Under current conditions, watching partial counts come in is like trying to guess someone’s weight as they put different body parts on the scale – but you are not told which part is being measured.
Give it time, everyone. This is not a normal election night.
— Sam Wang (@SamWangPhD) November 4, 2020
Live thread starting at 8:00pm. Comments are on!
8:54pm: Under 2020 conditions, television coverage is worse than usual. At baseline it’s low information but high distraction. Don’t watch TV!
9:11pm: The Virginia redistricting reform amendment is passing by a 2-1 margin, with two-thirds of the vote in. Yay!
9:16pm: Florida is looking like a polling miss – maybe as much as 4 points, and Trump is ahead. Surprising, but note that Florida is not considered the decisive state this year. However, I am looking at FL/NH/GA/NC performance to see if there is a pattern of over/underperformance.
9:19pm: NYT has called Virginia for Biden, despite the fact that the vote is 58-40 favoring Trump so far. Same story as the 2008/2012/2016 Presidential election in Virginia: the northern, more Democratic-leaning counties report late. But the nerds know this, and account for it.
9:26pm: Under current conditions, watching partial counts come in is like trying to guess someone’s weight as they put different body parts on the scale – but you are not told which part is being measured. Give it time, everyone. This is not a normal election night.
10:46pm: Sorry, was doing an event. Back now.
Word is that the Minnesota state Senate is likely to go Democratic. They needed to pick up two seats. They are on track to pick up SD-44, and are leading in SD-20 and SD-14. That was within the range of possibilities in our Redistricting Moneyball model (our only model this year). Depending on how the state House goes, it raises the possibility that Democrats will end up with single-party control. They could then hypothetically gerrymander – though our Gerrymandering Project has gotten pledges from some of the winning candidates to adhere to fairness standards.
It seems like all the warnings abt early results have gone unheeded.
— emptywheel (@emptywheel) November 4, 2020
11:05 pm: In North Carolina state House races, word is that Democrats are on track to losing four seats. They won’t be gaining control of that chamber. Note the significant remaining partisanship in the map, even after the court-ordered redistricting last year. However, the fact that they lost seats is pretty bad.
11:18 pm: Wisconsin won’t report results until tomorrow morning. Michigan needs until Friday. We don’t know when Pennsylvania will be done. Basically we’re not going to resolve the Presidential race tonight!
11:28 pm: New Jersey has passed Amendment 3, which delays redistricting every time Census data comes after February 15. That’s true under normal conditions, not to mention this year’s pandemic conditions. It means that population growth in N.J., including growth in Latino and Asian populations, will not be reflected in district lines until 3 years after the Census, in 2023. Not great.
On to the next thread.
The 2020 NYT needles are looking a lot like the 2016 ones.
fox news has biden @ 95%
nc @biden 91%
David McC. wrote: “If Biden loses, the interest in following election ups and downs on sites like this and 538 will completely dry up. Two elections in a row giving people the wrong impression about who’s winning. 2016 could have been dismissed as a fluke, but to happen again, people are going to conclude there’s something fundamentally flawed with how polls are done.” David, that’s not the point of what we did this year. Were you reading at all, or did you just show up today?
I mean, your histogram of EC outcomes showed nearly zero probability of a Trump win, so I think the criticism is fair. People want to know if the world is ending, not esoteric arguments about data.
Considering I’ve been blaring on for months about how that’s not our goal this year, I have no sympathy for you at all.
I assume you are just checking in today, and are unaware of our focus on getting people interested in the downticket and to focus on redistricting.
Prof. Wang, I know Daryl (not David) well, and I think you’ve mischaracterized his position.
Focusing on the downticket and redistricting is a great thing, but if the polls all have huge systematic errors, what’s the point in even using them for this purpose? This isn’t primarily about you, it’s about the pollsters, but *they* all need to step back and think about what they’re even trying to do.
I was inclined to refrain from comment. But Daryl/David’s next comment, which was fairly rude and is therefore gone, could not be interpreted so charitably.
I suspect a mistake in which Latino voters were reached, i.e. they’re a heterogeneous group. Most of the errors look like that. For example, if Latino voters are more like their white neighbors in North Carolina and also in New York, then they don’t vote the same way. However, I have to admit it is starting to get tedious figuring out which polling error was made in each successive election.
Also, to remind you, 2018 and 2019 surveys did fine in Kentucky, Louisiana, and House elections. Though not as weil in Senate elections. And there was a recent SCOTUS nomination. Hmmm.
Florida is showing where Republican strategists have been focusing: Brown votes. Will that carry across the country? We’ll see.
How could most polls be so wrong in Florida? Only ABC News/Washington Post, Tralfalgar Group, and Wick had it right with Trump +2 or +3. Let’s hope Tralfalgar Group and Wick will not be right again on PA, as they had Trump at +2, while ABC News/Washington Post had Biden +7
Predicting the result might not be the point, but if polling is bad *enough*, it’s not even a reasonable spur to action.
Well, that I agree with.
Wouldn’t you need polling to be very bad, randomly so, and also have a strategy of betting it all on the highest marginal state/race for it to wreck the strategy? You can avoid the last one just by diversifying your bets (donations/efforts)
One would have to basically ask how much the average error was. If it’s a lot, maybe pre-campaign race ratings and other data analytics would be better. It would allow effort optimization, but with much less reliance on polls.
Given how much progress machine learning(ML) has made in the last 8 years. (it was 2012 that Alexnet broke imagenet via the use of neural networks). I have to believe some ML based system will do better than polls at a much lower cost and with higher frequency signals. That being said the ML based systems i am looking at predicted FL to go Biden!
@paras the Canadian AI website got it way wrong too: https://advancedsymbolics.com/us-election-2/
My theory was that polls would be off by 5% for Trump, as 10% of his supporters are more honest with themselves and would find it shameful to admit to a pollster that he’s going to vote for Trump. This adjustment would explain FL and NC.
Looks like about 5 points off again in OH. 538 average said tie. Wick had +2 and Trafalgar Group +5 for Trump. Final vote count has Trump +8. :(
Looks like for the remaining states we need to look at those two agencies’ results.
Sam, I have a question: How would we know if parts of the voting was expertly hacked? Are there analytics that would necessarily reveal that? Would these take a lot of time before reaching reasonably conclusive findings?
Mind you, I am not making a claim, nor am I being conspiracy-minded… But are there reasonably clear answers?
I’m sorry, because I know Sam hates this kind of take, but if the polling error is 4% in Trump’s favor in every key state, I will believe that the polling is fine, but gerrymandering, suppression and who knows, hacking of results, is to blame.
The bottom has dropped out of the Fauci scale for me, at least at the moment. :(
Actually some encouraging news on the POTUS front so far. I’m strangely calm. Is that a sign of being in shock? ;)
I think it is OK to be calm.
I am looking through the Senate and state-legislative rubble, which is not a pretty sight at the moment.
I find it very hard to believe that the polls are consistently that far off. As I write this Trump is winning FL (95% votes in), MI, OH (78% votes in), NC(88% votes in). Sam did his very best to make this not about the polls, but this sort of outcome certainly implies that the polls are meaningless, so I have to agree that the question of whether the voting machines are hacked is a very good one.
“There is a potentially material issue in Georgia. A reported 80,000 mail ballots are delayed in being reported in Gwinnett County because of a software error. Those are likely very blue votes.”
A software error? Who woulda thunk!?
Thank you so much for all this great information over the years Sam! I did just turn on CNN for about 3 minutes and I think it took 3 years off my life. Won’t do it again.