I feel like these “probability” statistics are nonsense.
Polls are one thing… They are a measure of concrete things, and while they perhaps may have flaws, they are at least a measure of something real.
Probabilities for things like election outcomes seem misguided at best. They are entirely unverifiable, because it’s a single event. You can’t ever have the same election over and over again, to demonstrate that the results actually coalesce around that probabilistic prediction. No matter how wrong their model is, there is no way to really even establish it, because it’s just a chance. They could say there’s a 99.9999999% chance of something, and even if it doesn’t happen, their model could be correct. Or not. There’s no way to know.
With dice, the outcome are very tangible. It’s easily calculated mathematically the various outcomes. I just don’t see anything close to that kind of concrete math behind these kind of probabilities given for election outcomes. It really seems like it’s just some numerical value applied to some guy’s gut.
I realize there’s more to it, but part of me had a hard time putting it into the same group as mathematical probabilities for repeated events. I just don’t have faith that their modelling is that good.
There’s a ballot question in MA to repeal transgender rights. It’s worded in such a way that it’s confusing. Voting No keeps transgender folks as a protected group. Voting Yes repeals that law.
I’d be worried but I think the people voting for Good will take the time to understand the question and vote correctly, and the people voting for Evil will be much more likely to see “transgender rights” and reflexively say No.
EDIT: I have it backwards. An ad on TV just corrected me. Yes keeps the law (good), No repeals it (evil). Fuck I need a flowchart for this question.
I don’t even know how this gets on the ballot. We legalized marijuana years ago and still can’t buy it without a medical card. But this will get right through.
Also, sorry for being a downer regarding the probability and polls and stuff. I like trigger’s coverage of this stuff as much as anyone, and appreciate his efforts.
There is an interesting ongoing debate over whether certain events can have a probability. For instance, is it valid to say that football team A has a one in three chance to beat football team B, if the two teams will only play each other once?
But this does not apply to political polls. Pollsters ask people the same question, over and over again. When voters are asked the same question again tomorrow, it’s very hard to believe that their answer will be completely unrelated to the answers they gave in the past.
Now, it’s true that pollsters don’t talk to everyone, and respondents can change their mind. But there are valid ways to address those problems in order to arrive at a meaningful probability for a given election outcome. It’s the same methods that ensure your car is unlikely to explode, even though factory workers only test parts from other cars.
The probabilities aren’t like dice rolls exactly. The uncertainty isn’t in the dice roll of the election but rather the model. Nate discussed this somewhere but I’m on my phone.
You can judge them by looking at all the small predictions as a whole. If Nate makes 100 90% predictions, you should expect about 90 to be right.
In some ways, these predictions are very similar to sports odds (i believe nate silver was originally an odds maker for sports, wasn’t he?).
And as such, there is some rhyme and reason behind it, because otherwise bookies would be less well off than they are… But I’m not sure how scientific it is. Certainly there’s analysis behind it, but there’s no real “correct” way. Just different methodologies with varying tech records.
I think maybe that’s the real source of my skepticism. I’ve never seen really good, easily parsed days that shows me past track records for the various guys making these predictions that gives me faith. It still seems like a lot of it is trusting their guts on certain things.
Informed, well thought out things, to be sure… But still not the kind of probabilities I’m dealing with when playing poker, or dice, or analyzing how a program is going to run.
Yeah, absolutely. As i said, I’m unsure of that data though… Also, it seems like the landscape over which we are making these predictions is itself changing.
Unfortunately, this isn’t actually true and it’s what makes it so hard to know if the models are right. It would be great if the small events could be this independent, but they are highly correlated. Only by looking at how his percentages do over several separate election days can you get a good sense of accuracy on the predictions.
That said, there is some value to the model when done well - they look at historical examples of how far off polls are, how correlated those errors are, and how they are affected local factors vs demographics vs national factors. They use those things to establish a formula with a bunch of error terms “Joe Blow will get 45% + m + n + p + q” etc., then run randomized trials (or calculate closed form values - not sure which 538 is doing now) of those error factors, where m is specific to Joe and his opponent, n is based on turnout of white non-college voters in Joe’s state, p is Democratic turnout across the country, q is sampling error, etc. So the percentages you see are in some sense the result of a repeated experiment.
I’ve always had the sense that a lot of people stayed home in 2016 because of a lack of enthusiasm for the candidates, but also cause they thought the outcome was determined, which speaks to what you are talking about.
Well, I know that Nate believes it is valid to say that a team has a one in three chance of winning. He is what we call a “Bayesian” statistician. But not everyone agrees.
Note that bookies make odds using a different method, basically balancing bettors against each other without necessarily professing an opinion about what is likely to happen. In other words, if one in three people bet against a team then the bookie odds are one in three, but that is not necessarily their odds of winning.
Yeah, like i said, I’m not saying there’s nothing to it… i know better than that. I’m just expressing my skepticism regarding some of the stuff that backs such use of probabilities, and I’m aware that it’s just an opinion.
One other thing, despite my skepticism here, i also believe that silver is actually much now rigorous in his approach to all of this than lots of other folks on the field. I’m skeptical of him, but i don’t think he’s totally full of shit, whereas some of the other folks (like those saying Clinton had like a 90% chance of winning) were.
Also, all science aside, I’m wearing my lucky pink shirt tomorrow.
I made myself a little playlist on Spotify to warm my progressive libtard heart tomorrow. It’s mostly quiet, pastoral ballads and a few new age instrumentals, and is no way meant to batter away cynicism, defeatism, doubt, or in any way get the ol’ heart pumping a little.
I’m really freaking out that the gerrymandering is going to fuck the Dems despite winning state popular vote totals by 6 or 7 percent, robbing them of even winning the House.
It might, but if enough people vote Dem down ballot and change state legislatures and the like, that can be fixed.
You’ll have to wait a few years, but gerrymandering can be corrected every 10 years… if you bother to stop electing the fucks that keep doing it.
Or maybe the Terminator will save us all, but I wouldn’t count on it. Godspeed to him though.
Started listening to a podcast by Rachel Maddow on Spiro Agnew. This guy was trump before there was a trump - the more outlandish and offensive his rhetoric, the more popular he grew. It strikes me that the worse a Republican politician is, the more Republican voters like him (or her, see Sarah Palin.)
I had planned on taking tomorrow off to watch the returns tonight, but cancelled it. I can no longer stomach media coverage of politics, or how like radioactive osmosis their blather starts infecting others.
Win or lose, going to try me best to just ignore everything for a while.
