538's House model has launched!

[JKellyMap]

@SantaMan

At that point, I started personally framing it as two coin flips away from a Trump presidency.

I love this! I’ll start using it, instead of my sour-milk analogy.

[pulykamell]

@SantaMan

I believe that 538 had a discussion after the last election on how they should describe chances of winning. IIRC, the thought was that people ar a gut level conflated 70% chance of winning with 70% of the vote, or some such.
When they put the (Aug 2016) notes about “Hold on, this election is NOT a lock.”
At that point, I started personally framing it as two coin flips away from a Trump presidency. So I was surprised but not shocked.

Yep. I was the nervous towards the end, especially when I saw Nate’s numbers and, well, it looked like I had good reason to. A 70% chance of winning isn’t that comforting when we saw what we ended up with.

@Locrian

I was a Nate 538 junkie in 2016, even signed up for updates. Maybe it was the graphs. Now I consider Nate less than accurate than Jimmy the Greek

I sure wouldn’t say that. Here’s the thing: let’s pretend for a moment there are 100 state-level races, and, for simplicity, let’s say every single race has a front runner that the model reports has a 70% chance of winning. If the model is correct, then somewhere around 30 of those races should be losses for the model to be correct and accurate. If the model correctly predicts all 100 races, then the model is severely underestimating the probability of the frontrunner’s win (or is extremely, extremely, extremely, extremely lucky. It would be like rolling a 10-sided die a hundred times and never having 1, 2, or 3 show up.) I wouldn’t consider that a very accurate model that describes the state of the election, but one that severely underestimates the chances of winning.

I’m sure’s he’s done it somewhere, but I would judge the accuracy of Nate’s model or any model not based on whether it predicted the winner in a single race, but rather whether overall the “winning percentage” of the model matches up with the predicted winning percentage of the candidates.

[Lance_Turbo]

The idea that the 2016 election was some kind of epic failure on Nate Silver’s part is bizarre. Here’s what he posted on election morning…

Final Election Update: There’s A Wide Range Of Outcomes, And Most Of Them Come Up Clinton

Here’s one quote, but there are several where he make’s the case for how Trump could win.

First, Clinton’s overall lead over Trump — while her gains over the past day or two have helped — is still within the range where a fairly ordinary polling error could eliminate it.

[Sage_Rat]

@Lance_Turbo

The idea that the 2016 election was some kind of epic failure on Nate Silver’s part is bizarre. Here’s what he posted on election morning…

Final Election Update: There’s A Wide Range Of Outcomes, And Most Of Them Come Up Clinton

Here’s one quote, but there are several where he make’s the case for how Trump could win.

Yep, I for one was freaking out that 538 was giving Trump a 1 in 3 or 1 in 4 chance of winning. I’ve done a fair amount of work with and within random number generators to have a good sense for how bad those odds actually were for the sanity of the world.

[DigitalC]

@Buck_Godot

In general I don’t think this would be true, as both A and B supporters would be likely to not bother voting. The case where it likely makes a difference is when, like in 2016, there are a significant protest vote component, where people aren’t voting so much to elect person A or B but to “send a message to Washington”.

Then you have person A voters that they can safely send a message to the establishment by staying home, voting third party, or even voting for B because they know A will win anyway. Meanwhile person B’s supporters continue to come out because it wasn’t really about electing person B but instead simply voicing their opposition.

After the 2016 debacle I don’t think that voters are going to be so careless for some time to come.

I think people are more likely to go out and vote for someone they think is going to win than for someone predicted to lose.

[BigT]

@Sage_Rat

Yep, I for one was freaking out that 538 was giving Trump a 1 in 3 or 1 in 4 chance of winning. I’ve done a fair amount of work with and within random number generators to have a good sense for how bad those odds actually were for the sanity of the world.

I was, too, but I was trying to tell myself it couldn’t happen, and hoping that Nate’s numbers were off, since there were others out there.

What really bugs me is that I don’t know if he was off. We can’t go back and run the election multiple times to see if his percentages are right. Even the 95% chance I was hoping for meant that Trump could win 1 in 20 times.

[octopus]

@DigitalC

No, you’d pretend to support the Democrats, because the day the Democrats actually take over they are going to fuck Russia right up the ass.

With a reset button and would they have enough room to maneuver?

[octopus]

@JKellyMap

Thank you, Lord Feldon and iiandyiiii. I’m sad to see that so many (including Locrian and, it seems, asahi) still don’t get it. Two years ago, in the run up to the 2016 election, I tried to explain this over and over. Here, for example, I likened a 75% chance to having four bottles of milk in the fridge. If you knew one contained brutally sour milk that would make you sick, would you just blindly reach in and grab one of the four bottles and start gulping, because it was a “foregone conclusion” (to quite asahi) the milk would be fine? Of course not!

In fact, as Nate Silver keeps saying over and over, he would be flat-out WRONG if, say, he predicted certain candidates to have a 75% chance of winning in each of 10 separate races, and all ten did end up winning. He’s MORE RIGHT — his prediction was MORE ACCURATE — if some of them LOSE. He was very clear about the many paths to a Trump victory, and Trump waltzed right down one of those paths.

My brain makes all kinds of mistakes — humans are humans — I’m as fallible as anyone, but it would be nice if this concept could penetrate more the consciousness of American voters who look at 538.

It’s like how that 3 doors with 2 goats and a car puzzle still stumps even the educated. And that’s not even a hard problem.

[Miller]

@JKellyMap

Thank you, Lord Feldon and iiandyiiii. I’m sad to see that so many (including Locrian and, it seems, asahi) still don’t get it. Two years ago, in the run up to the 2016 election, I tried to explain this over and over. Here, for example, I likened a 75% chance to having four bottles of milk in the fridge. If you knew one contained brutally sour milk that would make you sick, would you just blindly reach in and grab one of the four bottles and start gulping, because it was a “foregone conclusion” (to quite asahi) the milk would be fine? Of course not!

Anyone who’s ever played X-Com should understand this.

[DigitalC]

@Miller

Anyone who’s ever played X-Com should understand this.

Freaking X-Com taught me anything under 70% is a miss.

[Chisquirrel]

@BigT

I was, too, but I was trying to tell myself it couldn’t happen, and hoping that Nate’s numbers were off, since there were others out there.

What really bugs me is that I don’t know if he was off. We can’t go back and run the election multiple times to see if his percentages are right. Even the 95% chance I was hoping for meant that Trump could win 1 in 20 times.

No, but you can look at all of the results of the model versus reality. If, over a large enough sample, 538’s predictions hold close to actual results, you can probably trust the model. If 70% of elections his model predicts at 70% end in the affirmative, you’re probably relatively safe to trust the results - as long as you take the actual meaning of probability into account.