Nate Silver

Nate Silver @NateSilver538

Over the past 2 weeks, the US has moved up from 17th to 14th on the Worldometers list of COVID-19 deaths per capita, having overtaken Luxembourg, Iran and Saint Martin. But overall, the relative position of the US v. other countries has been fairly stable. https://t.co/5plSdXJbMK — PolitiTweet.org

Nate Silver @NateSilver538

e.g. You might have a paper that *assumes* from other research that the IFR (infection fatality rate) is 0.5% to estimate some other interesting quantities. But then people will point to the paper and be like "see, this clever paper shows the IFR is 0.5%." No! It just assumes it. — PolitiTweet.org

Nate Silver @NateSilver538

There are lots of papers where researchers make assumptions about some parameters (e.g. the fatality rate) & have "hard"/novel data on other parameters and then blend those in some way to infer some third set of parameters. People on here need to be careful about which is which. — PolitiTweet.org

Nate Silver @NateSilver538

(Too-rare-these-days sports tweet:) Working on RAPTOR last year convinced me that Jordan > LeBron. — PolitiTweet.org

Nate Silver @NateSilver538

RT @ameliatd: Some forecasters are predicting a fast economic recovery later this year. Here's the thing, though: research shows that econo… — PolitiTweet.org

Nate Silver @NateSilver538

Overall though, this data puts some boundaries on what the real death counts are in the US and Western Europe. Unless places are explicitly trying to count probable deaths (e.g. NYC), there is clearly likely to be an undercount. But it is not likely, say, a 3x or 5x undercount. — PolitiTweet.org

Nate Silver @NateSilver538

There could be much more underreporting in poorer places with less access to medical care and worse record-keeping systems, by contrast, or places that are in some form of denial about the extent of their outbreak. — PolitiTweet.org

Nate Silver @NateSilver538

I'd be especially careful about making inferences outside of that narrow band of countries/places. There is possibly less undercounting in wealthy Western places with *mild* outbreaks (e.g. Canada, Utah) because their medical and record-keeping systems are less overwhelmed. — PolitiTweet.org

Nate Silver @NateSilver538

Also, all of those qualifiers I put at the top are important: "wealthy Western countries with bad outbreaks of COVID-19". That's where these numbers, at least the ones that are getting highlighted in the media, mostly come from. — PolitiTweet.org

Nate Silver @NateSilver538

Plus, jurisdictions can vary in how thorough they are. France has done an admirable job of trying to count nursing home deaths, for instance. And if a place *is* trying to count probable deaths, as in NYC, there's going to be considerably less undercounting. — PolitiTweet.org

Nate Silver @NateSilver538

Still, inferring from excess mortality data can be tricky. Being in lockdown with heavy strain on hospitals can have various effects that push other types of deaths up (e.g. you have a heart attack, the ambulance takes longer to come & you die) and down (fewer traffic accidents). — PolitiTweet.org

Nate Silver @NateSilver538

For instance, NYC *is* trying to count *probable* COVID-19 deaths, and their total (confirmed + probable) death count is about 1.5x as high as their confirmed count. https://t.co/FzTtPAVIGR — PolitiTweet.org

Nate Silver @NateSilver538

Does that mean you can just multiply confirmed COVID-19 deaths by 1.5 to get the "real" death count? Well, in some cases, it might not be a terrible approach. But you'd have to be careful. — PolitiTweet.org

Nate Silver @NateSilver538

It seems like in wealthy Western countries with bad outbreaks of COVID-19, total excess mortality is about 1.5x higher than confirmed COVID-19 deaths. Quick thread on how I think about this. https://t.co/ybmRpb7wRQ — PolitiTweet.org

Nate Silver @NateSilver538

It's interesting to read this as a time-capsule one month later, as it's a pretty good summary of highbrow discourse around COVID-19 at that time. Some things hold up quite well, others not as well. https://t.co/m1IhFoVcMQ — PolitiTweet.org

Nate Silver @NateSilver538

@kevin For sure but even so, hard to claim there's a downturn. — PolitiTweet.org

Nate Silver @NateSilver538

A lot of the context re: Georgia is that they *haven't* had a sustained decrease in cases. It's been noisy, and in fact, today was the 2nd-highest day for new cases. They are doing a bit more testing than before. But even with that... it has been noisy. https://t.co/qY4tzmbMAE https://t.co/me1th1V7SX — PolitiTweet.org

Nate Silver @NateSilver538

Sure, if you flipped a switch overnight, that would be one thing. But opening up a limited segment of businesses that are fairly strictly adhering to social distancing protocols while many people stay home anyway... the data will probably not be precise enough to tease that out. — PolitiTweet.org

Nate Silver @NateSilver538

I'd be careful about trying to dunk on states that may be lifting certain restrictions, because the impact, although not trivial, may get lost amidst the noise/uncertainty, and you're setting yourself up for "See, State X tried *this* and it wasn't the disaster people claimed." — PolitiTweet.org

Nate Silver @NateSilver538

RT @538politics: New podcast! 1. Anti-quarantine protests 2. Veepstakes! https://t.co/QREoEG56Kr — PolitiTweet.org

Nate Silver @NateSilver538

No super pithy, definitive conclusion. Just think this sort of data may be worth looking at in a less quick-and-dirty fashion than I did here. — PolitiTweet.org

Nate Silver @NateSilver538

In New York City, the ratio is 0.26 using confirmed deaths, although it jumps up to 0.39 if you use the city's count of *probable* deaths. It's not obvious to me which count to prefer here (how are other states/cities handling cases that NYC would consider probable deaths?). — PolitiTweet.org

Nate Silver @NateSilver538

Nationally this ratio is 0.19 in the average state to have recently updated its cumulative hospitalization data, with a standard deviation of +/- 0.06. New York state is at 0.25 which is one SD above the mean, so it's definitely on the high side but not some crazy outlier. — PolitiTweet.org

Nate Silver @NateSilver538

A worthwhile question is whether NYC had higher fatality rates than it otherwise would because of overwhelmed hospitals. One indication could be if the ratio of deaths to hospitalizations is high, which might indicate i) people avoiding hospitals or ii) not getting adequate care. — PolitiTweet.org

Nate Silver @NateSilver538

@asymmetricinfo The rate actually has begun to decelerate pretty quickly in NYC and the dynamics of what you'd expect to see happen given various infection rates are fairly complicated given different R0s among different subgroups, as you mention. — PolitiTweet.org

Nate Silver @NateSilver538

@asymmetricinfo Did 50% of NYC get COVID at some point? I'd bet against it, but it's not crazy to think the numbers could be pretty high. — PolitiTweet.org

Nate Silver @NateSilver538

@asymmetricinfo People getting tested is a biased sample, but FWIW NYC's tests were ~60%+ at the peak. Also, these tests can have high-ish false negatives. And, people who get the disease will only test positive for a few weeks whereas antibody tests will pick up on whether they ever had it. — PolitiTweet.org

Nate Silver @NateSilver538

@R_Thaler @dmorey I tend to agree that the behavior is maybe 70-90% voluntary. But I wonder if the signals sent by state and local officials matter also, though. Also wonder what happens when the fear tends to subside because people see & hear less about rising case counts. — PolitiTweet.org

Nate Silver @NateSilver538

@kmedved @CountingBaskets I think there's a question because of how many excess deaths there were in NYC because of the crash on hospitals, which may not have reached Lombardy levels but could still lead to less-than-ideal levels of care. — PolitiTweet.org

Nate Silver @NateSilver538

US daily numbers via @COVID19Tracking: Newly-reported deaths: Today: 1,528 Yesterday: 1,654 One week ago (4/13): 1,450 Newly-reported cases: T: 23K Y: 27K 4/13: 25K Newly-reported tests: T: 138K Y: 167K 4/13: 129K Share of tests positive: T: 17% Y: 16% 4/13: 19% — PolitiTweet.org