Posts

Bayesian and Frequentists Approaches to Statistical Analysis 2021-04-26T00:18:00.668Z
Bayesian Inference with Partially Specified Priors 2020-03-01T05:02:38.075Z

Comments

Comment by Christopher Clark (christopher-clark) on Credibility of the CDC on SARS-CoV-2 · 2020-03-08T11:53:23.842Z · LW · GW

I agree with some of the sentiments in this post, but I think the claim in the second paragraph "Unfortunately, the CDC has repeatedly given advice with lots of evidence against it", is poorly supported. It suggests that the CDC has given advice that is not just incomplete or somewhat off-base, but that is ineffective and should be ignored. I don't think the points that deal with advice meet that standard:

Packages: The CDC quote explicitly refers to packages from China, so this more a matter of missing advice about what to do in other cases than bad advice.

Masks: At the end of the day, "don't buy masks" seems like good advice that ought to be followed. I get the annoyance that the CDC or others might be trying to downplay the fact masks can help healthy individuals, but that doesn't mean the recommendation is wrong.

Genetics and Environment: The general sentiment of "please keep in mind the odds of a Chinese-American having COVID-19 is very similar to anyone else having COVID-19" is pretty good advice. Sure, you can nitpick the language and say the CDC implied "exactly equal" instead of "very similar", but I think it's pretty pedantic to use that to justify calling this "advice with lots of evidence against it". The general point that is trying to be made here is correct.

Gave False Reassurances About Recovered Individuals: We should have some epistemological humility here, one paper published a month ago and a handful of anecdotes shouldn't give us a lot of certainty that the 14 day period suggested by most experts is a mistake. Even if it was, its possible people not in quarantine and going about their day-to-day lives will have a higher chance of having COVID-19 than people who were exposed and then quarantined for 14 days anyway due the increased chance of other, incidental exposures. If the general bit of advice here is "treat people who left quarantine as you would anyone else who you have no reason to believe is infected", that seems like pretty good advice. I suppose you could argue that, given the uncertainty, one should be slightly more carefully about people who left quarantine very recently, but again in my mind that is an insufficient caveat to justify calling this "advice with lots of evidence against it".

We should remember that much of the CDC's website is meant for the general public, and is mostly trying to remove naive misconceptions people might have (e.g., people who have been quarantined and Chinese-Americans are very likely to be infectious). It is not trying convey very precise statistical information or make detailed technical claims about COVID-19, and shouldn't be interpreted that way. From that perspective, the CDC's advice discussed in this post seems fine. There might be some issues with the phrasing or the details, and maybe there are things missing, but I think it's a stretch to call it erroneous.

Comment by Christopher Clark (christopher-clark) on Bayesian Inference with Partially Specified Priors · 2020-03-01T20:31:35.606Z · LW · GW

That makes a lot of sense, but it does require you to know your utility function ahead of time. When this is not the case we might still want to propagate whatever you know about the prior forward to the posterior as a kind of caching operation for use in future decisions.