I’m sitting here writing this as of 1:55pm EST on Wednesday, November 4th. All the data you see was collected as of 11:00 AM EST. I will update both as we get more results.
Here is what the election looks like so far. Biden will be President. He will win, literally, by one Electoral Vote.
Trump looks like he will win in Pennsylvania, North Carolina, and Georgia. Biden looks like he will win in Arizona, Nevada, Michigan, and Wisconsin. This creates a dead-even tie at 269 electoral votes each. But, Nebraska and Maine award their electoral votes by Congressional District, rather than winner takes all for the state.
**Historical side note: This was the way the Founding Fathers intended the Electoral College to work. It began to change when Thomas Jefferson convinced Virginia to go winner-take-all to help him win the presidency. James Madison and Alexander Hamilton freaked out, because this was not how electoral votes were supposed to be tabulated. Other states followed, and now this is how 48 states work.**
So, because Nebraska allows a split vote, Lincoln, Nebraska voted for Biden, giving him one additional Electoral vote. This would put him over the top at 270-268. Expect Republicans to fight this as hard as they can in the courts.
It is still possible that Biden can win any three of North Carolina, Georgia, and Pennsylvania. The votes left to be counted appear to be mail-in ballots from heavily democratic areas like Charlotte, Raleigh, Atlanta, and Philadelphia. It is still possible that Trump can win Nevada and/or Arizona, so things could change. Again, I’ll update this as we get more results.
What I wanted to do here was an early wrap up for everyone who has been asking me for lots of detailed info. So, here it goes.
WAS THIS WHAT YOU EXPECTED?
No.
My model had a median outcome of Biden winning 325ish electoral votes. The most common outcome, however, was below the median—305 electoral votes. In this most common outcome Biden won North Carolina and Pennsylvania, which it currently looks like he would lose.
Instead, it looks like Biden will squeak by while losing those two states, and win by one electoral vote.
I had expected things to go more or less along the lines of the model, but they did not.
HOW LIKELY WAS THIS OUTCOME?
The model did show this outcome was possible. The fact that Biden could lose Pennsylvania, Florida, and North Carolina but still end up winning was a big reason why the model only gave Trump a 16 percent chance to win. Trump always had a tough hill to climb, even if he outperformed predictions (which he did).
In the model, an outcome where Biden won by less than 10 electoral votes was more likely than a Trump win—19 percent to 16 percent. This particular outcome was not more likely than a Trump win, but that is because there were multiple different ways Trump could have won. This particular outcome was more likely than most of the Trump-winning outcomes.
I'M CONFUSED, HOW DO
YOU MEASURE LIKELIHOOD?
In my model, I factored in the polling for each state, then ran 1,000 simulations (each time I added new polling data I did this again) to randomly assess the likelihood of outcomes for each state. Electoral votes were given for each outcome.
So, for example, the model gave Biden a 51 percent chance to win Florida, and a 53 percent chance to win Pennsylvania. So, over the 1,000 simulations we would expect Biden to win Florida in 510 of them, and Pennsylvania in 530 of them. But, the most important thing to understand is that these were random draws, so in any given simulation it was more or less a coinflip for each. It could be Biden-Biden, Trump-Trump, Biden-Trump, or Trump-Biden, and that is perfectly fine.
**Side note: Other models build in dependencies on other states to factor in voting trends. I did not; for me every state is independent of the other. This is something I’m happy to discuss the merits of, but this is how my model operated**
Over the course of 1000 simulations, the model balances out the randomness to provide likelihoods because things become less random on average the larger a sample gets. That’s stats nerdy, sorry if it’s a bit confusing. Essentially, you can think of it like a coin flip. You know a coin has a 50/50 shot at head or tails. But, if you flip it 20 times you should not be surprised to get 13 heads. That does not mean the coin is broken, there is a statistical probability formula that actually says this is expected a large percentage of the time. However, if you flip the coin 1,000 times you would expect to get really close to 500 heads and 500 tails. That’s what I mean by “less random on average the larger the sample gets”.
In the model, given that Biden had a 51 percent chance to win Florida and a 53 percent chance to win Pennsylvania, he won them both more often than he lost them both.
So, likelihood in the model was just how many times out of 1,000 a result happened in a random draw, where the probability of a candidate winning was based on polling numbers.
I adjusted polling numbers to reflect historical accuracy, but that’s not terribly important for this wrap up, so I won't bore you with the details.
So, when I say “In the model, an outcome where Biden won by less than 10 electoral votes was more likely than a Trump win”, it means that this happened more often in the 1,000 simulations.
OK, ENOUGH NERDY STUFF.
WHAT THE HECK HAPPENED?
Here’s where it starts to get really interesting. In 2016 Neither Trump nor Clinton really did all that well. Wisconsin is the perfect example. Trump got fewer votes in Wisconsin than Mitt Romney got in 2012, but Mitt Romney lost by a big margin, and Trump won. I’ve always explained this by looking at Trump and Clinton’s favorability numbers. They were the most unpopular and untrusted presidential candidates (as far as I can tell) to ever run for office.
The polling on Trump’s performance during his first term indicated that he really didn’t win any new fans. Think about Wisconsin in 2016; the common knowledge (which I agreed with) was that Trump would have to expand his base to win in 2020, but polling indicated he was not expanding his base.
Turns out that was incorrect. Check out these two graphs. These are overall Republican vote totals in each state. The first is 2016 vs 2012, the second is 2020 vs 2012. You can see that in 2020 Trump really did pull out a lot of new voters across a lot of states. (look at the difference in the X axis)
Both Trump and Biden pulled out millions of new voters for their party. Trump did far better in 2020 than he did in 2016. This was unexpected. Biden is far more likeable and trusted than Clinton was, so the common wisdom was that Trump would struggle to expand very far beyond his 2016 base. As it turns out, he didn't really add much to his base. He really struggled with women, minorities, young voters, and really everyone except white men, and particularly white men without a college degree. What we learned yesterday is that his 2016 base of white men was far larger than people had anticipated—and far larger than the vote he got in 2016.
Why did this happen? There are a few possible explanations, but it is really hard to know which one is correct without large-scale detailed polling of voters.
There really were a huge number of “shy Trump voters” who loved Trump, but were afraid to say so. I have always been skeptical of this, so it may be time for me to eat some crow. But this is not the only possible explanation, so don't cook any up just yet.
Expanded early voting and mail-in voting was not the advantage for Democrats people assumed it would be. It is true that lower income and minority voters have a harder time getting to the polls because they cannot always take enough time off work on election day. Allowing them to vote early or through the mail makes it much easier (and probably much more likely) that they will vote. This was understood (myself included) to be a disadvantage for Democrats. However, there has been a pattern of electoral shift. Now, college educated white voters are more likely to vote for a Democrat, and non-college educated white voters are more likely to vote for a Republican. This means that making it easier for low income voters to vote would not have just made it easier for minorities to vote, but would have allowed a larger number of non-college educated white voters—Trump’s biggest base of support—who are also more likely to be poor, to vote.
In 2016 there may have been a huge number of Trump voters who just stayed home because the media narrative was that Clinton would win easily. They may have liked Trump, but figured he never had a shot, and didn’t bother “wasting their time” voting. **Side note: I hate the idea that voting is a waste of time. It’s not. You should do it every election.**
It is also possible that Trump simply won over new voters. Like I mentioned, this looks less likely in the data, so the real explanation is most likely some combination of the first three factors (I’ll take a small serving of crow).
Just for fun, here are those same two graphs for Democratic votes. You'll notice again how low turnout was in 2016.
Biden also pulled out millions more votes than Clinton. Mostly because he was more well like and more trusted, but we also cannot discount the amount of anti-Trump vote that Biden benefitted from. Very far left voters really didn't like Biden, but they REALLY didn't like Trump, so they were motivated to vote for Biden anyway. It appears the number of new votes he pulled out above 2016 will be just enough to give him the win.
SO, WERE THE POLLS WRONG AGAIN?
If you have been reading my stuff, or had a conversation with me, you know I have been saying since 2016 that the polls were not wrong that year, with the exception of Wisconsin and maybe a tiny bit off in Pennsylvania.
But this year it seems like they were off. I will not make a final judgement on that until all the votes come in, because we really can't judge the accuracy of the polls until then. But, as it stands they do seem to have been off this year (again, this might change once all the votes are counted).
Nationally, Biden had a consistent 7-9 point lead in the polls. Right now, his national lead is about 2 percent. There are still a lot of votes to count in California, but I doubt it is enough to get Biden to an 8 point win in the popular vote. Even if Biden adds 3 million more votes in California, and Trump adds 0, Biden’s national vote lead would only be 4.5ish percent. That is a decent sized polling error, and would be outside most national poll’s confidence intervals.
This was also true in individual states. Most states appear to be inside the confidence interval, which in nerdy stats talk means they technically weren’t ‘wrong’. However, they consistently missed pretty far on the Trump friendly side of the confidence intervals.
For example, in Florida, as discussed above, my model showed Biden with a 51 percent chance to win. The average was Biden 51, Trump 49. However, the confidence intervals were roughly Biden 48-54, and Trump 46-52. Right now, Florida is 51/48 for Trump, so that is inside the confidence interval. Technically, that means the polls were still correct, but only barely.
However, it seems like Pennsylvania, Wisconsin, and Michigan were all decent size polling errors. By decent size I mean 2 percentage points or more outside the confidence intervals. Biden’s lead was simply large enough to withstand this polling error in Wisconsin and Michigan, but—at least as it stands now—not in Pennsylvania.
In one state, being in the confidence interval, but far to one side would not indicate polling error. However, given that this has happened in a good number of states, as well as in the national popular vote polls, it looks like there was a polling error that gave Biden a bigger lead than he actually had. North Carolina, Ohio, and Arizona have turned out (again, so far) to be almost exactly what the polls predicted. But there are 3 states with large misses (so far), but only 1 (possibly 1.5 depending on how you looked at Pennsylvania) in 2016. Biden’s total popular vote share is also outside the confidence interval, which was not the case for Clinton in 2016. So, yes, it looks like there was a polling error, but let’s wait until 100 percent of votes are counted to say that for sure.
WAIT! DID I HEAR YOU RIGHT? BIDEN IS WINNING BY MILLIONS VOTES? WHY ISN'T HE THE WINNER?
Yes, you heard me right. Biden is 3 million votes ahead right now. He is not the winner because popular vote does not matter. Fun fact (maybe not so fun if you’re a Democrat), in the last 30 years Republicans have won the popular vote only one time, in 2004. Prior to that you’d have to go back to 1988, when your favorite political stats nerd was only 3 months old, to find a time Republicans won the popular vote.
In the United States votes are awarded based on the Electoral College. One vote for every federal representative (Congresspeople and Senators) a state has. This is one reason the Census is hugely important. Election winners can distribute elected officials in a way that heavily benefits their party. Look at this map of the 2010 electoral college. Red are states where Republicans have more Electoral votes than they should based on their population; Blue is the same for Democrats. Yellow is where Republicans have less votes than they should; green is the same for Democrats.
While yellow and green appear to be close to even (big states just have a disadvantage) you’ll notice there is a lot more red than blue. That was not an accident. Republicans largely won control of the last redistricting process in the 2010 elections. The electoral college gives small states a much larger proportional impact than big states, and they leveraged that to help their party. If you think that’s not fair, blame slavery.
Yes, I said slavery. The Constitution has what’s known as the three-fifths compromise. It says that a slave counts as 60 percent of a human (odd, I agree, but that is what we are working with).
So, let’s think about a hypothetical slave state called, I dunno, Rississippy. Suppose they had 1,000 ‘people’ and 500 ‘60-percent people’. Their population for the Electoral College would not be 1,000, but 1,000 + .6*500. Or, for you math nerds following along, 1,300 people. So, only 1,000 people could vote, but the Electoral College makes those votes all count 1.3 times as much as a vote in a non-slave state.
Thankfully everyone is a real ‘people’ now that we fought a war over this, but small states still have a huge advantage in the Electoral College, and this advantage is compounded by the winner take all approach not intended by the founders—the one that Madison and Hamilton thought was a disastrous development.
I won't get into the merits of the Electoral College because they are not relevant to this election. I can do that another time if you’d like. The fact is, whether you hate it or love it, the Electoral College is how we choose our President, so Republicans can—as Trump expressly did—make a strategy of winning the White House that has no intention of winning the popular vote.
If you have any other questions about the election, i’d be happy to blab on about those as well. Just send me an email:
31mattyoung AT gmaildotcom.
One last personal thing. I’m considering starting a newsletter that would be fairly similar to what you just read. Policy, economics, and stats explanations of what is happening in the real world, written in plain english. If this is something you’d like this—full of typos and all—delivered to your inbox every day, please email me and say so. It would help me make up my mind about whether or not I want to do it.