Joe Biden, most experts say, is beating Donald Trump. If the election were held today, we are told that Biden would win comfortably. Biden’s lead is “the steadiest on record,” says Harry Enten of CNN: “Biden’s up 52 percent to 42 percent over President Donald Trump among likely voters nationally, and he has a 50 percent to 44 percent edge over Trump in the key battleground state of Wisconsin as well.” Biden’s lead is so strong that some are saying Biden may simply “coast to victory.” The Biden campaign has not opened field offices in battleground states and is not doing in-person canvassing, but they do not appear worried. After all, they have the steadiest lead on record. Michelle Goldberg of the New York Times cautions people against “freaking out,” saying that while “as of this writing FiveThirtyEight gives Trump a 24 percent chance of winning” (in 2016 they gave him 29 percent), and while some on-the-ground organizers are “terrified” at the campaign’s inaction, party officials and the Biden campaign have offered assurance that a “digital” strategy will work. Donald Trump’s campaign may be knocking on a million doors a week to Biden’s 0, but Biden is winning in every poll.

I haven’t been nearly as reassured by this, because in 2016 everyone was telling me Trump couldn’t win and I thought they were delusional. I’ve recommended that the Biden campaign get its ass in gear and step up the fight, because after all, it never hurts to have too many votes. But when I’ve said this to people, they’ve responded to me with the signature piece of data: the polls. The polls. And, admittedly, I hadn’t really scrutinized the polls closely, so I assumed they were right that Biden’s lead was comfortable and steady.

Well, now I’ve looked at the polls. And I’m far more worried. So worried that I don’t think it’s even responsible to say that “Biden is winning.” Our working assumption should actually be that Biden is losing.

First, every single time you see a “national” polling average of Biden versus Trump, put it out of your head. It’s meaningless, or at least on its own it can’t tell you whether he’s likely to win the election. This is because we do not live in a system where the person who gets the most votes wins. Instead, we have the Electoral College, which is massively unfair and totally indefensible on rational grounds (people use absurd arguments to stick up for it). But those are the rules under which the game is played and thus the ones which will determine who wins. Hillary won the popular vote. She did not win the election. Biden could lead Trump 60-40 in the national polls and still lose the election if his voters were concentrated in deep blue states.

So what we actually need to look at is states. Specifically “swing states,” the ones where the outcome is fairly uncertain and which might contribute significantly to turning the Electoral College result one way or the other. Let us, then, open up the New York Times’ latest results from critical swing states and see what we find: 

How reassuring! Blue blue blue blue blue blue. +7 in Michigan! We got this. Don’t freak out. Except: what’s the column with some red in it? “If polls were as wrong as they were in 2016.” Hmm. What the Times is saying, then, is that the deep blue column on the left is only a reliable predictor of the result if we make one specific assumption, which is that the polls will not be wrong by the same amount in the same direction as they were in 2016.

And if they are as wrong as they were last time? Well, the Times shows us what the result would be: 

The map on election night would therefore end up looking like this: 

Biden gets Michigan, Minnesota, Arizona. Trump squeaks by in Pennsylvania, Florida, Ohio, Wisconsin. Trump wins a second term. 

So this is the outcome that, via the Times’ calculations, will occur if current polling stays the same but is off in exactly the same way as it was in 2016. You’d better hope to God that doesn’t happen, then.

How concerned should we be about this? Every time I voice a worry, I am deluged online by a class of people I call “Poll Guys.” Poll Guys do not suffer from doubt. They know statistics, and you do not. They laugh at you when you wonder if the polls could be off. No. You dummy. That was 2016. The polls are better now. We’ve fixed them. And then they tell you why the polls can be relied upon this time, why the same thing cannot happen again, because there are fewer undecided voters or the sampling is better or whatever. 

(Actually, to be more precise, that is Poll Guy Type A. There is a second type, Poll Guy Type B, who responds: “Well, of course we think Trump could very well win, Nobody Says Otherwise, we have always said it is a probability. This is Nate Silver-ism. As I have pointed out before, Silverism is where you give a very strong impression of one thing, but then you also say that the opposite could happen too. What happens is that people are left thinking you are reassuring them one outcome can be pretty well relied upon—especially since sometimes you drop the caveats entirely or bury them in the body text—but if it doesn’t, you get to say that you also told them the opposite thing was possible. It is a good way to never technically be wrong while also being functionally useless and giving a dangerously misleading picture of reality to people who will act according to the impression you give while overlooking just how important the caveats were because you did not frontload those caveats or put big flashing emergency lights next to them like a responsible commentator should have done.)

It is not productive to argue with Poll Guys of Type A, because Poll Guys do not allow the submission of evidence other than polls. So, if you have some concern about sampling technique, they will discuss it with you. But if you say something mushy and qualitative like “I’m a little concerned about the fact that I see a billion Trump flags in Florida and hardly any Biden signs, and the lack of canvassing seems troubling,” you will be told that this is not statistics. Which, indeed, it isn’t.

Let me give you a list of three assumptions, though:

Assumption A:

We can confidently assume that the polls will not be wrong in the same direction and degree as they were in 2016. Biden is therefore winning and winning comfortably.

Assumption B:

We cannot assume that polls do not have the same bias as 2016. But we also should not assume that they do. Perhaps the polls are off in Biden’s favor, or perhaps Trump’s. We should therefore be cautious about making any assertion about who we think will win, even if we think it probably favors Biden. 

Assumption C:

We should assume that what happened in 2016 is likely to happen in 2020. Last time the polls underestimated Trump, so we should treat them as if they will do it again. 

Poll Guys, at least those who have called me a fool on the internet, tend to hold Assumption A. Their faith in state-level presidential polling has actually increased since 2016. This is because they believe that polling gets better when we notice the errors, and adjustments are made to ensure the same oversight will not occur again. Thus 2020 Democrats need not have 2016 Democrats’ worries. 

I, however, tend to favor Assumption C. But this is not actually because I believe it to be the “true” or “correct” assumption. Rather, it is because I believe it is the assumption you should work from when you’re running in an incredibly high-stakes election. You want to be extremely cautious and conservative, and it is a perfectly sound form of reasoning to say: “In this election, we will operate on the basis of the assumption that what happened last time may happen this time. This is because we do not want to risk making a terrible mistake. The stakes are too high. Last time we relied excessively on the comforting predictions of pollsters and it was a horrible idea. This time, the pollsters offer us reassurance that everything is fixed. But given what happened last time, that reassurance is insufficient to stake an election on.” (This is also why “existential risk” matters so much. We may think the chance of a nuclear war is on the lower end but if it happened it would be so catastrophic that we must be proactive in trying to prevent it and operate on the assumption that it could very well happen.)

I want Joe Biden to win the election. I have made that very clear, even though I despise Joe Biden and think he will suck as a president. The consequences of Donald Trump’s reelection for the climate, immigrants, electoral democracy, and nuclear proliferation are far too severe for us to contemplate letting it happen. This means that I do not want Biden to screw this up. 

Not screwing up means assuming worst-case scenarios. It means fighting like you think you’re going to lose. That’s especially the case if there is actual hard data showing that all it would take is for the same fuckup to occur twice in order for Biden to lose. 

Ok, but how could the polls be wrong? Why specifically would they have underestimated Trump? Simple: Biden has made a decision that no other presidential campaign has ever made. He’s not running in-person campaign operations in critical swing states. No offices. No door-knocking. No tabling. Nada. Polls might have corrected for whatever errors led pollsters to underestimate Trump against Clinton, but there might be an entirely new error. Certainly, I think the decision not to have a ground operation is a huge X factor that could definitely bump things a few points in the direction of Trump. I don’t know that it will, of course. But unlike Poll Guys, I try not to be too confident.

We must also recognize that everyone is prone to bias. People who don’t admit that they have bias should never be trusted, because bias is most dangerous when it is unexamined. I freely admit that I was biased in favor of Bernie Sanders in the primary. This led me to underestimate Joe Biden’s chances of winning at the time. I endeavor to interpret facts accurately rather than the way I would like them to be, but it’s hard. 

The most important thing is not to be overconfident. After all, the world is extremely complicated. Are you sure there’s nothing you’ve missed? Are you sure your increased faith in polls this time around is warranted? Or do you perhaps want Joe Biden’s “do very little and hope Trump implodes” strategy to work, thereby influencing which numbers you put stock in and which you set aside? Michelle Goldberg’s “don’t freak out” article begins by quoting a ground-level organizer with a lot of experience at voter persuasion, who is deeply alarmed by Biden’s absence from Pennsylvania. But then she talks to lots of party officials, who say that they are confident and we should not be so troubled. Is it obvious, however, that they are right? And is it so obvious that we are willing to gamble everything on it?

I hope I am wrong. I hoped I was wrong in 2016. People don’t think I mean that, because there’s often satisfaction to be had from being right about things. Actually, what I found out in 2016 was that there is no satisfaction whatsoever. That’s the worst part, actually. It feels like warning someone they’re about to get hit by a train, and they don’t listen, and then you watch them get hit by the train. Only a psychopath enjoys that. No, when you’re right about something horrible, you don’t feel good; you just feel sad and scared. I keep having nightmares lately about watching the map turn red on Election Night. If that doesn’t happen, I will be absolutely elated. I will send Michelle Goldberg a big basket of cookies that say “I’m sorry I doubted you!” on them (although I won’t be sorry, because doubt is critically important and it’s far better to doubt too much that you’re winning than to be over-assured of victory). 

In 2017, we published this little cartoon about Poll Guys by Pranas Naujokaitis and I still like it: 

In other words: be careful. The fact that your data said you could never be attacked simultaneously by a bear, an octopus, three crocodiles, a snake, and a shark will be small comfort to you the moment it happens. “My certainty was rational” is not a reassuring statement if it turns out to be your last words.

Don’t count your chickens before they hatch. Don’t wander into the deadly swamp confident you won’t be eaten by animals. Don’t assume Joe Biden is winning. The data suggests that actually, he might very well not be winning at all. If Trump’s past electoral performance is an indication of what his upcoming performance will be like, Biden’s solid lead is a mirage.