Nate Silver

Nate Silver @NateSilver538

@mattyglesias If polling-based models considerably outperform fundamentals-based models in true out-of-sample prediction, that would tend to suggest that campaigns are idiosyncratic and that it's hard to develop reliable stylized facts about elections, which does seem like a valuable insight. — PolitiTweet.org

Nate Silver @NateSilver538

@mattyglesias Yes, I disagree that there's a difference between "modeling designed to improve forecast accuracy and modeling designed to explore causal theories". I think the best way to test whether causal theories are true is generally through true out-of-sample prediction. — PolitiTweet.org

Nate Silver @NateSilver538

Ehh. There's no get-out-of-jail-free pass for designing a p-hacked/overfit model because you say you're not using it for forecasting. Models that don't hold up well in true out-of-sample prediction are usually dubious, and the theories they proffer are often going to be wrong. — PolitiTweet.org

Nate Silver @NateSilver538

@jonathanchait Well, that's why it's a subtle shift! But generally speaking there's a lack of articulation of exit strategies from lockdowns, and to some degree, also a lack of acknowledgment about how costly they are (especially the noneconomic costs). — PolitiTweet.org

Nate Silver @NateSilver538

It's a subtle shift, but at some point in the US discourse, closing/locking things down transformed from something you implement for want of better alternatives—and to be clear, in the US, we probably still lack those alternatives—into the default way to fighting the coronavirus. — PolitiTweet.org

Nate Silver @NateSilver538

Our national polling average has tightened a bit to Biden +7.6. We haven't had that many super high quality polls lately though. Also a bit less clear whether state polls imply much tightening. Still, we aren't seeing as many of those double-digit leads. https://t.co/cy51vc5isJ https://t.co/mmREWWCmfV — PolitiTweet.org

Nate Silver @NateSilver538

Wednesdays seem to be producing high death counts, on the other hand. A very bad number today (1,401) after another very bad one last Wednesday (1,447). With that said, the rate of increase has flattened out a bit. — PolitiTweet.org

Nate Silver @NateSilver538

A decline in cases vs. last week. But this is the fewest # of tests reported in several weeks, which likely reflects some combination of (i) tropical weather reducing testing and reporting in some states and (ii) California having some "technical issues": https://t.co/i1uI2wkFBR — PolitiTweet.org

Nate Silver @NateSilver538

US daily numbers via @COVID19Tracking: Newly reported deaths Today: 1,401 Yesterday: 1,176 One week ago (7/29): 1,447 Newly reported cases T: 52K Y: 52K 7/29: 67K Newly reported tests T: 664K Y: 696K 7/29: 853K Positive test rate T: 7.8% Y: 7.4% 7/29: 7.8% — PolitiTweet.org

Nate Silver @NateSilver538

False negatives and false positives may not be all that common as a share of the general population, but once you get into edge cases, you're not talking about the general population anymore. — PolitiTweet.org

Nate Silver @NateSilver538

I've seen a few too many mainstream media stories of "unusual" COVID cases where the most likely explanation is a false positive or a false negative test and the article doesn't really even explore the possibility at all. — PolitiTweet.org

Nate Silver @NateSilver538

@CathyYoung63 Yeah, I should have been clearer about that. Since the start of the pandemic. — PolitiTweet.org

Nate Silver @NateSilver538

How often are you seeing (in person) friends and family who don't live with you? (Please answer only if you're in the United States.) — PolitiTweet.org

Nate Silver @NateSilver538

I do think there could be pretty decent TV viewership. People will be sitting at home without a lot of alternatives. But probably less coverage in print and digital outlets, in part because the reporters will mostly not be there. — PolitiTweet.org

Nate Silver @NateSilver538

…both parties have now scaled back their convention plans to the point where I don't think we can pretend its business as usual. So we'll probably apply 50% of the usual convention bounce adjustment. I wish there were a more empirical way to do this, though. — PolitiTweet.org

Nate Silver @NateSilver538

This year is forcing a lot of ad-hoc decisions for election modelers. One of them concerns the party conventions—our model traditionally makes a "convention bounce adjustment" which adjusts for the fact that a party's standing is typically inflated in polls after its convention… — PolitiTweet.org

Nate Silver @NateSilver538

He Incorrectly Predicted That Trump Would Win The Popular Vote In 2016. But, Hey, Who Cares About Details Like That? https://t.co/FkWyWFcfAc — PolitiTweet.org

Nate Silver @NateSilver538

@Redistrict If only the opponents were Paul Kroger (D) and Mary Shoprite (I) then this would basically represent my childhood shopping choices. — PolitiTweet.org

Nate Silver @NateSilver538

A somewhat mixed bag. Fewer cases than last Tuesday, but also fewer tests, with some states' testing impacted by the hurricane. Deaths still increasing week-over-week, meanwhile (although only modestly so). — PolitiTweet.org

Nate Silver @NateSilver538

US daily numbers via @COVID19Tracking: Newly reported deaths Today: 1,176 Yesterday: 519 One week ago (7/28): 1,121 Newly reported cases T: 52K Y: 50K 7/28: 56K Newly reported tests T: 696K Y: 732K 7/28: 735K Positive test rate T: 7.4% Y: 6.8% 7/28: 7.6% — PolitiTweet.org

Nate Silver @NateSilver538

@jonmladd @BrendanNyhan Also, from a comparativist standpoint, many world leaders have become *more* popular despite extremely poor economic data because they were perceived to handle COVID well, which also creates issues for economy-driven models. — PolitiTweet.org

Nate Silver @NateSilver538

@jonmladd @BrendanNyhan And if you plug in the before-and-after economic data into most of these models, they predict a much more radical change in Trump's standing (mostly sharply downward -- but in some cases sharply upward!) than actually occurred in the polls, where he only lost a couple of points. — PolitiTweet.org

Nate Silver @NateSilver538

@jonmladd @BrendanNyhan Yes! I think that's absolutely an incorrect story and one not well supported by the evidence. The incumbent's fortunes were already poor, and then got worse because of COVID and maybe the protests. His approval for his handling of the economy is still pretty good! — PolitiTweet.org

Nate Silver @NateSilver538

@jonmladd @BrendanNyhan I'm saying that, in addition to doing a poor job of out-of-sample prediction, the fundamentals models also do a poor job of contributing to our understanding of what drives elections because the conclusions are based on heavily overfit data. — PolitiTweet.org

Nate Silver @NateSilver538

@BrendanNyhan @jonmladd Since you think the Abramowitz model is strong, happy to agree to make it the standard by which we judge these and settle the debate. Here is its prediction for the the electoral vote this year: Biden 996, Trump -453 I'll take the under on Biden. https://t.co/Nb2HFqaLR3 — PolitiTweet.org

Nate Silver @NateSilver538

@BrendanNyhan @jonmladd Since you think the Abramowitz model is strong, happy to agree to make it the standard by which we judge these and settle the debate that way. It's prediction for the the electoral vote this year is: Biden 996, Trump -453 I'll take the under on Biden. https://t.co/Nb2HFqaLR3 — PolitiTweet.org

Nate Silver @NateSilver538

@BrendanNyhan @jonmladd I really wish you'd responded in more detail to that article that I wrote 8(!) years ago. It's pretty devastating and pretty detailed, and was the culmination of months' worth of research. Has never really been adequately engaged with by you or other "fundamental"-ists. — PolitiTweet.org

Nate Silver @NateSilver538

@BrendanNyhan @jonmladd The fundamentals models have performed **terribly** out of sample and they're also a **complete mess** this year. I don't think this is adequately recognized by you or other people in the profession. https://t.co/uzhRzj9s2F — PolitiTweet.org

Nate Silver @NateSilver538

The reason we generally prefer polls>economy is because different modeling choices on the economy can yield radically different answers (i.e. a Biden landslide if you look at GDP or a Trump landslide if you look at income) while different polling averages produce similar answers. — PolitiTweet.org

Nate Silver @NateSilver538

So while we spend a lot of time thinking thru different model specifications, it's mostly *not* about trying to maximize fit on past data. It's more about thinking through edge cases and specifications that will be robust under a variety of conditions. — PolitiTweet.org