Given how the lab-leak hypothesis was treated initially and subsequently, a reasonable person must at least consider the possibility that this was not a naturally occurring virus.
This was posted here previously [1]. I'm just going to repost what I said about it last time [2], since my take on it hasn't changed. The lab leak hypothesis seems plausible, but this is a bad analysis.
-----
Disclaimer: I didn't read much of the analysis you linked. I'm not a virologist, but I do have a PhD in Bayesian Statistics.
As a general rule, I don't take this kind of analysis via repeated repeated repeated repeated repeated repeated repeated application of Bayes' Rule with somewhat-to-completely arbitrary probabilities at each step very seriously. At all.
It's a kind of gish gallop [3] with additional window dressing purporting to wrap it all up into one easy number. From a probabilistic point of view it ignores any dependency between observed facts, which is a serious issue but not necessarily the most damning one.
This particular case is even worse than usual. According to a quick Ctrl-F-aided skim of the document, the author doesn't identify a single fact that increases the probability of zoonotic origin, which is extremely suspicious. If every single available fact in a complicated issue points to the same conclusion that's probably because you are messing with your facts, not because that's how things actually are.
After noticing the one-sided-ness of the Bayesian updates I rolled my eyes and closed the tab.
If I'm understanding this, they have guessed the probability of various things, including events, and assembled them to assign some kind of probability for each possibility.
So it boils down to "if our guesses about some underlying probabilities are correct, this would be the answer". I see this a lot in financial analyses too, where some sophisticated analysis is done but it's just on top of guesses. It's not really accurate to say it "concludes beyond a reasonable doubt" - they have an argument, maybe a valid and supported one, buy the "bayesian analysis" is just smoke and mirrors on top of their initial assertions.
If someone has a different view of what the Baysian analysis adds, I'd be interested to understand.
5 comments
[ 2.7 ms ] story [ 23.8 ms ] thread-----
Disclaimer: I didn't read much of the analysis you linked. I'm not a virologist, but I do have a PhD in Bayesian Statistics. As a general rule, I don't take this kind of analysis via repeated repeated repeated repeated repeated repeated repeated application of Bayes' Rule with somewhat-to-completely arbitrary probabilities at each step very seriously. At all.
It's a kind of gish gallop [3] with additional window dressing purporting to wrap it all up into one easy number. From a probabilistic point of view it ignores any dependency between observed facts, which is a serious issue but not necessarily the most damning one.
This particular case is even worse than usual. According to a quick Ctrl-F-aided skim of the document, the author doesn't identify a single fact that increases the probability of zoonotic origin, which is extremely suspicious. If every single available fact in a complicated issue points to the same conclusion that's probably because you are messing with your facts, not because that's how things actually are.
After noticing the one-sided-ness of the Bayesian updates I rolled my eyes and closed the tab.
[1] https://news.ycombinator.com/item?id=25965751
[2] https://news.ycombinator.com/item?id=26129558
[3] https://en.wikipedia.org/wiki/Gish_gallop
This is just Hacker News-level commentary passing for science.
So it boils down to "if our guesses about some underlying probabilities are correct, this would be the answer". I see this a lot in financial analyses too, where some sophisticated analysis is done but it's just on top of guesses. It's not really accurate to say it "concludes beyond a reasonable doubt" - they have an argument, maybe a valid and supported one, buy the "bayesian analysis" is just smoke and mirrors on top of their initial assertions.
If someone has a different view of what the Baysian analysis adds, I'd be interested to understand.