Your friend Jim George thinks you'd be a great addition to Ricochet, so we'd like to offer you a special deal: You can become a member for no initial charge for one month!
Ricochet is a community of like-minded people who enjoy writing about and discussing politics (usually of the center-right nature), culture, sports, history, and just about every other topic under the sun in a fully moderated environment. We’re so sure you’ll like Ricochet, we’ll let you join and get your first month for free. Kick the tires: read the always eclectic member feed, write some posts, join discussions, participate in a live chat or two, and listen to a few of our over 50 (free) podcasts on every conceivable topic, hosted by some of the biggest names on the right, for 30 days on us. We’re confident you’re gonna love it.
I did a post on the same thing. Sorry to cross signals. Your title is better than mine, and as a Star Wars fan, I’m a bit sheepish that I didn’t think of it. :)
I think that you’re mostly right about this. A modified “really bad bug/less bad bug” theory might explain the lower IFR estimate for the Bay Area from this study, compared to the probably higher IFR in NYC. There’s not much of an age difference between Santa Clara County and NYC (this is covered in my post). It’s possible that NYC is just about through this, if the 50-85x figure applies there, though the higher number can’t possibly apply because I think that it would imply over 100% infection in NYC at present.
I really hope that someone is doing a similar study in NYC.
Nah. They were both great posts.
Jerry’s is better.
I write like a scientist: long and winding, exploring every thought that comes to mind.
Jerry writes like somebody who actually has to convince other people in order to get paid: short and to the point.
Okay.
They were both great posts.
Yes. As you seem to indicate, a number of demographic factors in Santa Clara County–not just age–can influence fatalities (downward). Absent further study, I don’t see this as a game changer for purposes of influencing policy across the country.
Thanks for your effort on this.
I see something interesting. You tell me what it might mean, gentlemen. How does the population of Facebook users differ from Facebook non- users? No non- users were subjects in the study.
I giggled… and then I couldn’t stop laughing.
could stress be a factor? Is NYC more stressed than Santa Clara?
I didn’t notice that but I think there is a world of difference between those populations. Personally, I suspect that all Facebook users are doomed. Or at least should be.
Sources of potential bias/concrn:
1. What percentage of Santa Clara County residents use Facebook? A sample of a sample.
2. In the paper there did not seem to be any consideration of age. Could this matter? I don’t know.
3. The denominator is really small which makes the 50-85 number really big.
4. There’s no indication of false positives rates that I saw in the paper.
I’d appreciate your thoughts on this critique of the paper. Since you actually know something about this topic, unlike me, you can evaluate its criticisms.
I think the virus was in the central valley by the end of Dec. first of Jan. We have a lot of contact here with China due to the ag. industry through consulting and “something was going around”. My wife and I had several bad weeks then and what I had was unlike any cold or flu I have previously had. We didn’t think that we had Wuhan flu at the time because we really hadn’t heard about it. I didn’t see a doctor but my wife had to go to ER because she was having a hard time breathing one night. This was before any cases here so no flu test was done. Reading about the effects sure points to that virus and the Stanford testing adds. We are trying to find out how one of us could get the antibody test to be sure.
I’ll post this on the other post too.
So, since this paper was released yesterday a number of well-respected and well-credentialed experts have voiced some doubts about the confidence we can place in these results. Bearing in mind I’m not a statistics guy, I’ll try my best to address the criticisms.
One thing to point out in advance, though, is that all of the serious critiques by highly-visible experts have been quite polite and respectful. Everybody seems to agree that there is likely a substantial percentage of undetected cases, everyone agrees that it’s good the study was carried out, and everybody seems to respect the fact that the authors laid out all of their raw data and showed their math (more than can be said for some other similar studies *cough* Heinsberg *cough).
I mention this because with all of the passion and finger-pointing surrounding the science of Covid-19, it’s refreshing to see people with quite opposite scientific views capable of addressing each other calmly and seriously.
With that in mind, let’s go through the individual points:
1. Recruitment by Facebook:
The authors placed ads targeting several very different communities in Santa Clara County (with the help of Facebook itself) in order to get a diverse set of volunteers. The ads were supposedly one-time ads, meaning they couldn’t be forwarded onto others.
Even with those precautions, it seems intuitive that people who thought they might have been infected but who were denied access to a PCR test for capacity reasons would have been highly motivated to sign up. The fact that all 3,300 slots were filled within a few hours supports the notion that many of the participants were very motivated.
While the “clean” way to perform a study like this is to randomly invite volunteers (i.e. using the phone book), everyone recognizes that such an approach would have taken much, much longer. The authors acknowledged this potential source of bias in their paper, and this seems like a factor that even most laypeople can easily judge for themselves whether or not it played a major role.
2. False positives:
This is the most pertinent criticism in my view. When the overall number of positive “hits” in a test like this is low (as it was in this case: only 50 out of 3,300 tests were positive, which is a crude rate of 1.5%), even a low false-positive rate can play a huge role. For example, many rapid antibody tests are known to have false-positive rates of about 1.5%. With a test like that, the authors’ 1.5% positive rate could theoretically be entirely explained by false positives, i.e. a true positive rate of 0%. So small differences in false positive rate matter a lot.
The authors determined their test had a specificity of 99.5%, meaning a false positive rate of only 0.5%. This is quite low for tests of this manner. This rate is based on 2 positives among 400 negative samples tested. However, that number of 2 is low enough to make the “99.5%” claim questionable – not in its accuracy, but in its reliability: there’s a good chance the actual false-positive rate could be somewhat higher of lower. In this case, “somewhat higher” would be enough to completely obviate their top-line results.
The authors acknowledge this issue by providing 95% confidence intervals – and indeed, these intervals are so broad that “0 cases in Santa Clara County” is among the possible interpretations of their results.
Bottom line: this issue is big enough that it doesn’t negate their results, but does mean that policy makers would probably require at least one more confirmatory study before taking action.
3. Demographic extrapolation:
Because Bendavid and Bhattacharya’s study population was self-selecting and thus not inherently representative of the residents of Santa Clara County, they had to normalize their results onto the known demographics. This by itself is of course standard and proper.
There are two issues that have been raised though: first, the fact that their final figures are so much higher than their crude figures (i.e. 1.5% => 2.5-4.2%, i.e. almost 3-fold greater), especially since the crude figure is so small to begin with. Basically, any time somebody uses statistics to transform a fairly low rate into a much higher one, eyebrows should be raised. This is the basic concern with almost all political polling and is a general criticism, not one specific to the methodology used for demographic normalization in this study.
The second issue is raised here. I’m nowhere near competent enough to judge this criticism, but the jist is that the crude rate of 1.5% needed to undergo two transformations: one based on the error rate of the test, and a second based on the demographics of Santa Clara county. The argument is that the order in which these two transformations are performed makes a huge difference on the results. In a nutshell, had the authors performed the two transformations in the other order, their results would have been much lower.
Apparently, there’s no clear-cut standard for which order is the better one, nor do the study authors explain why they chose their approach. Bottom line: there seems to be a valid statistical approach that arrives at a much lower figure even assuming the same test specificity. Which approach is more justified is well beyond my paygrade.
4. Fatality rates in affected areas:
As I mentioned in the original post, the infection fatality rate (IFR) of 0.12-0.2% proposed by Bendavid and Bhattacharya doesn’t completely square with the fact that places like NYC and Lombardy, Italy have seen absolute fatality rates of over 0.1% in the course of a month. In other words, in the areas most heavily affected, not just 0.1% of SARS-CoV-2 infected patients have died, but over 0.1% of the entire population has died. Since the maximum percentage of the population that can be infected is well below 100%, these absolute fatality rates point to an IFR considerably higher than 0.12-0.2%.
This real-world sniff test is an important sign that the IFRs derived by Bendavid et al. should be considered somewhat speculative. However, there’s an equally important lesson: for all of the navel-gazing surrounding the mythical “IFR” of SARS-CoV-2, we need to remember that IFR is not some figure set in stone. We know for a fact that different demographic groups have wildly different fatality rates. We know for a fact that different regions define Covid-19 deaths differently. We can presume that fatality rates will vary based on how overwhelmed a local healthcare system is.
We can also presume (based on other viral diseases) that more subtle factors like infectious dose and direct route of transmission may play a major role in outcome. In other words, getting the virus by rubbing your nose after touching an infected doorknob might lead to a different course of disease than sharing a bedroom with somebody constantly coughing the disease straight down your throat.
So the differences in real-world fatality rates should make us skeptical about Bendavid and Bhattacharya’s conclusion, but are not a show-stopper in my opinion.
Some people think the default policy should be freedom, which we might need to sacrifice in the face of overwhelming evidence.
Some people think the default policy should be hard lockdown, which we might loosen in the face of overwhelming evidence.
Some people think the default policy should be the status quo, which we might change in the face of overwhelming evidence.
All well and good.
In the episode of uncommon knowledge released yesterday, Dr. Bhattacharya mentioned 2 other studies. 1 submitted for publication and 1 he was writing this weekend. I think his tone would have been much different if those studies substantially contradicted the first one.
“considerably higher”?
Can you quantify what you mean by “considerably higher”?
The maximum percent is well below 100%… how much below 100%?
I’m glad to hear the maximum percent infected is under 100%.
Is that from the Imperial College model?
I don’t think there’s any correlation in a scientist’s certainty in his own conclusions and the probability of those conclusions being true, so I don’t find that particularly relevant.
In any case, I’m very happy for Dr. Bhattacharya’s work on Covid-19, I just don’t see it being persuasive enough to guide much public policy in the absence of more solid corroborating evidence (by more solid I mean more/larger/better controlled examples than pregnant women at Columbia Hospital or people in a homeless shelter).
My summary of Bhattacharya’s work on Covid-19 so far would be to say that he’s presented hypotheses/conclusions that are certainly plausible but do not inspire a high degree of confidence (in the scientific, not general, sense of the word). The reason is because they tend to postulate a scenario that runs counter to fairly large evidence without explaining the discrepancy between his conclusions and the conflicting data. That’s not a deal breaker, because almost none of the science so far earns a high degree of confidence, and there are certainly many scientists postulating an excessively high level of danger based on even less solid evidence.
So I see the main value of his work in pointing out that we need to keep an open mind about this virus, not that it is obviously and self-apparently only slightly more harmful than a seasonal flu.
By now we’ve probably all seen the general equation for determining herd immunity based on R0. Even though we don’t know R0 here with certainty, it seems like it will be somewhere between 60-75%. So if we view that as the upper bound of what percentage of any group can become infected, then a 0.12% absolute mortality rate would imply about a 0.16-0.2% infection fatality rate if every single potential susceptible person was infected.
I’m intentionally ignoring your favorite 20% statistic from the Diamond Princess because a) we currently have no way of knowing how many people would eventually become infected had they not put the guests under strict lockdown, and b) the French aircraft carrier has already surpassed its rate of infection.
I am hopeful that a few things together may change the scene. Showing a relatively low IFR is one, ramping up treatment options if their trial results actually are successful is another.
What I dislike is that some public figures seem to respond to (even cautious) good news with even stricter restrictions on liberty. This stimulates my libertarian skepticism and makes me think of potential ulterior motives and a reluctance to give up newfound power over a newly compliant public.
I thought empty hospitals would make a difference (at least regionally/locally), since the original goal was to not overwhelm the healthcare system.
But that goalpost definitely moved.
Very much. Our first lockdown was for 2 weeks. That was nonsense at the time. The reason was to “flatten the curve,” and that seems to have gone away completely. Not sure what the rationale is at this point, except that we need to get to zero deaths and no danger… I don’t think that makes much sense. It may well be that it doesn’t make any better sense to go back to business as normal, but quite frankly, nothing is making all that much sense right now.
I haven’t seen a single headline noting any flattening of the curves here or there. But there have been headlines about the death toll being higher than ever.
I wonder what they’ll do when the count of total deaths (since the beginning of the crisis) starts going down. They’ll probably run headlines complaining that not all the dead have yet been resurrected.
Two questions occurred to me today. I’ve figured out the answer to one, but was wondering if anyone knew about the other.
The first was whether flu infections are asymptomatic like COVID? The answer, acccording to a 2014 article in The Lancet is yes, and like COVID the number is substantial.
The second is whether the CDC includes an estimate for asymptomatic flu cases in its annual model of flu cases? I haven’t been able to find an answer to that.
I’m trying to figure out if we are making apples to apples comparison regarding infection and CFR between the flu and COVID.
So, this has been happening in my neck of the woods:
https://www.bostonglobe.com/2020/04/17/business/nearly-third-200-blood-samples-taken-chelsea-show-exposure-coronavirus/
(And yes it’s The Boston Globe, but the article does manage to do a competent job and keep the progressive axe-grinding to a minimum.)
This thing is widespread, and grinding along, and it’s premature at best to chalk that up as a positive.
As I’ve discussed elsewhere, I fled from my Tokyo abode back to the US in late February — had been hoping I would be able to take refuge in Fortress America while looking after my parents. I’ve been able to do the latter — the former, well, we know how that’s gone.
I was prompted to flee the Northeast Asia neighborhood because it was the easiest thing in the world to discern that what we’re confronting — even if naturally-occurring and not tweaked further by human hand — nevertheless was something obtained/collected, analyzed, and then selected for further work/attention in the lab precisely for its hyper-transmissibility and lethality attributes. It was meant for deployment somewhere. (My contention is that it was intended for a certain non-Han-Chinese population in a certain northwestern province in the PRC.)
Of course, subsequent to that selection by the lab management, someone on the frontline research teams goofed and commuted home from the lab with the virus on her (apparently the intel suggests a “her”).
And eventually the CCP regime then had to deliberately start implementing strictures not only in the epicenter of the leak but nationwide — even at risk of trashing their own GDP. And the instant when I saw from Tokyo that this was what the PRC dictatorship was doing, I knew it was time to get out of Dodge. (I just wasn’t counting on Dodge following me across the Pacific…)
Unless and until there’s either a proven vaccine or a proven palliative or both, this thing will keep spreading and ravaging. This is not to excuse the Whitmers and Newsoms out there — quite the contrary — but rather to emphasize that status quo ante is not an option at this juncture.