And/or the interventions have been working.

The model that was down revised already took into account the interventions

Got an article? CNN is saying almost exactly the opposite: https://www.cnn.com/2020/04/08/politics/deborah-birx-social-...

Birx continued, "is how important behavioral change is, and how amazing Americans are at adapting to and following through on these behavioral changes." "That's what's changing the rate of new cases, and that's what will change the rate of mortality going forward," she said.

Been following the guy from the Big Short, Michael Burry, who has been going deep in the data and models, and trying to raise awareness.

He tweeted today how Fauci and the mainstream media glossed over how the models already took into account the interventions:

https://twitter.com/michaeljburry/status/1248244146571116545

I can't find anything concrete there, but "already took into account" is a bit weaselly.

There had to be tons of uncertainty in a) how well people comply with the guidelines and b) how effective that compliance is. It certainly doesn't seem crazy to me more data would lead to revised values for those factors.

They could be more effective than originally expected. It could also just narrow the prediction interval--I'd bet that a lot of the "millions dead" stuff is reporting the high end, rather than the most likely outcome.

Yes, it seems it could either be "measures are more effective than thought" or "disease is not as severe as thought", or both.

Fwiw Michael Burry, someone who has earned the right to be listened to in the face of collective thinking, is strongly arguing for the "disease is not as severe as thought".

You might also be interested in this thread from Andy Slavitt, who ran CMS during the Obama administration.

https://twitter.com/ASlavitt/status/1248362891201392640

He makes the point that a) the new model has some optimistic assumptions, like no interstate travel and b) due to the exponential growth, a small uncertainty in R0 leads to wildly variable outcomes.

I guess I am uneasy with the "disease is not as severe" hypothesis because it's essentially unknowable. There's no objective measure of severity: it depends on knowing how to treat/prevent the disease and whether the necessary resources are available to do so.

> There's no objective measure of severity

While it varies with health care rationing, think CFR is still a useful measure of disease severity and allows us to situate COVID vs. the flu. Knowing the real CFR depends right now on how we calculate the denominator. Burry had analysis that 4% of the undiagnosed asymptomatic Danish population had COVID-19 per blood donation data, which suggests the CFR is 80x less severe than official figures in Denmark.

Anecdotally the disease entered countries weeks, arguably months before any cases were officially announced, making it likely that the denominator is an order of magnitude greater than official stats, and thus CFR an order of mag less severe.

Did you miss this around a week ago [0]? Birx said ~200K deaths if we did everything "perfectly." We were well into the lockdown at that point.

0: https://www.nbcnews.com/news/us-news/dr-deborah-birx-predict...

FWIW, she said "almost perfectly" and "up to 200,000 deaths." The revised model predicts up to ~127k deaths, which is certainly less, but not egregiously so (here: https://covid19.healthdata.org/united-states-of-america )

If you kept the model exactly the same, you'd nevertheless get tighter and tighter estimates (i.e., reduced uncertainty and a lower upper bound) as more data comes in. This is just how statistics works.

Moreover, we're presumably learning stuff as we go (e.g., putting patients prone seems to work better than on their backs), so the survival rate itself is (hopefully) not stationary.

> FWIW, she said "almost perfectly"

Has anyone seriously suggested that the US’s measures are being carried out anywhere near “almost perfectly”? Quite the opposite, there’s been lots of concerns voiced that people aren’t taking this seriously.

> The revised model predicts up to ~127k deaths, which is certainly less, but not egregiously so

It’s a nearly 40% reduction!

1) Are the policy implications of potentially killing off all of (say) Salt Lake City much different from destroying New Haven? I would argue no, not really.

2) Biological data is often a nightmare to work with. Estimates about behavior too. Getting something within an order of magnitude is often not too shabby.

3) Errors ('up to') are sensitive.

Here's a toy example. Suppose you think two numbers are each around 5, but the data are consistent with anywhere between 0-10. The sum of these numbers must be between 0-20 (low case: 0 + 0 = 0, high: 10 + 10 = 20), and their product between 0-100 (0 x 0 = 0; 10 x 10 = 100).

More data comes in and you can estimate each value more precisely: now you know they're somewhere between 4-6. You know the sum is actually between 8-12, and the product between 16-36. That's a massive decrease in the upper bound (64 percent for the product!) but literally nothing has changed except for the increased precision.

The COVID models have exactly this problem--none of the parameters are known exactly--and the outcome is some function of combining them. Moreoever, we're learning more about what factors matter AND how to fight the virus.

Any decent statistical model like this should include a confidence interval , if biological is so difficult then the CI would have reflected that . This just seems poor science to me

The 200k/127k people are harping about IS THE CONFIDENCE INTERVAL (well, the upper half of it, hence "up to").

That's half of my point--you'd expect the confidence interval to narrow with more data, regardless of what's going on. On top of that, you've got model error and non-stationarity (e.g., better care is discovered, driving the mortality rate down), which can't be reflected in the confidence interval.

Here's one of the modelling paper. The discussion of uncertainty starts on page 4: https://www.medrxiv.org/content/10.1101/2020.03.27.20043752v...