In the Election of Madness
Firstly, I would like to thank the people who read my first post. Initially I was hoping 3 or 4 people would read it, but it looks like there was a lot more — I had at least 30 unique viewers! Again, if you want to add anything please comment or message, it’s one of my favourite parts of writing :)
In the first part, I argued that polling performance in 2016 isn’t as bad as the media often portrays, and made the claim that there is a pathway for Trump to take the presidency this election. I am now going to make the case that Trump is polling fairly poorly and stick to my original claim that Trump still has a chance at taking the election. Hopefully I’ve convinced you that I at least trust polling. But it’s a tricky relationship.
To begin, let’s take a look at how Trump is performing on the state level
Trump’s state performance
On the state level, Biden is way out ahead, having captured republican states like North Carolina, Arizona, and maybe Iowa (it’s dead even). Further, it’s a coin toss on whether Biden wins Georgia, Ohio and Texas. All of the states I have listed so far, Biden is competitive in. The crazy thing is, he could lose all of these states and still have a good chance of winning. If we give Trump all the above states and throw in an alligator infested Florida, Biden is still leading at 60% to win so long as Biden can take Pennsylvania. That might sound a little arbitrary and weird, but what it shows is how important certain states are (and the assumptions that being made about similar states).
This graph orders states based on the probability of them being blue or red. The line needs to be crossed by either party to take the election.
This graph shows that as long as Biden takes Pennsylvania, the above states on the winding graph can be lost.
This is all to say that on election night, the important states that we should be drinking to are Pennsylvania, Florida, Wisconsin and Michigan. These states are the states that are the most likely to decide the election; although you might have noticed they are not listed in what I called competitive states — Trump is unlikely to take them.
On another note, you might think Biden taking Texas would be a really big deal, and it is, it’s very significant in the grand scheme of politics. But if Biden wins Texas, chances are that he’s already won the election, meaning that it’s not a ‘tipping point’ state.
To put it in numbers, Pennsylvania has more than a 25% chance of deciding the election, Florida almost 15%, Wisconsin 12% and Michigan just under 10% while Texas is roughly 1%. The reason being is that there are not many worlds in which the electoral votes are even and Texas has the remaining votes that would break the tie.
Here is the projected vote share of these four important states:
Biden is ahead in all of these states bar Florida by more than a normal polling error. There’s more to say about these numbers, but let’s talk about Trump on the national level first.
Trump’s National Performance
On the National level, things are looking fairly bad for Trump too. Biden has been pulling ahead steadily since early June, with major gains in October. Currently, he’s just over 9% ahead of Trump, this doesn’t happen too often at this point in the race. We should keep in mind that it wouldn’t be unprecedented, per se, if Trump came back. In 1980, Reagan came back to beat Carter in a swing that was over 10 points; Reagan up 8 points and Carter down 3.
If Trump remains 9% behind nationally, I don’t think there is an electoral map that makes Trump win. At 5%, it’s unlikely but it is possible. We can see in the graphs below that races usually tightens towards the end. I don’t know of a good reason that that won’t happen with this election, Nate Silver from FiveThirtyEight thinks that it most likely will, but we’ll have to see (it looks like it might have started happening over the last couple of days).
The other really important point here is that Pennsylvania is polling at 6% ahead. One way to think of this is that Trump doesn’t need to make up that full 10%, he just needs to make up 4% of it in Pennsylvania. This is a fairly crude metric, but what I’m getting at is that the National level isn’t the most important metric, it’s a kind of ‘how are we doing test’. The National poll says Biden is doing good, we then look at the state polling and see Biden is not doing as well as the National polls would suggest. The electoral college is weird.
2020 Prediction
With that all said, here’s the final prediction from FiveThirtyEight.
I cannot reiterate enough that this graph is not telling you that Biden is going to win the election. It’s saying that if we had to live through 100 election days then Trump would make us consume roughly 26 bottles of wine, two for each win. Importantly, there’s a 35% chance that Biden wins in a landslide, and a 9% chance that Biden wins the popular vote and loses the electoral college. That’s a 65% chance for some regular chaos and 9% chance for some 2020 level Chaos.
Speaking of 2020 chaos, I feel like we’ve been here before…
The Horror of 2016
“His coming was foretold and now the Messenger manifests. Behold his works writ large across the world, their thunder echoes from the first of seconds to the last chime of the final hour.” — Eldritch Horror, Prophecy Fulfilled
It might have been a long time ago, but I still remember early 2016. We were all so young and naïve. Britain was happily married, Harambe was alive, “clown invasion” wasn’t a credible news headline, and the US was about to have their first female president. Except that all turned to shit. Over the next few years we learned to live with the trauma, only to be slammed with 2016 round 2, 2020.
On a slightly more serious note, I think these events make it kind of hard to believe that Trump will lose the election this year. If it was some other boring Wednesday (Tuesday if you’re in the states) afternoon of a year, then sure, he doesn’t have a chance. But the perpetual disaster of 2020 gives Trump a home field advantage.
If there is a way out of the despair, it is through what the Lovecraft mythos has taught me; science and math keeps the horror at bay.
Let’s begin by asking the obvious; how accurate is polling this year?
In perspective of the 2016 polling error (which wasn’t that large), we can see the results of the battlegrounds states.
So even if we have a sizable state polling error that swings in Trump’s favour, we can see that Biden is still in the lead. Remember, Biden doesn’t need all of these states. He also doesn’t need the most competitive states; Texas, Iowa, Ohio, and Georgia. Biden is polling better than Clinton, and the states that will give Biden the victory are pretty much the same; Florida, Pennsylvania, Michigan, Minnesota. You can play around with this tool to create electoral maps and see how interesting maps might play out on election night.
It is also important to point out that Clinton never achieved Biden’s state or national lead in the entire race. Her highest was 8% and that was fairly early. This time in 2016, Clinton was ahead in national polling by 6%.
We can see that compared to historical trends, it would be almost an unprecedented polling error, should Trump take the election without polls changing.
Is there any reason to think there is a large polling error? The funny thing about polling errors is that if pollsters know about them, they fix them. One prominent theory this election is that there are shy Trump voters who are afraid to tell the nice lady on the phone that they are voting for Trump. It sounds plausible — especially to democrats. No democrat wants to say that they support Trump, so why would a Republican… right? This theory isn’t given any credit by FiveThirtyEight for good reason. Have you met a Trump supporter? I don’t think I would call them shy — but this isn’t good evidence, it’s only anecdotal.
One good piece of evidence we do have is that when polling occurs online or by mail, the results are practically the same. If people are shy about telling the person on the phone then they are shy about it in more anonymous settings too — I’m not sure why this would be. Perhaps they will be shy on the voter ballot too?
The other piece of context we have for this is the shy racist voter. In 2012 there was a theory that posited that people told pollsters on the phone that they supported Obama because they didn’t want to tell the nice lady they weren’t going to vote for the nice black man. Obama outperformed his polling average by ~4%
The Outcome of the Election
So far I think a lot of the statements I have made about Biden being ahead and not counting Trump out have been confusing. We live in a world where events are concrete; our intuition tells us events happen or they don’t. It would be good if we had a clearer way of thinking about the election.
In one way, Trump having a 12% chance means that he can come back. Trump recently held a rally in Pennsylvania, maybe that went really well? Maybe Covid-19 relief will come through soon and sway voters more than expected? Maybe news of a vaccine being closer will inspire confidence? On top of all that maybe there’s a polling error slightly larger than last year. In that world, Trump has a very real chance of winning the election; that world isn’t high fantasy.
In another way, it means that Trump is probably going to lose. Democrats have such deep penetration of republican voters and Biden is doing well where he needs to do well. In terms of polling error, we don’t know which way it will sway either. Perhaps it will swing Biden’s way and the election will be a landslide. We know that pollsters have made corrections from 2016, so perhaps they lean too heavily towards Trump? I don’t think we should psychologize pollsters too much and buying into a Biden landslide is a little naïve, but it’s certainly possibility.
Again though, the above two paragraphs don’t give you something material to hold onto, they don’t make me feel like I know what is going to happen on election night. This is, of course, due to rule #1 of polling; polling doesn’t tell us who is going to win the election. If they did, they would tell us who has won the election.
There is perhaps another way of thinking about this that can marginally save us from the immaterial nature of polling.
Trump is about a normal polling error from this being equivalent to rolling a 1 on a dice ~17% chance. On election day, roll a dice and imagine that on a 1 Trump wins. I’ve lost board games on those odds. Sometimes the world has been at stake in those board games.
