How could Trump Still Win
In response to the 2nd Iraq War failures, Rumsfeld famously described, "known knowns, known unknowns, and unknown unknowns". This statement was criticized, including by yours truly, because this was used to avoid taking responsibility. But his words still ring true. In fact, they help explain the paradox of our current feelings. As of now, President Trump is looking at his lowest chances ever of being re-elected. So, why are we still scared?
Well, because of 2016, obviously. At near this time, the Access Hollywood tapes had been released, and chances were just as bad. Then, the Podesta emails were revealed, and Comey announced re-opening the Hillary Clinton investigations.
So, are there other events waiting to save Trump this time? What is that we don't know? In other words, what are the known unknowns? (Obviously, I can't talk about the unknown unknowns because, well, I don't know anything about them. And what's the point in talking about the known knowns. We already know them.)
The Data
One question we are asking is 'could the polls be wrong?'. That's what a lot of people felt after 2016. Well, that's not me, considering I stressed that 30% chance of the Trump meteor hit. From my perspective, the polls were not wrong. They were misread. That is precisely the position of FiveThirtyEight. One of our problems as human beings is that we overstate a measurement's accuracy. A poll that showed Hillary Clint with a 2-point lead is a statistical tie. And as per my last post, if it was a national poll, she was behind. Where I got it wrong was that I misunderstood what a polling error means. I was thinking of it as if a weight scale said +/- 2 pounds. So, if the scale shows I'm 200 pounds (in my dreams), I could be 198 or 202. However, that's not how polling errors work. On a scale, the error margin is due to mechanical effects. So, if I measured myself on different scales, I would get a more accurate measure of my weight. However, the error margin in polls has to do with the information. So, while a polling average, like RealClearPolitics and FiveThrityEight, protects you from polling assumptions not shared, it does not protect when polls make the same assumption. It also doesn't protect you from missing data since most polls will likely lack the same.
Two ways polls can be disrupted is by a late swing by undecided voters and a drastic change in voting behavior. Both of which could explain 2016.
The problem with a late undecided voter swing is that Biden is not Hillary. Hillary Clinton rarely polled above 50%, where Biden is the opposite. That Biden is consistently above 50%, in fact, speaks better for his chances than having a 10-point lead. Even if Trump wins all undecideds, Biden still wins. But again, remember, it is states that matter. Even here, there is good news. The three states (Michigan, Wisconsin, and Pennsylvania) that would win Biden the election are regularly above 49%. Technically, enough undecideds for Trump to win, but he practically must get all of them. Not just most. However, don't get too comfortable. An undecided shift can change things when combined with other factors, as you will see later.
In the case of voting behavior changes, what happens is that pollsters can't guarantee that their sample represents the real population of voters. So, they adjust. These adjustments are based on past voting behavior, so if there is a drastic change, they will adjust incorrectly. The most often way that happens is by a change in who votes. Pollsters adjust by 'weighing' polls to make their samples match real demographics. The problem is that these weights can only be based on past elections. If voting were to change drastically, the weights will be wrong. This happened in 2016 because "uneducated" white voters had been under-weighted, especially in Pennsylvania. However, 2016 is now part of that electoral history and thus included in the polling weights. The kind of polling error would have to occur with another group. For example, older voters support Trump less for 2020 than 2016, which is part of Biden's lead. Now, what if COVID-19 made them vote less. Those polls would error in Trump's favor because older voters had been over-weighted. I'm not saying this is likely to happen. In fact, I don't think it will. This just presents the kind of scenario that could create the polling error we fear.
Another scenario is if a "non-battleground" state flips. Again, this happened in 2016 with Pennsylvania, Michigan, and Wisconsin. Not surprisingly, "non-battleground" states don't get polled as much, so the data there is less accurate.
Besides these three, pollsters don't work in a vacuum. They have incentives, financial and otherwise, that influence their results, sometimes overtly. This is where looking at polling averages helps, but not when most polls share motivations. The good news is that, because of 2016, most of the incentives are to give Trump a better chance.
And, of course, there is the tightening I talked about in the last post. Consider this math. If Biden is 51% in a poll and has a 10-point lead, Trump has 41% in that poll. And thus, 8% are undecided or third parties. So, as polling is done later, the lead dropping by 5% would not be strange. That's why we should be more worried if Biden drops below 50% than if his lead shrinks. If Biden stayed at 51% and Trump rose to 48%, the lead would shrink to 3%. Here, Biden's chances to win did not lessen. Instead, more Trump voters made up their minds.
Events
Of course, polls don't just change. Events happen to change them. And let's be honest. That is what is keeping us up at night.
So, what about the Russians? That's already happened with a supposed laptop from Hunter Biden. However, again, Biden is not Clinton. The Podesta emails fit with what we already believed about Clinton, but Biden is more trusted. Also, the response has been different. This story initially got blocked by Twitter and Facebook with only media that is basically propaganda taking it seriously. Even if it is not stopping, it is slowing it down. And clearly, slowing it down has had its effect That is why you haven't seen me include any links about it. I'm sure that we have not heard the last accusations about Hunter Biden, up to and after the election. Now the story is growing, but only in conservative media. And that's the problem for Trump. As it grows in conservative media, it helps Trump motivate those who already voted for him. In other words, it hastens tightening, not convert voters. Until the story is taken seriously on CBS Nightly News, it only helps in turnout. In the end, it will be more of a factor in 2024 than in 2020.
As for scandals, we had them as well. Including Biden being accused of rape in the primaries. Again, no effect. While I would be surprised if Ratcliffe or Barr does NOT announce an investigation, the result is likely the same. What is instructive is the real scandal about the Senate candidate, Cal Cunningham, for North Carolina. Still, his polls have not gone down. The takeaway is that we want Trump out, and we are not going to be distracted.
Now, there's the more subtle use of targeted social media to spread misinformation. Again, happening now, though, by definition, I can't say how much Russians are secretly infiltrating us because, well, it's a secret. But just as I don't know the influence, I argue that we overestimate the effect. Again, Biden is not Hillary, and this is not 2016. We are aware of them now. People ran to third party candidates because they were both sure Hilary would win and disgusted with her at the same time. Let me ask. Who is the Green Party candidate? Who is the Libertarian candidate? Don't know? Neither do I. And none of us care.
Of course, there are the unmentionables. You know. Things that begin with 't' and end with 'ist'. Well, do you really think an event that makes us feel less safe would change our minds about this President? I would also include the domestic kind. Here I go back to my continuing stance. Such tactics require more organization than the Trump team is capable of. I can easily imagine a horrific event happening at a critical polling place, but only one. Would that stop voters across the country? Across a state? Would it be more likely to prevent those voting for Trump who has not voted early or by mail? The unmentionable that does give me pause is the death of a candidate. To happen within the days left until November 3rd is thankfully unlikely, but I can't discount it. Its outcome is more complicated than I am capable of going into, so I recommend this link.
One election affecting event that is occurring is COVID-19. Right now, this event only hurts Trump. But it has also created a massive change in voting behavior. Precisely the kind of changes that generate polling errors. Then again, it could make a scenario where we find Biden's lead was even more. Regardless, we won't know which until after the election. Then, there is the possibility that a vaccine is announced. OK, let me correct that. There will be an announcement of a vaccine by November 3rd, probably by tweet. However, this has been preempted. Both Biden and Harris have expressed the same sentiment I feel. I'll take it when someone other than Trump tells me. Given that attacks against their statements have had no effect, I'm quite sure that most others feel the same way.
Another event just recently is the announcement of a peace agreement between Sudan and Israel. However, foreign relation successes usually do not beat out domestic failures. Just ask Carter and H. W. Bush.
One event that could have happen, and I'm surprised that it didn't, is to pass a stimulus bill. When it comes to politics, You should trust Mitch McConnel's instincts over mine, but that still baffles me.
Of course, there could be more events I have not thought of, but those would be unknown unknown. Here's the thing. Remember that it can take as much 2 weeks for an event to show up in the polls. That means any event that would change the polls has already happened. With anything else, we won't know until all the votes are counted.
The Rallies
Here is what the Trump campaign is saying, and unfortunately, there is some logic to it. They can win with rallies. OK, follow me. If you can ignore the racism and fascism, those rallies do look fun. Kind of like the 1997 move, Starship Troopers. And, when Trump gets thousands to attend, they get excited and vote for him, even those that just wanted to see what the fuss was about. The numbers may be small compared to a national debate, and the effect may not last long. Still, he's doing them near the election, in target areas, and a lot of them. Shifting his fortunes in enough states to win a 2nd term is possible. Now there is a lot of motivational thinking happening here and risks he's ignoring. He's also vulnerable in more states than in 2016. Still, we should not be flippant. I still claim that his win in 2016 was through luck, but it was not accidental. His campaign did plan to win by flipping rust belt states, and these rallies were a large part. This is one of the few times I can't say Trump was wrong.
The Debate
So, the 2nd debate has happened, which I intentionally waited for before releasing this post. Unfortunately, it's still too early to know the real effect, so I can only give my opinion, not data. To start, Trump did help himself. The question was, 'is it enough?'. I'm sure Republicans are happy right now, especially senators and congressmen up for reelection. Trump debated like an ordinary Republican politician. Granted, more like a stupid Ted Cruz than a Jeb Bush, but one that conservative could feel comfortable supporting. But again, Biden is not Hilary. With Hilary, that was enough. With Biden, Trump needs to win other than Republicans and conservatives.
And Now Some Corrections
My reasoning has 2 assumptions. Trump will not flip a state he lost in 2016, and he will not win by popular vote. I still stand by this, but Nate Silvers gives a better chance: About a 1 to 3 chance of flipping a state and, if he wins, 1 to 2 doing so by popular vote. (Here's my math. As of writing this, the "Weird and not-so-weird possibilities" gives a 26% for "Trump wins at least one state that Clinton won in 2016". The chance to win by the popular vote is 4% vs. 8% that he wins by the Electoral College.) The odds are still on my side, but we have learned not to ignore a 25% to 30% chance.
So which states do we need to watch? Minnesota and Nevada are the most likely candidates.
They still look good for Biden. However, these are two states that have gotten overlooked by pollsters and have the potential for the kind of polling errors I described before.
Perhaps you caught me pulling a fast one. When I published the last post, I made a mistake and fixed it the next day. I had forgotten that Maine and Nebraska split their electoral votes, giving 1 each to Maine 2nd and Nebraska 2nd. You would think that only 1 electoral vote would not make a difference but enough to need Michigan, Pennsylvania, and Wisconsin instead of just Michigan and Pennsylvania. Both had been won by Trump, but Biden is better positioned to take them this time.
Cheating
What triggers my insomnia the most is that Republicans will cheat their way to a win. And the forecast models cannot include cheating. So, yes, cheating gives Trump a higher probability of winning than we are seeing. I am mostly talking about legal "cheating": reducing the places you can vote, aggressively revoking registrations, etc. This is why I have been tracking who controls critical states' elections and why I don't count on Florida, Texas, and Georgia. We are talking about Trump again being in power because the system is rigged. I'd feel better if it were because too many Americans are racist. That at least would still be a democracy.
The good news is that cheating only works because it is small, so it is more of a problem for smaller local elections. Given the current lead by Biden, cheating alone would not do it. There would need to be other shifts in Trump's favor. So, consider this. Let's say the state that decides the election has 10 million votes. The last poll had Biden at 50%, Trump at 44%, 3% 3rd parties, and 3% undecided. So, the lead was 6%. Let say there was a 2% polling error due to COVID-19 in Trump's favor, so it was really Biden 49%, Trump 45%, and a 4% lead. On election day, the swing voters went to Trump (possibly because they went to rallies). This is the result.
- Biden 49% => 4.9 million
- Trump 48% => 4.8 million
- 3rd Parties 3% => 300,000
However, 2% of Biden's vote had been taken away by cheating. Then, Biden has 4.802 million, and the post-cheat total vote count is 9.902 million instead of 10. The election day result will be this.
- Biden 4.802 million => 48.49%
- Trump 4.8 million => 48.47%
- 3rd Parties 300,000 => 3%
And we are in a recount.
We could also consider Russian interference on election day that same way because most hacks just simulate how state officials cheat: messing with voter registration, shutting down polling places, spreading misinformation, etc. Directly changing vote counts is practically impossible. I could get into the details of why. The short technical explanation is that our election system is already too screwed up to hack. The big difference is that Russians can choose to be blatant about it. Typical Republican cheating must be below 2% because otherwise, it will be found out. Russians don't care. They could choose to interfere, having an as large an effect as they could pull off. It wouldn't get Trump elected, but it would create the shit show they want. (Yes, I said shit show. If they can say it on CNN after the debate, I can say it here.)
My example has a few points to note. The first is that the election results will be closer than the polls. That is because there is no such thing as an undecided vote. Second, my example did not require a dramatic event. If fact, everything that changed a 6% lead to a tie is happening now.
What About Early and Mail-In Voting?
Short answer. Don't get overconfident. Yes, it is good news, but we should treat it as the pollsters do, no different than voting on election day. In other words, if someone has already cast a vote, their answer on the poll would be no different. Now, of course, it does mean the poll numbers will change less. But, remember, expect over 150 million to vote in this election. Even if 60 million votes are cast before November 3rd, that still leaves over 90 million on election day. And, not unsurprisingly, undecided voters are more likely to wait until election day because, well, they're indecisive.
An effect could be "likely voter" models because pollsters are not prepared to deal with this much early voting. If someone has already voted, they are 100% likely to vote, but that makes the polls only if the pollsters ask, which they sometimes don't.
As for cheating, it is good and bad news. Most of the ways Republicans (and Russians) cheat are undone by early and mail-in voting. However, that also gives them more time to think of new ways, and mail-in voting has more technicalities to use against Biden in a recount.
Summary
Basically, Trump needs to get the same people voting for him in 2016 and lesson some voting for Biden. Just turnout won't do it. This lets up break the known unknowns into two groups:
What can take votes from Biden
- Polling error
- A drastic change in voting behavior
- Trump winning a "non-battleground" state such as Minnesota or Nevada
- Cheating
- Affecting voter registration, closing polling places, etc.
- Misinformation through social media
What can add votes for Trump
- Scandal, real and manufactured
- Rallies
- The Final Debate
What is likely to have no effect
- Unmentionables
- Vaccine
- Sudan-Israel peace agreement
A lot scary stuff here, so let me put this all in perspective. Polling errorss could also be for Biden. In fact, when we talk about him winning Texas and Georgia that is the scenario. Except for cheating, the other events could hurt Trump as much as help him. In fact, I'll say this. Yes, Trump could win, but keep your pitchforks ready because he likely cheated.
