It would be trivial to reverse the image left for right. Sadly that’d be all wrong for pipes. Good luck convincing @Colibri to turn the hummingbird around
This pretty much is exactly me.
Did you notice that they are now forecasting “Democrats are favored to win the Senate”?
So I’ve been cross-comparing the Fivethirtyeight forecast with the Economist forecast, and I think Nate’s got the uncertainty on too high.
For example, take New Hampshire. Not a lot of high quality polls, but some recent B-ranked ones, and the polling is consistent around +12 Biden (range +8 to +19). Nate’s model has the probability of Biden winning NH at 89%, or the same probability of winning the whole election. Really? Does Trump really have a 11% chance of an upset against a +12 polling average?
The Economist, using very similar methods (same polls and averages) but a different “uncertainty formula”, is putting NH at 97% for Biden. This seems to be far more reasonable for a +12 lead at this point in the race. Blow up these state-by-state differences, and there’s a lot more (possibly unreasonable) uncertainty in the corners of Nate’s model that add to too much uncertainty in the bottom line.
Now of course you’ll say “but but 2016! Nate’s the only one that got it right!” Well, in 2016, there were a lot of less sophisticated models reporting 99.9% certainty or similar for Clinton. So there, Nate was obviously “more correct”. But he had Trumps chance of winning at 33% or so. Maybe, but perhaps Nate was still overestimating Trump’s odds - perhaps Trump only had 10% chance - winning a 1 in 10 chance will happen, too.
For the overall model right now, Fivethirtyeight gives 89% and Economist gives 95% chance (still a possibility for Trump rolling a natural 20 on a 20-sided dice, so not 99.9% or anything). I think the Economist is closer here.
You go with what the elves of the internet provide.
Question about chance. My math-fu is extremely weak, but I’m having a little confusion about comparing a candidate’s odds in a national election with a one-off dice toss or Russian roulette. The dice toss is a completely random single event. Whereas to win the election, a candidate needs several things to break their way across several states. When 538 says they ran 40k (or so) simulations, the attempt to test with every combination of identifiable factors (i.e., if polling in MI is faulty, but tons of black women vote in GA, etc.)?
Is there a simple way to explain how - say - a 10% chance of winning the presidency is the same as rolling a 10 on a 10-sided die?
I’m no expert but taking Silver’s interactive election map as the reason I think it’s because if a State swings a certain way it likely means similar States will also swing a similar way. If you click on a State and turn it Red, it alters the probability of nearby States also turning Red. No State is in a vacuum.
Let me give it a try.
As I understand it, the 538 simulation models try to take into account various permutations of a number of variables (polling errors, turn-out, etc.). Think of each of those as an alternate reality. Currently, in 10% of those alternate realities, Trump wins the election – the odds of that are, as you note, the same odds of rolling a 1 on a 10-sided die (or a 01 through 10 on a 100-sided die).
However, we only get one reality to live in. While 89% of the realities that 538 is modeling yield a Biden victory, there are still those other 10% (right now, about 1% of their models are returning an electoral college tie). When that 10-sided die (representing which of the possible realities we wind up with) gets rolled on Tuesday, there’s still that 10% chance that it lands on a 1.
Tossing three dice separately is three distinct and random events, so the probability of say, three 6s coming up is 1 in 216.
But how neighbouring states vote aren’t distinct events. There are shared communities, shared media outlets, similar economic / social /racial issues in neighbouring states. So what happens in one state can have influence on what happens in another state. Nor is it random, like a dice toss. People are weighing the candidates and deciding how to vote. If the people in Michigan are collectively tending towards a particular candidate, there’s a good chance that the people in Wisconsin are doing the same.
All the combined possibilities of the 50 states have complicated cross-correlation factors with lots of dice that are “connected” to each other (not independent). The 538 model rolls all these dice 40K times and comes up with 40K results (40K different maps). The 40K are put into a bag, and you draw one map as the “true” election. 10% of them are Trump wins so that’s your chance of drawing a Trump win.
Of course, having figured out those probabilities, there’s no statistical difference between grabbing one of the 4K wins out of a bag of 40K versus rolling a 10 on a 10-sided dice.
Thanks all. So do those "complicated cross correlation factors address the fact that not all of the 10% of chances Trump wins is equally likely. I mean, he COULD win California, NY, and IL, but I think it pretty damned unlikely.
I have no problem trusting the offs related to an ostensibly fair die/coin. But those “cross correlation factors” seem to me like some degree of potential human error/guesswork.
As I understand it, they’re based on what 538 has seen in examining previous elections, and the correlation that they see between certain events. As @Northern_Piper notes, the Upper Midwest states tend to be affected by the same sorts of factors (blue-collar employment, for one), and there’s a level of correlation between Wisconsin going red or blue, and its neighboring states doing the same thing. OTOH, the correlation between what happens in Wisconsin and what happens in Arizona, for example, is likely much lower.
To expand this to a two-step process with dice.
There are actually lots of bags of dice. Assuming you have to roll a 1 on a die to win a state, you might have a “rust belt pessimistic” bag were all the rust belt states are 12-sided. And you have an optimistic rust-belt bag where all the rust belt dice are 6-sided. So for each map, you first pick a bag, then roll the dice.
So yeah, you have to assemble the bags of dice! These are done by looking statistically at past elections to ask “which states moved together?” There is a method here (not guesswork) but it’s based on past elections, so a new realignment of states moving together might be missed. And it’s definitely better than “no correlations”, that’s where a lot of those 99.9% predictions came from in 2016.
Thanks all. I think I’ve got it and am likely just driving myself crazy, fearing the worst. Man, Wed can’t get her soon enough!
I would support that ticket completely.
Unfortunately we will never know how good Obama could have been without Mitch McConnell trying to obstruct every single thing he ever attempted.
Speaking of cross party tickets, I really believe McCain might have offered Vice President to Hillary. There was a long period in that cycle when she was out of the running (Obama could win but she couldn’t) when she refused to concede. I think during that time she was mulling over McCain’s offer. When she finally did drop out and supported Obama she was very anti McCain. I think she leverage the VP offer for her support and the Secretary of State gig.
I feel a bromance coming on…
Not a freakin’ chance. McCain really wanted to name Joe Lieberman as his running mate in 2008 as a bipartisan gesture (and because Lieberman was one of his best friends in the Senate, although a Dem) just eight years after Lieberman had been Al Gore’s running mate. McCain was strongly warned by his top aides - correctly, I think - that he’d have a floor fight at the GOP convention if he did (Game Change by Halperin and Heilemann has an excellent account of this). Hillary was and is absolutely toxic to the Republican base. No way in hell McCain would ever invited her to be his running mate.
I think it would have been greatly beneficial in the long term to the Republican party if they’d had that fight then. They could’ve started cutting out the racism before it had metastasized everywhere. They knew they had a problem then, but it’s too late at this point. Now that they’ve chased out the few remaining reasonable people, there’s nothing left to save.
Yup. I teach Political Geography to undergrads, and we used this tool today as a class exercise.
How did it go?
I have a 15-year-old at home. She and her friends are very angry at all of us, and I can’t say I blame them. We have failed them.