're worried. I get it. Polls could be wrong. But no matter how you crunch or combine the polls, you come up with a result that Biden is up by 8-10 points nationally.  If anything close to that holds, then he's the next president.


  1. What a fantastic blog. Hope you don't mind if I post here a bunch in the comments.

    RE your post about the dislocations occurding on PI. It's funny, as I was just analyzing arbing between 538 and PI and how successful that strategy would have been historically, especially in the top 6 or so states that are battlegrounds.

    What I find to be utterly hilarious is

    AZ is at 6c, and yet of all the markets on PI is the close to 50/50. Serioiusly, these type of shallow dislocations speak to the very poor predictive capability of PI.

  2. People are making crazy money on PI. 850 per contract, ~1000 contracts? That's 850K total tradeable.


Post a Comment