Nate Silver

Nate Silver @NateSilver538

A 3-week national lockdown to (i) contain discovered *and undiscovered* simmering hotspots and (ii) get testing volume up to speed everywhere ... and *then* applying recommendations on a county-by-county basis *after* those 3 weeks... that would be a much better policy. — PolitiTweet.org

Nate Silver @NateSilver538

So the places that look good now may look bad in 2-3 weeks *especially* if they're deemed low-risk and interventions are relaxed. (To some extent, the reverse may be true also and places that were bad before may be bending the curve.) This thing changes *fast*. — PolitiTweet.org

Nate Silver @NateSilver538

Another problem is that counties that have a lot of cases *today* may be on the flat part or even the downslope of their curve, while counties with relatively few cases today may be on the sharpest part of exponential increase. — PolitiTweet.org

Nate Silver @NateSilver538

Some of the most dangerous hotspots in the US right now are probably places where there are few reported positives because there hasn't been much testing ... and the lack of reported positives also hasn't compelled much change in behavior. — PolitiTweet.org

Nate Silver @NateSilver538

To categorize counties as high-, medium- and low-risk, you need an adequate amount of testing everywhere. The US testing situation has improved, but it very much varies from location to location. — PolitiTweet.org

Nate Silver @NateSilver538

One thing I don't know about the Italian data, FWIW, is whether they're reporting the number of tests *as they're conducted* (meaning that many results may still be pending) or only once results are returned. — PolitiTweet.org

Nate Silver @NateSilver538

Not fantastic. But Italy reported a large volume of new tests, 37k, which is considerably larger that the number of tests they usually report (in the low-mid 20k range). https://t.co/3mA6RYQQON — PolitiTweet.org

Nate Silver @NateSilver538

@R_Thaler @PTetlock I agree that the ranges for number of deaths seems too narrow. But... if more cases are unreported it also means that the true case fatality rate is lower. So measures such as deaths and hospitalizations may not be that sensitive to underreporting. — PolitiTweet.org

Nate Silver @NateSilver538

The experts we surveyed also think that under-reporting could be anywhere from 3x (that is, 3 actual cases for every reported case) to 50x (that is, 50 actual cases for every reported case) in the US. That's a big range, and one of the most important unknowns right now. — PolitiTweet.org

Nate Silver @NateSilver538

Experts expect 117K *reported* cases of coronavirus in the US by Sunday. They also think only about 10% of cases are actually being reported, though, so that would mean more than 1M actual cases. https://t.co/oMWfVdQVVd — PolitiTweet.org

Nate Silver @NateSilver538

Basically *none* of the coronavirus data is being reported *exactly* in real time. There are lags, bottlenecks, etc., in reporting, and sometimes a good day one day means there's been a bottleneck which will yield a *worse* day the next day. — PolitiTweet.org

Nate Silver @NateSilver538

Big increase in hospitalizations in NY state today, per Cuomo's presser, to 5,327, after slower growth for the past two days. My advice is to wait for 3-4 days' worth of data before declaring a trend, especially a downward trend. — PolitiTweet.org

Nate Silver @NateSilver538

There is also likely significant underdetection in Trump states. In Trump states that report on hospitalizations, 20% of detected cases are resulting in hospitalizations, compared to 12% in Clinton states (10% without NY). That probably means a lot of mild cases are being missed. — PolitiTweet.org

Nate Silver @NateSilver538

If detected cases in each set of states continue to increase at their current rates, the number of cases in Trump states would surpass the number of cases in Clinton states on 4/11... the day before Easter. — PolitiTweet.org

Nate Silver @NateSilver538

Mississippi hospitalizations for COVID-19 have tripled in the last three days. https://t.co/crzp9Jf7dQ — PolitiTweet.org

Nate Silver @NateSilver538

Here's another interesting comparison. Yesterday, detected cases increased by 31% in Trump states as compared to 21% in Clinton states. — PolitiTweet.org

Nate Silver @NateSilver538

A 3-week national lockdown would probably do more good than a 6-week lockdown in some places and a 0-week lockdown in others, but we're probably going to get the latter and not the former. — PolitiTweet.org

Nate Silver @NateSilver538

These might not *seem* like big differences. But multiplied over days/weeks, they matter a lot: 30% = daily growth is quite bad; steepest part of exponential curve 20% = slope nudged downward but still exponential growth 10% = rather encouraging, may mean Ro<1 and spread slowing — PolitiTweet.org

Nate Silver @NateSilver538

Increase over yesterday in detected coronavirus cases. Hopefully, grouping things by region helps even out disparities in testing rates. South: +32% Northeast (excluding NY): +30% Midwest: +27% West (excluding CA & WA): +23% New York: +20% California: +20% Washington: +11% — PolitiTweet.org

Nate Silver @NateSilver538

Louisiana now with 14.3 deaths per million residents, which is comparable to New York (14.6 per million). And the number of hospitalizations is rising very rapidly there. — PolitiTweet.org

Nate Silver @NateSilver538

But on a log scale the countries barely look any different, even though A's intervention is hugely important and already reduced cases by 50%. Don't assume the log scale is always the way to visualize this data; sometimes it is, but sometimes it obscures the most important info. https://t.co/ob7w2UX0Qi — PolitiTweet.org

Nate Silver @NateSilver538

Say you have two countries that each have cases growing at 30% per day. However, on Day 10 of the outbreak, country A undertakes an intervention where growth slows to 10% per day. This makes a huge difference, even after a few days—and you can see on a chart with a linear axis. https://t.co/EnJSMLNlX7 — PolitiTweet.org

Nate Silver @NateSilver538

I have a 🔥 take about how charts showing the cumulative number of coronavirus cases on a logarithmic scale are actually pretty easy to misinterpret. This is the quick half-assed version of it because I'm too lazy right now to make nice-looking graphs. — PolitiTweet.org

Nate Silver @NateSilver538

This is key. The federal, state and local governments all need to be thinking about what the post-lockdown phase looks like and not just circling a date on the calendar and hoping for the best. — PolitiTweet.org

Nate Silver @NateSilver538

The rest of Italy has seen a 9% increase over the same periods, by contrast. https://t.co/3mA6RYQQON — PolitiTweet.org

Nate Silver @NateSilver538

The slowdown in new cases in Italy has been concentrated in Lombardy, the worst-affected region. Lombardy has seen a 27% drop in new cases over the previous 4 days (3/22 thru 3/25), as compared to the 4 days before that (3/18 thru 3/21). — PolitiTweet.org

Nate Silver @NateSilver538

We still need a lot more testing capacity. A lot more. But even countries like South Korea have tested only ~1% of… https://t.co/l8fkeZhBWt — PolitiTweet.org

Nate Silver @NateSilver538

The US has conducted about 400k tests now. We are ramping up to speed, quickly. This has been a positive developmen… https://t.co/T2B9MhYlPy — PolitiTweet.org

Nate Silver @NateSilver538

The NYC data is decently close to the overall age distribution in the city, although with few cases among people <1… https://t.co/HHGxEuZLL5 — PolitiTweet.org

Nate Silver @NateSilver538

There a lot of caveats to how well econometric models can handle coronavirus. But there are also some objections th… https://t.co/LQlJDx7S13 — PolitiTweet.org