The best kittens, technology, and video games blog in the world.

Wednesday, August 26, 2020

Why 538 election model is wrong and Biden has an easy win ahead

#TrumpYourCat by jillccarlson from flickr (CC-BY)

538 releases a model predicting what they think is going to happen every elections, and this time they released a crazy one, where "Biden is only slightly favored to win the election", and Trump has insanely high 30% chance of  second term.

This is in stark contrast to everyone else's predictions. For example The Economist gives Trump just 9% chance of winning, and this is a much more sensible number.

In this post I'll explain why 538 believes elections are quite close, and why it's completely wrong.

It's not a terribly sophisticated analysis, but it's too long for a tweet, so I thought I'd put it here.

If you're interested, I wrote "A Guide to 2020 Elections for non-Americans" a while back, and I dislike both Trump and Biden very much.

Why 538 model believes what it does

Their model is very sophisticated, but its basic assumptions leading to their crazy conclusions are fairly straighforward:

  • Biden is far ahead in polls, however:
  • there's still a lot of time until elections
  • things are likely to change until then
  • especially since pandemic year is full of events
  • Electoral College benefits Trump, so he will likely win even if he losses popular vote narrowly
  • the economy went to hell earlier this year, but they think it will improve, and that will somehow benefit Trump

Why 538 is completely wrong - not much time

The only thing they got right is that Biden is indeed far ahead in the polls. They're wrong about everything else.

The elections are a lot closer than they think. In normal times, most people vote on Election Day. This time due to the pandemic a lot more people will vote by mail early, and mail voting opens about a month before Election Day, depending on the state. So for a lot of voters, time until elections is only half of what they think. And people might be voting even earlier as there's a lot of talk about Post Office having trouble processing votes in time.

Even if Biden says the dumbest thing he ever said one week until elections, or Trump single-handedly invents coronavirus vaccine, enough votes will be cast by then that it probably won't matter.

Why 538 is completely wrong - pandemic prevents most campaigning

People don't change political beliefs based on clocks ticking, they do based on events and campaigning. This time due to pandemic there's going to be a lot less campaiging, and people's attention will be largely on other issues.

It's not even certain if the usual debates will happen or not. And even if they do, only the first of the three is supposed to happen before mail voting starts.

Why 538 is completely wrong - people already made up their mind

If you compare Trump's approval ratings with all past presidents, Trump's are the least volatile.

People love Trump, or hate Trump, and there's a lot fewer people in between that could be convinced by the campaigns. So no matter how good Trump's campaign is, it will likely not matter. And likewise, while Biden will almost certainly win, he's also very unlikely to win by 1984 style landslide.

Why 538 is completely wrong - Electoral College advantage can go either way

In 2000 and 2016 elections, Democrats won the popular vote, but Republicans won the presidency. While this might give an impression that Electoral College system inherently benefits Republicans (the way Senate arguably does), if you look back both parties benefited from the system about equally often.

Right now it might look like 2020 map will be similar to 2016 map, and seems to benefit Trump. This thinking is completely mistaken.

If very little changes, Trump will lose terribly, and while he might indeed lose Electoral College by less than popular vote, it will be completely inconsequential.

For elections to be even competitive, fairly big changes must happen. But while it's possible, there's no reason to expect that they'll uniformly apply to all states (in proportion to their "swinginess").

For example if Biden says something that will offend a million Californians and makes them stay at home or vote third party, this will significantly reduce his popular vote share, but it won't matter for Electoral College one bit, since California will go Democratic regardless.

Trump needs not only a really big swing - he needs a swing that's either eerily uniform, or happen in just the right states.

In reality, if a major swing happens, it throws all assumptions about Electoral College advantage out of the window. Even if much bigger swing than could be reasonably expected happens, and Trump somehow manages to reach vote share similar to what he got in 2016 (losing popular vote by 2%), it's actually still quite unlikely that this swing will leave his Electoral College advantage intact, and get him another victory.

Why 538 is completely wrong - the economy will not favor Trump

A popular way to predict elections is by looking at "economic fundamentals". The thinking goes that whenever economy is doing poorly, people blame whoever is in charge for it, and vote for someone else. This is generally true, and after every major economic crisis like 1929 or 2008, governments all across the world got a lot less popular, regardless of which part of the political spectrum they were on, and how much or how little they had to do with causes of the crisis. There's no reason to expect 2020 to be any different.

538 really hates fundamentals models, and for a good reason. While bad "economy" affects people's voting preferences, there's no obvious way to pick a single number for what is the "economy". You can use unemployment, or GDP, or disposable income, or stock market, or so many other indicators, of combinations of indicators. So in year were economy is doing so-so, you can't really use those models to predict much.

Instead of stopping there, 538 somehow decided to ask experts where the economy will go just before elections, and since most expect some recovery from the worst of the lockdowns, they count this as benefiting Trump. While there's a room for reasonable disagreement how to model this, there's no way in hell 2020 economy could possibly benefit Trump, people's memories are longer than a few weeks, and US is still and will likely be in state of mass unemployment.

Are 2020 Elections a foregone conclusion?

Democrats are overwhelmingly likely to win the House - they'll win the popular vote, and there's not even a possibility of Electoral College stopping them. Republican House is about as likely as Nicola Sturgeon becoming the next Prime Minister.

Biden is extremely likely to win presidency - things would need to go really damn well for Trump.

Senate is another matter. Only 1/3 of it is reelected each time, and Republicans do have a meaningful structural advantage there, so it seems fairly close to even who's going to win. And it probably matters more than anything else this year.

If Democrats control the Senate, they'll be able to abolish filibuster, pass pretty much whatever they want, and then possibly also pack Supreme Court with ten new liberal judges that will pronounce all of that Constitutional. They probably won't go that far, but in principle they could.

In Democrats keep the House, and Republicans keep the Senate, then it will most likely be four more years of gridlock regardless of who's the president. The parties hate each other so much, they won't compromise on anything beyond the most routine operations of the government. Biden's policies will be different than Trump's, but presidential policies don't matter as much as people thing, an the next president in 2024 might very well reverse most of them.

And yes, I obviously put money where my mouth is.


Mr. Stone said...

Well, I guess time will tell, but last time there were multiple polls 1 day before election showing 99% Hillary, while 538 had 65%... They have lot of credibility - but if you are so sure, there are betting markets, you can make good money there.

taw said...

Their 2016 model was much more reasonable.

This time they made a lot of weird assumptions that all benefit Trump, probably as some kind of overreaction to 2016.

And yeah, I'm going to make a bit of money on this (not that much, I'm just not used to the whole betting thing, as very rarely bets at such crazy rates are offered).

taw said...

I'm not the only one who noticed 538 being crazy.

538, Aug 26 2016: Clinton +4, 18% other/undecided, 81% to win
538, Aug 26 2020: Biden +9, 7% other/undecided, 70% to win

Their 2016 model made sense. Their 2020 model is just madness. Biden has 2x the lead of Hillary Clinton, there's not even half as many undecided voters to persuade, and somehow they give Biden a lot lower chances. Pure madness.

Quickshot said...

Guess I'll agree with you on that 538 seem a bit to cautious in their assessment here. Your points are reasonable and I was thinking it seemed a bit weird that a polling advantage that large seemed to lead to no real difference in win chances.

Though I'll note that your link in the comment seems from some one with some potential biases as well.

taw said...

Oh hey Quickshot. I wonder where you're hanging out these days.

Looks like my first prediction already came true in just one day, and top Democrats are already talking about not debating Trump, to avoid any risk of upsetting Biden's lead.

Of course they don't phrase it that way, but it's pretty transparent why they're considering it, and they definitely wouldn't if Biden was polling behind.

Not like it wouldn't be that unusual, similar thing happened in Poland a few months ago, and it happened in other countries too.

That tweet might be phrasing things in a confrontational manner, but Biden having lower modelled chances of winning with much bigger polling advantage, much lower number of undecided voters, and all the other factors I mentioned is crazy.

taw said...

And I missed that story, but it looks like Trump wants even more debates than usual. Interestingly he uses same argument as me:

"The Trump campaign had argued that the current schedule would render the debates useless to the many Americans who will vote by mail before they take place. [...] Members of the Commission on Presidential Debates on Thursday rejected the Trump campaign’s request for changes to the fall debate schedule, declining to shift the debates earlier or add a fourth debate to the calendar."

For a lot of voters only 1 debate will happen before they vote instead of the usual 3, and unexpectedly good debate performance was his best shot, so Trump is really screwed here.

538 completely ignores this.

Quickshot said...

Hi taw and I'm not really hanging out anywhere new, same old places really. I did notice recently that Trump gained a little polling wise, not sure what caused that, though Biden remains well in the lead.

Guess we'll see if that erodes any more... or increases any more in the coming time.

taw said...

Quickshot: It's totally plausible for Trump to gain or lose a few points.

My point is that the gap is way too big, and realistic changes are too small, for Trump to have any shot at actually winning.

RCP numbers show Biden between +7 and +9 from June onwards, a few points better for Biden than his pre-recession lead of +4 to +6. Pretty much the same for 538 numbers.

(RCP graphs are very noisy due to their methodology; most of this noise is not based on voters actually changing their mind; 538 shows pretty much no change at all, ever)

Quickshot said...

I agree, time is getting short for much of a change, especially with mail in ballots starting much sooner. Beyond that I just saw that Biden got a small favorability boost and Trump a small favorability decline from their conventions. Well assuming favorability means much of anything in the polls, but still it seems like something that makes a turnaround by Trump less likely yet.

My previous comment as such is just being a bit cautious, more wary of some kind of unknown unknown or known unknown some how changing things.

taw said...

Quickshot: And the voting already started last week! People can mail-in their votes in North Carolina, and more states will soon join.

Trump needs something huge to happen, and he needs it fast. Probably during the first debate, anything after that will have very little impact as a lot of votes will be cast by then.

Now this is completely speculative, but people who analyse election spending are saying Republican donors pretty much gave up on Trump and is putting their most spending in states with competitive Senate races. Which honestly is the best strategy for them.

If they keep the Senate, I think Republican establishment will be very happy to see Trump gone.

Quickshot said...

Interesting, I guess that could make some sense, yeah. Put the money where it will still have the most useful effect.

Two months or so still before the election result I guess, so it'll be a little wait till we can see the actual result, but yeah.