Hacker News Reader

Comment by epistasis | Hacker News Reader

There has been something very strange going on at Stanford since this began. Tons of really really bad stats, stuff that should destroy a person's career, even if it wasn't during a global pandemic and the bad stats resulted in the death of people.

Politics have overtaken some people's scientific sensibilities. Fortunately there are still lots of great people at Stanford, but they aren't the ones who are going on mass media.

View on HN

It's not all of Stanford, just one group (Jay Bhattarcharya's) that published a paper that betrayed extreme ignorance of binomial confidence intervals and modeling in general and is now the subject of this article. Why anybody would listen to this group despite after that is a good question (although, since poor understanding of statistics is rampant in the Medical field, let alone among the general population, perhaps not too surprising).

(disclaimer: Stanford '09)

It's Battarcharya and a few others, including Ioannidis. However Stanford is completely complicit, and when I search for these people I get results like this that promote bad information and reasoning:

https://fsi.stanford.edu/news/coronavirus-deadly-they-say

Rather than anything from the university that promotes the more accurate statements from the rest of the faculty.

So I would say that Stanford carries a lot of blame for letting their name be used for disinformation, and is not pushing back at all, which is a huge disgrace. If you are an alum, make some noise and refuse to make contributions until they start at least giving equal PR time to the scientifically based viewpoint.

Edit: and how could I forget Michael Levitt, who is now following in Kary Mullis' footsteps as a Nobel winning biochemist that promotes for crazy viral theories that damage public health

https://www.stanforddaily.com/2020/05/04/qa-nobel-laureate-s...

Well the Daily is the independent student newspaper and has little to do with official University policy. I don't know what FSI is, perhaps some department that Bhattarcharya is part of. I agree, if it was being pushed by official Stanford news sources (a search for Bhattarcharya on https://news.stanford.edu/ reveals nothing, although plenty of other COVID-related research), that that might be problematic. I agree it's not a good look for Stanford, but but if people conflate researchers from a university with a university, then that's on them.

That “conflation” is the exact same thing as the University’s reputation. If they don’t value their reputation, fine, but these bad researchers are only getting press because they are attaching the univerity’s name to their ideas. That’s not something to blame readers for. And if there are researchers at these universities that oppose them, it’s up to the university and other researchers at Stanford to stop their name being use if they disagree with its use.

Well I might be too knee-deep in academia to be a fair judge, but every university is a bunch of independent researchers of variable quality with wide latitude to pursue whatever and it seems silly to judge a university's reputation based on some individual researcher (or alumni, or whatever). I think any conflation is more the fault of media articles about the research than anything else. Yes, you write your affiliation on papers, but that does not really construe any sort of endorsement. I'm sure one can find many Stanford faculty expressing contrary viewpoints, but those are not (for whatever reason) being amplified by the media.

I can put a crazy paper on the arXiV tomorrow with my affiliation and if some media report says "The University of Chicago is claiming a perpetual motion machine" or whatever, that's the media's fault for misrepresenting the situation.

Curious, explain to me why I should listen to a random person on the Internet over a university research group? "betrayed extreme ignorance of binomial confidence intervals and modeling", lots of words but zero evidence supporting your assertions. If you have any concrete evidence to contradict their findings rather than just a blanket "extreme ignorance of binomial confidence intervals", that would be far more useful.

Don't trust one research group, trust the rest of the scientific body rising up to demolish them for obviously bad work:

https://arstechnica.com/science/2020/04/experts-demolish-stu...

I think calling it "extreme ignorance of binomial confidence intervals" is incorrect and far too charitable. The authors were intentionally deceitful. Co-authors of the preprint requested that they be taken off the pre-print because they had warned that the statistics were bad. Prior the the prevalence study, they had cherry picked weird data sets and published an editorial. They are clearly trying to bend the data to their preconceived notions, in a really public way.

To see journalists pick up their work again, just because they are at a big name institution, is a huge betrayal of the public trust.

> scientific body rising up to demolish them for obviously bad work

Come on man, let's be fair: a couple of tweets and a few snarky emails hardly consist of the "scientific body demolishing them." I'm not sure if I agree with Bhattacharya's study, but it's hardly been universally repudiated. This just low-effort politicizing by Ars Technica (which has a spotty record to begin with).

Even this critical blog post[1] has all kinds of colorful discussion (mostly by academics) showing that things are simply not as cut-and-dry as we'd like them to be.

[1] https://statmodeling.stat.columbia.edu/2020/04/19/fatal-flaw...

Worth looking through here and related links: https://statmodeling.stat.columbia.edu/2020/04/19/fatal-flaw...

No reason you should listen to me in particular (no doubt I have made egregious statistical errors myself in my research), but a good place to start might be https://en.wikipedia.org/wiki/Binomial_proportion_confidence... or, as others have posted, it's worth listening to Andrew Gelman (unlike me, a statistics expert) about this! https://statmodeling.stat.columbia.edu/2020/04/19/fatal-flaw...

Essentially the problem with the study is that the bounds on the false positive rate of the test used was such that the data are completely consistent with all of the positives being false positives.

Thanks!

>Politics have overtaken some people's scientific sensibilities. Fortunately there are still lots of great people at Stanford, but they aren't the ones who are going on mass media.

They've always leaned far more conservative than almost any other academic institution I can think of. Probably it's to do with the pervading influence of big business there. The Hoover Institution is one of the oldest conservative think tanks in the US.

It's weird because I had never heard that Jay Bhattacharya was super political. But early on he designed a study with unethical participant recruitment, juked and terrible stats, to push a bad pre-print for a contrarian number on the death rate that has proven to be very very very wrong. And he went out into media and pushed his bad study, at a crucial time. That the Mercury News is publishing his words is a huge black eye both for Stanford and for the Mercury News. It's clear that his COVID work is not driven by science, but rather strange political beliefs that can not be justified by data. Shameful.

> ... a contrarian number on the death rate that has proven to be very very very wrong.

What was that "very very wrong" number? I found [1] from April in which Bhattacharya estimates that the IFR is probably one tenth of the CFR of 3-4%, which would mean an IFR of 0.3-0.4%. That doesn't seem very contrarian or wrong. The CDC in May estimated 0.26% IFR [2]. The WHO just the other day [3] said that 10% of the world's population may have been infected with the virus. They also estimate ~1,000,000 deaths globally. That would imply an IFR much lower than 0.3%.

[1] https://padailypost.com/2020/04/06/stanford-experts-say-covi...

[2] https://www.usatoday.com/story/news/factcheck/2020/06/05/fac...

[3] https://apnews.com/article/virus-outbreak-archive-united-nat...

[4] https://www.who.int/docs/default-source/coronaviruse/situati...

The CDC estimate was from before we had any good studies using antibody prevalence. It was too low.

The WHO estimates a fatality rate of 0.6%. If you are simply dividing deaths by (an estimate of) current cases then you will get an inaccurate number because many of the current cases will eventually die of the disease.

If CDC's mis-estimate from May can be forgiven by virtue of there not being enough good studies at the time, then why is Bhattacharya's even earlier (and more accurate) estimate being characterized as unconscionable?

> The WHO estimates a fatality rate of 0.6%.

I don't see how that makes an estimate 0.3% to 0.4% "very very very wrong", especially considering that we haven't heard the final word on this.

> If you are simply dividing deaths by (an estimate of) current cases then you will get an inaccurate number because many of the current cases will eventually die of the disease.

I didn't do that, but if you look at the numbers you'll see that it wouldn't make a meaningful difference anyway. WHO numbers show over a million deaths since the start of the epidemic, and around 5000 deaths per day over the past month. Whether you use 0.13% or 0.14% for the implied IFR, the WHO's comments the other day are out of whack with the 0.6% fatality rate estimate.

I don't quite know what to make of the WHO estimate. I can't find further details about it. I don't think it makes sense to take this estimate and try to extract an IFR from it. They could be suggesting that the IFR of the virus is much lower than expected, but they could also be suggesting that deaths have been badly undercounted (e.g. in areas with poor health infrastructure), or that it has spread more widely among younger populations in places like Africa, where IFR would be expected to be much lower.

Bhattacharya's estimate is being criticized because, even though he was taking a contrarian view on the virus and getting results that were out of wack with other lines of evidence, he rushed out a highly flawed study (the Santa Clara seroprevalence study). This came out when we had little data about antibody prevalence, his study was one of the first ones, and it had the big name of Stanford behind it, so it was reported very widely and misled an awful lot of people about how dangerous this virus is. In his position he should have been bending over backwards to make sure he was on sure footing and not misleading the public, and instead he let this paper full of basic mistakes go to press.

Could you clarify your statements and provide references to the bad stats?

I'm not sure if you are accusing the Stanford doctors in the article of bad stats or those arguing against the doctors in the article.

I'm talking about the Stanford doctor in this article. Here's a summary of what he's been doing since the start of this, and he's been very wrong:

https://arstechnica.com/science/2020/04/experts-demolish-stu...

> In fact, criticism of the two studies has woven a damning tapestry of Twitter threads and blog posts pointing out flaws of the studies—everything from basic math errors to alleged statistical sloppiness and sample bias

Unless I'm missing something, it's Twitter and blogs against published studies. Sounds like he's been proven wrong all right...

Damn, these are the same people who came up with that ridiculous antibody study which fueled the crackpot "widespread immunity" theory? They must have no shame to keep pushing their contrarian covid takes.

Like what are you talking about specifically?

How much of this is coming from https://www.hoover.org/?

Reply on news.ycombinator.com