Nate Silver

Nate Silver @NateSilver538

Which means often reporting conclusions as provisional or "the best evidence available right now". It also means avoiding overly confident predictions and proclamations and avoiding the impulse to weave everything into a neatly-packaged narrative. — PolitiTweet.org

Nate Silver @NateSilver538

But it's a new disease! there's a lot we don't know! The experts are learning as they go! While there are some places where we should have relatively strong priors, in general experts need space to revise their conclusions and the media should be attentive to those revisions... — PolitiTweet.org

Nate Silver @NateSilver538

2. The next major category is not accounting enough for uncertainty. I actually think the experts have done a very good job of accounting for the major traits of COVID-19, especially if you (literally) read the fine print on what they were saying rather through the media filter. — PolitiTweet.org

Nate Silver @NateSilver538

There are a lot of other quirks in reporting. The data is proceeded by humans; it doesn't appear automatically. Lags are unavoidable, but some data series are more lagging than others. Many places have "lumpy" data, i.e. they report data in batches rather than in a steady flow. — PolitiTweet.org

Nate Silver @NateSilver538

e.g. experts estimate that only ~10% of coronavirus *infections* have become detected *cases* in the US. So accounting for how well we're doing at detection (e.g. test volume) is very important. Deaths are also significantly underreported, more so in some contexts than others. — PolitiTweet.org

Nate Silver @NateSilver538

1. I think there's not nearly enough recognition in the media that the data we have on coronavirus is highly imperfect and this sometimes leads to misleading conclusions. — PolitiTweet.org

Nate Silver @NateSilver538

So, here's a podcast with @brianstelter where I talk about media coverage of coronavirus data. I think media coverage of coronavirus overall has been good in many respects, but there are basically three things that frustrate me from a data point of view. — PolitiTweet.org

Nate Silver @NateSilver538

@gelliottmorris Right and other smart people have worked on the problem and come to different conclusions. It is intrinsically a difficult problem to solve without more information on the types of people being tested, and I'm not persuaded you've found a better method than say rt-live — PolitiTweet.org

Nate Silver @NateSilver538

@gelliottmorris I don't know what you mean by "even if you control for testing & seasonality". The positive test % has slowly but steadily *declined* in Texas. That's important evidence, though not dispositive, but much better than looking at the raw # of cases when tests have rapidly increased. — PolitiTweet.org

Nate Silver @NateSilver538

@gelliottmorris It's not at all clear that the later claims are wrong. Many analyses suggest that R<=1 in Texas, which means the reported increase in cases is in fact due to testing. https://t.co/oAQ2y08bJ5 — PolitiTweet.org

Nate Silver @NateSilver538

@gelliottmorris That's moving the goalposts a lot, though. You get media segments about cases spiking in Texas. Instead it's a question of a slight decline, a plateau or perhaps a slight increase when you look at the data more robustly. — PolitiTweet.org

Nate Silver @NateSilver538

@gelliottmorris @COVID19Tracking Texas only tracks hospitalizations for people who have a positive COVID test. So expanding testing will probably increase their reported hospitalizations as well. @SeanTrende — PolitiTweet.org

Nate Silver @NateSilver538

For various reasons I don't think the media has updated its priors quickly enough on the much improved availability of testing in the US, and that may lead people to not seek tests when they'd be relatively easy to obtain in many places. — PolitiTweet.org

Nate Silver @NateSilver538

Cuomo says New York now has *unused testing capacity* and that people with symptoms or in contact with COVID+ people should get tested if they can. — PolitiTweet.org

Nate Silver @NateSilver538

Quick explanation from me here about why death counts are likely an undercount, not an overcount. — PolitiTweet.org

Nate Silver @NateSilver538

RT @ABCPolitics: .@Natesilver538 beaks down why experts say that coronavirus deaths in the U.S. have been undercounted. https://t.co/4T9uJi… — PolitiTweet.org

Nate Silver @NateSilver538

RT @cmyeaton: Let’s not give too much air time to the minority of people who don’t want to wear masks or social distance. Most people are d… — PolitiTweet.org

Nate Silver @NateSilver538

@PatrickRuffini They're basically just running a big regression analysis, so their research is somewhat agnostic about how much of this is due to weather per se as compared to indoor vs. outdoor activity. — PolitiTweet.org

Nate Silver @NateSilver538

With numbers like these, as they emphasize, weather is unlikely to stop COVID transmission all on its own in the US. However, it probably does have *meaningful enough* effects that they may be worth accounting for when planning policy and when interpreting data. — PolitiTweet.org

Nate Silver @NateSilver538

Interesting attempt to get specific about the effects that weather has on COVID transmission. They estimate that for the US as a whole, it transmits ~25% less efficiently at the peak of summer than in the winter but this varies a *lot* from place to place. https://t.co/jxkDIlvgIY — PolitiTweet.org

Nate Silver @NateSilver538

RT @brianstelter: My guest on this week's Reliable podcast: @NateSilver538. On Covid-19 data, "known unknowns," and the need for humility >… — PolitiTweet.org

Nate Silver @NateSilver538

I still wouldn't put too much emphasis about any numbers reported on a Saturday. But, this is now 8 straight days with a lower death toll than what was reported than the same day a week earlier. So, that's a little something. Have a good Saturday night, folks. — PolitiTweet.org

Nate Silver @NateSilver538

Cases and testing data about the same as in recent days. But the deaths numbers are on the lower end of what we've seen recently. Saturdays are typically an average day for reporting, not as bad as Tuesdays and Wednesdays but not as slow as Sundays and Mondays. — PolitiTweet.org

Nate Silver @NateSilver538

US daily numbers via @COVID19Tracking: Newly reported deaths Today: 1,286 Yesterday: 1,645 One week ago (5/9): 1,529 Newly reported cases T: 25K Y: 26K 5/9: 25K Newly reported tests T: 357K Y: 378K 5/9: 296K Positive test rate T: 7% Y: 7% 5/9: 8% — PolitiTweet.org

Nate Silver @NateSilver538

@mattyglesias It is sobering but the question wording requiring 25+m doses is important. Keep in mind that there are various good and bad ways that COVID-19 could be "solved" through mechanisms other than a vaccine, and hence there might never be all that much demand for a vaccine. — PolitiTweet.org

Nate Silver @NateSilver538

@thecity2 A lot of Asian countries seem to be beating it thru contact-tracing + testing + mask-wearing + general awareness and without a ton of social distancing. So I don't think it's crazy once you get the numbers low enough to make contact tracing feasible. — PolitiTweet.org

Nate Silver @NateSilver538

FWIW, generally I prefer the @COVID19Tracking way but making a mental note that there are meaningful lags in reporting. — PolitiTweet.org

Nate Silver @NateSilver538

The CDC way makes the curve look like it's sloping down faster than it actually is because some deaths that have occurred recently haven't been recorded yet. But the other way to some degree has the opposite problem and can make the peak look later and flatter than it really is. — PolitiTweet.org

Nate Silver @NateSilver538

The problem is that, because there can be a fairly long lag between when a death occurs and when it gets recorded, both ways paint a somewhat incomplete picture. And there's probably not a great third way/middle ground unless you want to do a lot of fancy modeling/estimating. — PolitiTweet.org

Nate Silver @NateSilver538

While this may not be the only issue, one major difference is that @COVID19Tracking is showing deaths when they were *reported* whereas the CDC is showing deaths when they actually *occurred*. — PolitiTweet.org