Twitter | Search | |
Nate Silver
YouGov has Dems with an 89% chance to win the House. (They didn't list the probability in their write-up but you can calculate it from the chart.) I think their range of outcomes is too narrow (and I'm not super in love with MRP) but there's something pretty interesting here…
Reply Retweet Like More
Nate Silver Nov 4
Replying to @NateSilver538
...namely, that they show Democrats winning 30 seats (they need 23 for the majority) with a popular vote win of "only" 5.5 percentage points. Suggests the breakeven point for Democrats (when they become more likely than not to win the House) is a 4-5 point popular vote win.
Reply Retweet Like
Nate Silver Nov 4
Replying to @NateSilver538
People have been using a 7-point popular vote win as a benchmark for D majority, but that's *probably* out of date, and a little too high. Doesn't account for PA redistricting, GOP retirements, Democratic $$, and other ground-level factors that have helped Democrats.
Reply Retweet Like
Nate Silver Nov 4
Replying to @NateSilver538
Our Classic and Deluxe models have the breakeven point at 5.7. Our Lite model, on the other hand, expects Dems to waste a lot of votes in blue districts and has it at 7.2. YouGov looks like they have it at 4-5. So, there's considerable uncertainty. But 7 may be on the high side.
Reply Retweet Like
(((David Shor))) Nov 4
Replying to @NateSilver538
I'm glad to finally see someone else on twitter in the anti-MRP caucus
Reply Retweet Like
Nate Silver Nov 4
Replying to @davidshor
It's a good way of getting you 80% of the way to a good answer, but we already have a lot of ways to get to 80%.
Reply Retweet Like
doug rivers Nov 4
Really? How do you correct for selection bias in the polls you’re averaging? MRP avoids the phantom swings in your model estimates in 2012 and 2016.
Reply Retweet Like
Nate Silver Nov 4
Replying to @doug_rivers @davidshor
There was a *real* swing toward Trump late in 2016 and that's one reason our model was more accurate than most other people's. It didn't overly smooth the data.
Reply Retweet Like
Nate Silver Nov 4
Replying to @doug_rivers @davidshor
I get that selection bias is a problem. I think there's also a point at which you smooth away a lot of the information in a dataset. Also, as says, this doesn't have much to do with MRP, which doesn't do too great out-of-sample when you don't already know the result.
Reply Retweet Like
doug rivers Nov 4
You don’t seem to understand how it works. It’s empirical Bayes without informative priors.
Reply Retweet Like
Nate Silver Nov 4
Replying to @doug_rivers @davidshor
I think the confidence interval on your House forecast is **obviously** wrong (too narrow by a factor of 2-3X) and suggest you're a little too caught up in the technicalities and have lost the forest for the trees about understanding the real-world uncertainties here.
Reply Retweet Like
doug rivers Nov 4
There are sources of error that are hard to calibrate, but your claim that the moe of +/- 13 is 3x too small is **obviously** wrong unless you think Republicans winning 249 seats (our estimate of 210 + 39) is a real world possibility.
Reply Retweet Like
Nate Silver Nov 4
Replying to @doug_rivers @davidshor
What do you think the odds are that Dems win 45+ seats? You really think the odds are 50:1 against, or anywhere in that ballpark? That's what your model says.
Reply Retweet Like