He’s pretty much right though. It spreads worse than the flu and Dr. Ioaniddis recently published a serological study showing it seems to have a mortality rate similar to the flu. An overwhelming majority of people who get it will not die. Ioaniddis et. al. suggests the evidence points towards 1~2 out of a thousand mortality rate.
We should not be panicking. We should be mitigating the disease based on the evidence at hand.
[citation needed] about cherry picking. The limits of their scheme was acknowledged and discussed in the paper. What cherry picking are you referring to that wasn’t addressed?
"I think the authors of the above-linked paper owe us all an apology. We wasted time and effort discussing this paper whose main selling point was some numbers that were essentially the product of a statistical error."
I'm skeptical whenever I see a teardown like this which fails to mention that the offical case counts have all the same problems. Maybe we should dismiss this paper - but that means committing ourselves to radical skepticism about the prevalence, not going back and believing the numbers printed in the news.
PCR tests are asymmetric. Positive result almost certainly means that this person has the virus, but negative result might mean a number of things. Bad swab (less than 3000 virus copies), temporary remission (Korea has at least 160 such cases by now), etc.
Test kit availability adds another layer. At some point New York had 200 confirmed cases, and two weeks later 400 deaths, yet CFR is unlikely to be as high as 200%.
So yes, all statistics should be taken with a grain of salt, but magnitude and direction of that grain may be different.
While Gelman does point out to issues that may invalidate this paper (test specificity, mostly, and noisy weights, potentially) there does not seem to be any "cherry picking" involved.
I would say even if the study is merely statistical error, using it to give an implausibly low estimate of the IFR that just happens to fit the agenda of one author qualifies as cherry picking. A priori you can not use a test with high false positive rate to do a study like this unless the prevalence is much higher than the false positive rate.
If I were to make an informed estimate based on the limited testing data we have, I would say that covid-19 will result in fewer than 40,000 deaths this season in the USA,
He also predicted about 10,000 deaths in total in his mid-March article -- and he even meant that number without any special measures like social distancing and WFH.
No, Imperial gave an estimation of what would happen without measures, but there were measures taken. So in their case, it's natural that the actual number (with measures) would be much less.
Ioannidis' already 4x surpassed prediction was 10K deaths without measures.
Given that we have 4x WITH measures, this means we would be many times more wrong if we followed his advice and didn't take any...
If he's off by 1.5x (IHME model predicts 60k fatalities attributed to COVID-19), surely that's miles better than the what - 50x predicted at the start (I recall seeing 2mil fatalities passed around by Imperial for the US)? 40k deaths is a little less than a week of natural deaths in the US, for scale.
It's worth putting his actual statement here, so people can decide whether you're misrepresenting him.
From his article [1]:
> If we assume that case fatality rate among individuals infected by SARS-CoV-2 is 0.3% in the general population — a mid-range guess from my Diamond Princess analysis — and that 1% of the U.S. population gets infected (about 3.3 million people), this would translate to about 10,000 deaths.
He ventured a prediction of the CFR. And then just calculates the amount of deaths from a pretty arbitrary infected percentage of the population (there's no mention of the time scale either). Never does he predict that 1% will be infected. He's arbitrarily picking a number to illustrate the amount of deaths we'd see, but that all depends on how the disease spreads. Oddly enough his serology study ends up being somewhat close - instead of 1%, they saw 1.80-3.17% (in Santa Clara).
But he uses exactly this in his argument, in the same paragraph: "If we had not known about a new virus out there, and had not checked individuals with PCR tests, the number of total deaths due to “influenza-like illness” would not seem unusual this year. At most, we might have casually noted that flu this season seems to be a bit worse than average."
Then later: "Some worry that the 68 deaths from Covid-19 in the U.S. as of March 16 will increase exponentially to 680, 6,800, 68,000, 680,000 … along with similar catastrophic patterns around the globe. Is that a realistic scenario, or bad science fiction?"
Then he claims that "The most valuable piece of information for answering those questions would be to know the current prevalence of the infection in a random sample of a population"
But I claim he already had, at the moment he wrote that article, much better data than that already available: specifically, that all the statistics everybody could find even in the Wikipedia already gave much more information that he claimed has to be obtained by "a random sample of a population."
One can evaluate "how random" all already known cases, at the time he wrote the article, were. But also one can evaluate, if these known cases, even if they weren't random, were actually saying more, not less, by the nature the numbers were obtained.
And that was exactly the case: time and again, in country after country, the statistics included much more people than the small randomness based study would include, and it gave reasonable estimates about both the speed of the spread and percentage of the people affected.
His argument was not based on analyzing already available data, but on "not knowing" by *refusing to even look at the already available data.
Which is fraudulent, ignorant or both. But there were some big names doing exactly the same, exactly at the time he published that article. So his article was just political, not scientific at all.
at what multiplier would you consider being skeptical about what he says.
your comparison of the 2 mil which was the worst case scenario months ago, and now irrelevant, with 40k which was his prediction from 10 days ago is wrong.
You have to read the study’s details, not just the number in the headline. The IC report’s highest number was looking at what would happen if strong countermeasures were not taken, and they subsequently were — it’s like criticizing the justifications for mandating seatbelts because so many fewer people die in car crashes now.
that's what I am trying to tell him. comparing his prediction a week ago with measures to a prediction a month or more ago without measure is pointless.
The Ioaniddis survey was a total sham. The true fatality rate will wind up being something like 1-2%, if counted by excess deaths.
A town in Italy, Castiglione d'Adda, had 1.4% of its population die in March. Normally 0.1% dies in a month. That town did a serological survey of blood donors, and 70% came up positive. That exact number won't apply everywhere because there are different age distributions, different comorbibitities, different genes, etc., but that is the ballpark we're looking at.
0.15% of New York City has already died. That should tell you how reliable Ioaniddis' numbers are.
We should not be panicking. We should be mitigating the disease based on the evidence at hand.