Fivey Fox will do for 538 what Poochie did for the Simpsons.
This article is pretty interesting. My takeaway is that what made 2016’s interface more useful also made it much easier to misinterpret: a lot of folks saw Clinton’s 71% chance of victory, and misinterpreted it as getting 71% of the vote or something. When Clinton lost, they thought that meant the model was wrong.
Now it looks like they’re burying the lede, so that before you get to the number you want (% chance of victory), you have to go through some explanatory graphics.
I dunno if they’ll stick to that. It’s pretty annoying, and I hope they change it. But I understand why they’re doing it.
Is anyone actually complacent or confident? I sure as hell am not, nor is anyone I talk to IRL. We all remember what happened in 2016, when we were so damn sure Hilary would win. Now, I cower in terror. I mean, sure, encouraging poll data is better than something showing Trump winning. But I won’t believe in a Biden win until it happens AND Trump is so out of options to contest it that he’s actually out on January 21. Until then, I remain exceedingly cautious.
Timmy is…
Heh heh. Indeed. I guess “complacency” exists in various political flavors. Me, I’m not convinced that ANYONE should be complacent. I don’t care if you are sure Trump will win or you are sure Biden will win. Unless you are a Russian operative who has access to information indicating the fix is irrevocably in, you shouldn’t be complacent no matter what you hope or believe.
This is true. The Cubs actually won a World Series in my lifetime, so I now officially believe that anything can happen.
Not in a single election, since the states’ results are likely to be correlated. If Biden is a 66% favorite in two states, and he wins one of them, then he’s more than 66% likely to win the other one too.
I like the new design. I liked the old one too, but it was too crunchy for most people to understand. Probability and statistics are hard to get right and most simply can’t understand the details. Even some professionals got it wrong by ignoring potential correlations among states.
The new design is trying to ease the reader into a statistics lesson. Good to try, although I see some problems, which I think I’ll suggest to them.
Yeah, and that’s where 538 got it right and everyone else got it wrong. They saw Trump needing to win 4 or 5 states that he only had about a 33% chance in each and figured that there was no chance he could win them all. 538 were pretty much the only ones saying that if he won one of those, he was more likely to win the rest. They’re not independent events.
I’ve not really seen the new design but if that’s the reason for changing it, I support that change. A lot of people completely misunderstood Silver’s numbers, but he said consistently that he was working with limited data and that most polling at the state level was within the margin of error, which left room for a Trump win.
4 months? The election is now about 2 3/4 months away. Was that a typo, or are you expecting that it will take 1 1/4 months to finish counting the votes?
Time and space have no meaning in the time of COVID-19. What year is it again?
That’s fair. I was more thinking of a layman’s verification technique. 50 independant predictions where he should get some that hit over longer odds.
Of course this is also correct, there is a lot of correlation so even 50 samples isn’t independent. Since silver continually tweaks his model I’m not sure how much a comparison between elections is meaningful either.
That’s true, but transparently showing how well calibrated he was in the past gives some confidence in present calibration.
It very well could. We’ll probably know the winner well before that (thought there’s a small chance we won’t), but “counting the votes”? Certainly several weeks, and a month isn’t out of the question.
Anyone else following the forecast?
One interesting thing I’ve noticed is that there’s “spike” in the simulation results around Biden winning 410-420 electoral votes. I call it the “hope spike”. When the model has Biden below about 70% chance of winning, it’s not very likely. However, as Biden goes above about 72% that hope spike becomes the single most likely outcome (that is, the mode, not the mean nor median).
I haven’t seen an explanation of this particular effect of the model. I might have missed it. My hope is that if the actual result is that high, the coattails will be strong enough to take the Senate as well. I know not everyone likes to express hope, but I still have some.
My guess is that effect is due to Texas and it’s 38 electoral votes swinging blue in those scenarios. The other swing states, having closer to 10 EVs, probably make for a smooth curve. When TX flips that makes it look like a large spike.
Biden is back up to a 76% chance in the models up from a low of 67% after the GOP convention. This is the highest level he’s been since June (79% max).
Which is weird because the model was released August 12. I guess they just ran the model backwards to get earlier results. Maybe if they run it back far enough an Obama pops out.