Nate Silver

Nate Silver @NateSilver538

Of course, because of mediocre testing, we're still only catching a small-ish share of infections, by most estimates. If, per @youyanggu's model, we're catching one-fifth to one-sixth of cases, that means the *infection* fatality rate (IFR) is somewhere in the range of 0.3-0.4%. — PolitiTweet.org

Nate Silver @NateSilver538

One thing I like to look at is the lagged case fatality rate (CFR), which is the number of deaths divided by the number of cases some time ago. Over the past week, the lagged CFR has been 1.5% using a 2-week lag. If you prefer a 3-week lag, it's 1.9%. — PolitiTweet.org

Nate Silver @NateSilver538

Good chance deaths will increase further. But it does seem likely that the death *rate* from COVID is decreasing, which is good news, even if the number of deaths is increasing because we have so many freakin' cases. — PolitiTweet.org

Nate Silver @NateSilver538

Other numbers are looking more pleateu-y lately, especially in FL/AZ/TX, but a plateau isn't good news given how many cases there are currently; these states should really be aiming for declines and ideally fairly steep ones. — PolitiTweet.org

Nate Silver @NateSilver538

Very typical of recent numbers. While fewer deaths were reported than on most recent days, reporting on deaths is generally slow on Sundays; deaths were up slightly from last Sunday, in fact. — PolitiTweet.org

Nate Silver @NateSilver538

US daily numbers via @COVID19Tracking: Newly reported deaths Today: 523 Yesterday: 872 One week ago (7/12): 476 Newly reported cases T: 64K Y: 65K 7/12: 61K Newly reported tests T: 769K Y: 762K 7/12: 727K Positive test rate T: 8.3% Y: 8.6% 7/12: 8.4% — PolitiTweet.org

Nate Silver @NateSilver538

@kmedved @DanSoch @jburnmurdoch @MattZeitlin @felixsalmon @notanastronomer @Samfr Yeah, I'm not totally sure what's going on. It's a weird story either way. And it doesn't help that Sweden has been pretty half-assed about testing, especially early in the epidemic. — PolitiTweet.org

Nate Silver @NateSilver538

@kmedved @jburnmurdoch @MattZeitlin @felixsalmon @notanastronomer @Samfr Half-assed distancing + being in a place where you can do everything outdoors in the summer + nontrivial seroprevalence could be enough to keep R<1 for now, even if Sweden isn't very close to herd immunity if it attempted business as usual (in cold-weather months). — PolitiTweet.org

Nate Silver @NateSilver538

@kmedved @jburnmurdoch @MattZeitlin @felixsalmon @notanastronomer @Samfr Sweden's also interesting in that there's been relatively little behavioral change over the course of the pandemic. It's NOT that they didn't do any distancing, etc. They did. But they haven't oscillated back and forth between different behavioral regimes as other countries have. — PolitiTweet.org

Nate Silver @NateSilver538

It's sort of interesting which COVID-19 studies get widespread media attention. Not very well correlated (in fact, perhaps negatively correlated) with the ones the subject-matter experts I follow seem to find compelling. — PolitiTweet.org

Nate Silver @NateSilver538

One other thing to be aware of is that most national polls currently are conducted among registered voters, while most state polls are among likely voters. So unless you're accounting for that, you may overestimate the state poll vs. national poll gap. — PolitiTweet.org

Nate Silver @NateSilver538

In fact, this is something we re-examined and tweaked in our averages this year. We still make a likely voter adjustment. But it sticks more to a prior (which in this case, assumes a *slight* Trump gain among LVs) because it's not very predictable how it evolves over the cycle. — PolitiTweet.org

Nate Silver @NateSilver538

Consistent with this, but one thing I've noticed empirically is that the likely voter vs registered voter gap tends to fluctuate a lot over the course of the cycle, and from pollster to pollster. It's a fairly noisy measurement. — PolitiTweet.org

Nate Silver @NateSilver538

RT @ABCPolitics: .@NateSilver538 says that mail-in voting is a concern for President Trump, “but not because of abuse or fraud.” “By consi… — PolitiTweet.org

Nate Silver @NateSilver538

Got in a few arguments yesterday with folks who IMO are underestimating uncertainty and a bit overconfident Trump will lose. But one thing that does seem nearly certain, and that lengthens his odds, is that he won't change his approach to the presidency. https://t.co/2g2yP7yegK — PolitiTweet.org

Nate Silver @NateSilver538

One thing I'm increasingly convinced of is that when it comes to working with data, building models, etc., it likely helps to be a bit self-taught. — PolitiTweet.org

Nate Silver @NateSilver538

@RachelBitecofer @gelliottmorris @DanRosenheck @Nate_Cohn @collprof @Michael_Cobb68 I don't know it kind of seems that by having had by far the most conservative model in 2016, I might have some valuable insight on this topic. — PolitiTweet.org

Nate Silver @NateSilver538

RT @HelenBranswell: @NateSilver538 @CT_Bergstrom @mbeisen Too many people jumping to conclusions for which there currently aren't data. It'… — PolitiTweet.org

Nate Silver @NateSilver538

@HelenBranswell @CT_Bergstrom @mbeisen I don't know what it is, but the coverage around this set of issues (immunity and its implication for vaccines) in the mainstream press has mostly been quite bad. Places that invest more in dedicated health/science coverage seem to be doing a much better job, though. — PolitiTweet.org

Nate Silver @NateSilver538

Also in the spirit of not making predictions: I'd consider next week to be pretty important, data-wise. You can squint at this week's data one way and think we're headed to 100K+ cases/day. You can squint at it another way and see a plateau. More uncertainty than usual. — PolitiTweet.org

Nate Silver @NateSilver538

I try to avoid making any predictions in these threads. But I'll link to those made by others! The @youyanggu model expects deaths to continue to rise though early-mid August, although it could be more of a long plateau. https://t.co/Qxu8rSPoV1 https://t.co/39Ivhptz5q — PolitiTweet.org

Nate Silver @NateSilver538

The other metrics are a bit more equivocal; cases are increasing, but there are also more tests (though still huge shortages/backlogs/delays in many states) so the positive test rate is staying relatively constant. — PolitiTweet.org

Nate Silver @NateSilver538

Basically we're getting a pretty consistent story here. Deaths continue to rise week-over-week. An average of 742 deaths were reported over the past 7 days, as compared to 678 for the 7 days before that. So it hasn't been getting *massively* worse, but it's still getting worse. — PolitiTweet.org

Nate Silver @NateSilver538

US daily numbers via @COVID19Tracking: Newly reported deaths Today: 872 Yesterday: 951 One week ago (7/11): 757 Newly reported cases T: 65K Y: 77K 7/11: 63K Newly reported tests T: 762K Y: 852K 7/11: 635K Positive test rate T: 8.6% Y: 9.1% 7/11: 9.9% — PolitiTweet.org

Nate Silver @NateSilver538

@gelliottmorris @jbview @Nate_Cohn @DanRosenheck @collprof @Michael_Cobb68 Or going further back (in our case to 1880) to build the fundamentals part of your model, in which case you'll encounter a lot more weird stuff (World Wars! Depressions! Panics!) that may be a better match for the conditions we face today. — PolitiTweet.org

Nate Silver @NateSilver538

@gelliottmorris @jbview @Nate_Cohn @DanRosenheck @collprof @Michael_Cobb68 It's not clear that you can't account for COVID empirically. We have some ways to account for it. But that notwithstanding, it might lead one to make more conservative choices in model-building overall, certainly including trying to leverage the entire data set... — PolitiTweet.org

Nate Silver @NateSilver538

@gelliottmorris @Nate_Cohn @DanRosenheck @collprof @Michael_Cobb68 I think 25:1 (96%) vs. 100:1 (99%) are pretty different forecasts. And the gap in percentage terms will increase if the election gets closer, e.g. maybe you'll have Biden at 95% to win the popular vote with a 6-point lead when he "should" be at 80%. — PolitiTweet.org

Nate Silver @NateSilver538

@gelliottmorris @DanRosenheck @Nate_Cohn @collprof @Michael_Cobb68 For example, it could be that polling gets more accurate over the course of say 1948-2012 but is now becoming less accurate again. That's also a plausible story that fits the data well but it has pretty different implications for 2020. — PolitiTweet.org

Nate Silver @NateSilver538

@gelliottmorris @DanRosenheck @Nate_Cohn @collprof @Michael_Cobb68 Notwithstanding that maximizing fit is not necessarily the best objective when n=18, there are some other issues with putting all your eggs in the polarization basket, among them that polarization ~linearly increases over the course of 1948-2016 and so is basically a time trend. — PolitiTweet.org

Nate Silver @NateSilver538

@gelliottmorris @DanRosenheck @Nate_Cohn @collprof @Michael_Cobb68 There's no such thing as out-of-sample if you know the results ahead of time. *Especially* not when n=18 and you're intimately familiar with the data and the entire data set influences the choices you consider for specifying your model. — PolitiTweet.org