No. I really don't.

Sure, you could design a probability distribution that works this way but

A. How you objectively decide where the line is other than feels?

and

B. How much error does it really introduce to say there's a 0.00000001% chance of Dems going +100? It's like the canard that you'll only lose a small amount of EV if you always fold royal flushes.

Basically your criticism is valid but sort of nitty and could be applied to all kinds of statistical analysis.

@DrModern. Ok, but FiveThirtyEight's probability that any actually impossible result happens is less than 1%.*

A gain of more than 60 seats is definitely live. Dems are 25 seats behind a majority currently. A gain of 65 doesn't even yield a super majority (it gives 59% of the chamber), so your historical anomaly argument doesn't apply.

There are over 60 non-D-incumbent seats where the polling has D's up or within low single digits. A not extreme polling error correlated among similar districts could result in +60.

Yes and the 80’s Jazzercise music makes it 100% better I’m coming for that ass.

I could live to be a thousand and I'll still never get how leg warmers ever caught on

Then why has neither party had >60% in the House since the early 90s? Random chance leading to convergence on the mean? Surely we can do better.

In other words, I'm saying that a number of the non-obviously impossible outcomes are also effectively impossible. I'll grant that a gain of 65 is within the bounds of reason (this is what I meant by "I may be setting the line incorrectly"), but my point is that the 538 model systematically fails to properly discount this scenario and scenarios like it. In this election, it would not be a large mistake to 0 out every outcome above Dems 261. The point of the historical data is just to illustrate certain constraining facts about the underlying reality that the model is trying to estimate. See also pvn's post about Tennessee.

This is absurdly wrong.

Good poast.

Depends what you all mean by large. 538's model puts the chance of Dem > 261 at about 5%. Is it a "large" mistake to zero those probabilities out? Is it "absurd" to suggest that 5% is not large?

It feels like Dr. Modern is taking a macro view of the situation in noticing that, whatever's going on at the level of individual races, the balance stays within 10% of an even split. This could be because political forces and candidate positions will shift to maintain that balance.

I want the Dems to get 270 now just to shove it in Modern's face where as before I would have been happy with like 220

Yes.

What purpose does it serve to assign a zero probability to something that has a 1 in 20 chance of happening? It's foolish.

Agree it makes the forecast worse to zero those out.

At the same time, the 95% likelihood of Dem <= 261 is congruent with Dr.M's confidence. The basic response to that has been that gerrymandering can have an amplifying effect that wasn't as strong in the 90's, so we can't point to historical data. Could be wrong...

No it isn't. He didn't say 5%; he said zero.

Have avoided this cancer forum like the plague recently but always fun to check back in around election time to watch the meltdown. Republicans aren't losing the house and the meltdown will be epic

ya up for some props?

I mean, sure? I agree, it would not be a "large" mistake to zero-out every unlikely outcome, but that seems a bit weasel-y. The Red Sox probably won't score 20 runs tonight, I doubt that's ever happened in a World Series game, so in my model that predicts the score of Game 1 tonight, we'll just have the scale go to 15, it's a small mistake to make my model go to 20 runs when it's very unlikely they'll score more than 15? Ho hum...?

You're of course correct, discounting unlikely events will always be small mistakes, by definition. But how would it really inform readers do that? Can't you just do that on your own if you want to imagine a model with shorter distribution tails?

I don't really understand the criticism here. I agree with Trolly that yours is a sort of is a very bland take. At its most interesting, you're saying Nate is making some modest mistakes displaying unlikely outcomes because his readers might (?) mistakenly assume a very unlikely outcome is more probable than it seems, because they can visualize it, and the human mind takes any extent possibility and aggrandizes it or something?

Ain't following you here. You say we can do better but then seemingly admit these are small mistakes.

I mean, you just did some standard racism, got yelled at then slinked away like a dumb racist last time, what will be different this time?

dunking on people in this forum for being losers?

ofc if dems take the house he won't take it.

You seem to be asking why does FiveThirtyEight give a 1 in 12 ish chance to an outcome that historically hasn't occurred in... well about 12 election cycles. It just seems like a ludicrous question. I'm sorry, and I'm not trying mischaracterize your point. I just don't get it.

I also find the argument "it hasn't happened before (at least for a while)" to be uncompelling. Reagan and Nixon within living memory won every state save one, an outcome that has no chance of happening in the current political climate. The sample of recent, relevant election cycles is necessarily small, and to say some model-predicted outcome can't happen just based on history seems really shaky to me.

Yes. Thank you. This is a good observation. Hope you're doing well, btw.

To be clear, I am also rooting for me to get owned here.

Sorry if I'm not explaining myself clearly. My point is that, unlike in baseball, a number of the allowed-as-possible outcomes are literally impossible, and that more still are virtually impossible. By itself this is not a big deal, but it's indicative of a larger problem, which is that the model itself is systematically overestimating certain outcomes and hence spitting out empirically implausible answers. Here's an example. North Carolina 5th. 538 currently says that the Democrat has a 5% chance of winning. Nonsense. In actuality, the race is essentially already over; Virginia Foxx, the Republican, has essentially already won. Allow a bunch of terms like that across your model and you'll end up spitting out errors with great confidence.

Different context but similar points here: https://www.huffingtonpost.com/entry...b0334571e09e74

So just chop off the 5% tails of the distribution? 1%? Seems sort of pointless.

Well one house seat that probably won't get added to the Dem side is Paul Ryan's district in WI. National money people got behind Randy "Iron Stache" Bryce because they thought he checked a few identity politics boxes and dethroning the mighty Paul Ryan would've been a huge feather in their cap, or at the very least they could keep him occupied on his own race. However, after Ryan announced he wasn't running, it also came out that Bryce had been behind on his child support for a few years and only paid up after announcing his candidacy, had a lengthy record, allegedly used a campaign staffer as sexual bait for a notable endorsement (or at least, turned a blind eye), and was generally a **** candidate. Dems had an opportunity to prevent all of those talking points from being the main focus of the race by simply backing the other run-of-the-mill democrat in the primary. Cathy Meyers would've been a boring candidate and probably lacked a cool nickname, but she didn't have baggage and maybe the blue wave could've carried her to victory. She got to 40% in the primary with almost no help.

Instead, they're now doubling down on all of Randy's shortcomings by running ads touting him as an "everyman" and explaining how he's just like us, the deplorable voters. Bold strategy, Cotton.

This is why you guys can't have nice things.

Also, without using the words Trump or Kavanaugh, would anyone here like to take a stab at explaining why dems continued to support Randy Bryce after it was discovered that he hates women? Clearly, hypocrisy can't be involved here as that only exists on the right.

lmao bro you vote for the party of white identity politics in every election

Can't have nice things...like the House? Care to bet, since you are so sure? Or didn't you already slink away from a betting opportunity by accusing someone else of welching before even making a bet?

But, OK, big step back. Assume you're correct that in the NC-5, the true likelihood of a Democratic victory is ~0% and not 5%.

Didn't we acknowledge that these sorts of errors are kind of minor? The criticism -- assume I share it, we have consensus -- seems to be that they are systemically overstating extremely improbable events and showing them as merely unlikely. I think you admit that's a small mistake. So why are we focused on this again? Nate's model makes systemically but minorly bad ...but ultimately, practically worthwhile predictions, except on the margins? Sounds like a nice model.

Even if I'll concede this point (and I don't know enough about NC-5 to say either way), just how many of these races do you think there are? Even if there are 100 of these 5%->0% elections (there aren't), that's only a predicted increase of 5 seats from "reality".