Nate Silver

Nate Silver @NateSilver538

@joshtpm Right, that's the counter: that he pursued a high-ceiling, low-floor strategy, in which he was likely to wind up with a lot of 2nd/3rd place finishes, but relatively few wins. I mostly agree with that. But I also think many things about what Bernie did were very impressive. — PolitiTweet.org

Nate Silver @NateSilver538

FWIW, the White House's model—if this is actually a representation of it, this is from their press briefing—*also* seemed to assume a lot of symmetry. I'm worried that models that imply symmetry will overestimate how fast deaths will fall, even if they get the peak about right. https://t.co/Guf4nYmkSN — PolitiTweet.org

Nate Silver @NateSilver538

I think Bernie made a lot of mistakes, but one of the funny things is you get a lot more criticism if you're the 2nd-place candidate than if you're the 17th-place candidate, when it should be the other way around. Bernie did a lot of stuff that worked pretty well. — PolitiTweet.org

Nate Silver @NateSilver538

Indeed asymmetry is what you might expect given best guesses about how well social distancing works. Here are curves from Imperial College, for instance, which imagine turning social on & off. Note that they're somewhat asymmetric; cases don't decline quite as fast as they rose. https://t.co/JVFSAlhmCM — PolitiTweet.org

Nate Silver @NateSilver538

Instead, we're seeing signs of asymmetry. In New York, things do look like they're getting noticeably better, but they're not getting better as quickly as they're got worse. Same in Italy; there's a decline in deaths, but it's not particularly symmetric. Spain is more ambiguous. https://t.co/D6B0cGr8Pu — PolitiTweet.org

Nate Silver @NateSilver538

Thread. It looks like the IHME model is fond of drawing curves that are quite symmetric, i.e. they expect deaths to fall about as fast as they expect them to rise. There's no particular reason to expect this, empirically or theoretically, however. — PolitiTweet.org

Nate Silver @NateSilver538

@joshtpm Right, and it all gets very complicated because there are not just political but also epidemiological tipping points. — PolitiTweet.org

Nate Silver @NateSilver538

@jbarro We should start a Social But Also Sort of Antisocial Club for mediumverts. — PolitiTweet.org

Nate Silver @NateSilver538

Poll: — PolitiTweet.org

Nate Silver @NateSilver538

OK, which of these is closest to the mark for you? (Poll in next tweet.) https://t.co/fSJhMpraQY — PolitiTweet.org

Nate Silver @NateSilver538

The change in the rate of deaths has been pretty reliably lagging about a week behind the change in the number of newly-reported cases in the United States. In other words, whatever trend you see in the cases data tends to show up in the deaths data a week later. — PolitiTweet.org

Nate Silver @NateSilver538

@CarlBialik ^^^ Mostly talking about the models that look less rigorous here, often done by non-epidemiologists. The ones that IMO aspired to a higher standard, i.e. Imperial College, tended to be on the lower end of the range (i.e. 4-5% hospitalization rates). — PolitiTweet.org

Nate Silver @NateSilver538

@CarlBialik TBH it looks to me like *some* of the models just sort of plugged hospitalization rates from Wuhan/other early outbreaks without trying to account for underdetection, whereas conversely they were more thoughtful about underdetection when considering fatality rates. — PolitiTweet.org

Nate Silver @NateSilver538

@CarlBialik I'm not sure that can explain all of it, though. No doubt there was some triage in which people got admitted. But NYC also has a lot of temporary/overflow hospitals that never really got used or were very lightly used. — PolitiTweet.org

Nate Silver @NateSilver538

The number of new cases is down by around 15% from last Tuesday, however, with a similar number of tests conducted as last Tuesday. That's rather good* and *isn't* driven solely by New York, for what it's worth. * Though, we really need to be increasing testing. — PolitiTweet.org

Nate Silver @NateSilver538

I was worried about this report, because of likely Easter weekend backlogs in reporting; the last two Tuesdays have been quite bad. Indeed, the deaths number is a new daily high—but to be honest also about where you might have expected it to be given those issues. — PolitiTweet.org

Nate Silver @NateSilver538

Daily US numbers via @COVID19Tracking: Newly-reported deaths: Today: 2,299 Yesterday: 1,450 One week ago (4/7): 1,926 Newly-reported cases: T: 26K Y: 25K 4/7: 30K Newly-reported tests: T: 147K Y: 129K 4/7: 148K Share of tests positive: T: 18% Y: 19% 4/7: 21% — PolitiTweet.org

Nate Silver @NateSilver538

We won't really know the hospitalization rate until we know the number of undiagnosed cases. The lower end of that range might end up being pretty good though maybe not if there are a *lot* of undetected cases (say 50x rather than 10x). The upper end…I have questions about. — PolitiTweet.org

Nate Silver @NateSilver538

If I were looking for things to critique about the COVID-19 models, I'd be looking at their assumptions about hospitalization rates, which among other things seem to vary a lot from model to model. (As low as ~3% to as high as ~20% in some I've seen.) — PolitiTweet.org

Nate Silver @NateSilver538

A lot of the "trick" in building models is in figuring out which assumptions you can get mildly wrong and which ones you can't. — PolitiTweet.org

Nate Silver @NateSilver538

To put it another way, you should *not* necessarily expect a bell-shaped curve where the descent is as rapid as the ascent. Instead, the descent could take a while— even if social distancing measures are maintained. That's what we're seeing in Italy and now NYC. https://t.co/yI1XEQPM6l — PolitiTweet.org

Nate Silver @NateSilver538

But also: though the decline has been *steady*, it has also been pretty *slow*. It suggests there is not necessarily a ton of margin for error, and NYC/NYS will have to be rather careful when they eventually decide to start unwinding social distancing measures. https://t.co/rklKUVJZAt — PolitiTweet.org

Nate Silver @NateSilver538

Pretty robust decline in positive test rates across the different regions of the state, though less so upstate than downstate. These charts do some smoothing, but the numbers over the past 2 days have had particularly low (relatively speaking) rates of positive tests. https://t.co/gOXQvTPRWE — PolitiTweet.org

Nate Silver @NateSilver538

It increasingly looks like New York, both the state and the city, have not merely flattened but probably bent the COVID-19 curve downward. https://t.co/bMGWOzS8lC — PolitiTweet.org

Nate Silver @NateSilver538

RT @JamesSurowiecki: The fact that the president is comparing himself to Captain Bligh - and, for that matter, kind of is like Captain Blig… — PolitiTweet.org

Nate Silver @NateSilver538

@BenDWalsh @TheStalwart No, the trendline isn't awful at all. It's pretty good. New hospitalizations have fallen quite a bit. The share of people testing positive has also declined quite a bit. People stay in the hospital a long time, so it will take longer for the net number to decline. — PolitiTweet.org

Nate Silver @NateSilver538

Getting data on *when* and *where* (hospital, nursing home, at home) these deaths occurred would make interpretation and policy responses easier. Cuomo is now breaking out nursing home deaths in his press conferences; could use more of that. — PolitiTweet.org

Nate Silver @NateSilver538

Yep. For many reasons, we really want states to go back and find COVID-19 deaths they may have missed before. Even if that makes the numbers look superficially worse based on when those deaths are *reported*. — PolitiTweet.org

Nate Silver @NateSilver538

@TheStalwart Number of people coming into hospitals* without subtracting discharges (and deaths). * Or, perhaps more properly, getting a positive test for COVID-19 while in a hospital; I'm not 100% certain how this gets defined. — PolitiTweet.org

Nate Silver @NateSilver538

However, deaths are at a plateau and have not really begun to decline yet. At the same time, Tuesdays are often pretty bad days for reporting (clearing a weekend backlog) so this could have been worse, I suppose. https://t.co/a2OaZPX9ad — PolitiTweet.org