Ref. A wasn't crazy about my "Informal Reflections" but the editor liked it and ref B had no problem. Jamie Robins seemed to particularly like it. People could read the paper and draw their own informal reflections. You say "the likes of Alina and zeynep" as if it were a bad thing.
I'm interested in your response to the paper's content, describing how to fix a basic error in statistical logic. econjwatch.org/articles/an-...
For a less-technical summary, here's the interview econjwatch.org/podcast/mich...
with Jamie Robins (www.rousseeuwprize.org/2022)
True!
People have bad intuitions for what happens when something flies off a merry-go-round, poor intuitions for conservation laws, etc.
They should introduce him by playing this and dozens of other similar interviews from 6 years ago.
Turns out ChatGpt lies repeatedly when hammered on an easy apolitical question.
michaelweissman.substack.com/p/interrogat...
It doesn't even say that the probability that your null is true is less than p. They can be orders of magnitude different. In medical experiments with new treatments small tests often give p >0.05 though the null (no effect) cannot be true, although you don't know the effect sign or magnitude.
see:
The p-value is the chance that a result as far or farther from the value expected for a null hypothesis wold be obtained if that null were true. p<0.05 definitely does not say "your hypothesis is true." It disfavors the hypothesis it tested, the null & doesn't even refer to any alternate hypotheses,
(95% confidence interval). But these numbers simply tells us the likelihood of an effect, they are not a stamp of certainty."
of uncertainty in these analyses is less than 5%. We report this as a probability value (p-value). If the p-value is less than 5% (p<0.05), we interpret this as a significant effect. Our hypothesis is true. In association studies, we also report the flip side: the 95% of data we’re confident in ...
1/few A really thoughtful, useful essay. But one paragraph, a bout p-values, is all screwed up. "When researchers test whether an effect is likely real they perform statistical analyses that assess how likely the results could be found by chance. Across nearly all of science, the accepted rate ...
🧪Dunno if this is useful to anybody but I scrounged some old notes introducing likelihood methods and Bayes, with some exercises included. A supplement to an intro stats course.
michaelweissman.substack.com/p/likelihood...
A naive question: Do their methods allow comparison of selection rates at different periods even if not well-calibrated for how strong or what type of selection was involed?
Maybe it would be good to keep this more technical, less of those personal asides.
Worobey et al. picked that statistic to feature. I pointed out it had the wrong sign for their model. The point was that, as you say, their model was wrong.
No post-hoc p-value is generally considered publishable unless it's off-scale low. Noticing unusual low-p features is a natural thing to do but due to implicit multiple choices a marginal p-value isn't considered significant. That's what I meant by it.
The preprint you mention explicitly describes post-hoc tuning of the properties chosen. E.g.
ps. I also wear a mask indoors, have long covid, etc. But I'm not going to look for excuses to butcher basic logic in service of my team. Policy should be based on objectively evaluated facts, not group identity.
So I said that overturning the P2022 result had a smaller effect on the L/Z odds than you'd think from its original pro-Z hype. I gave no quantitative estimate of those odds, way beyond the scope . I did point people to my estimate of them, clearly labeled "opinion" in the "informal reflections."
The paper was widely advertised as increasing Z/L odds. That means that if the result was upside down it should decrease Z/L odds. I pointedly stay away from saying whether that effect was as big as the authors claimed or whether, based on these data, the odds favor Z or L.
July 21 1935: The Führer orders a new reception hall and ballroom to "entertain foreign diplomats" at the Chancellery. Underneath the dance floor is an air raid shelter. A 2nd deeper bunker is built next to this in 1944, where he would spend his last days as the Reich collapsed.
By "not publishable" I mean what the ref seems to have thought, that p=0.036 for a carefully post-selected property is not publishable grounds for rejecting a hypothesis.
You could submit something about the interesting clock reversal known before P2022 (academic.oup.com/mbe/article/...) to the Discussion at EJW, useful for context tho irrelevant to the main point. Mentioning that P2022 went to NYT before it was publishable just seems mean.
I guess you mean zenodo.org/records/6342... in which they in effect made no requirements of any sort on their two-spill hypothesis. Apparently a referee required some balance, but then they still left a couple pieces out. You really want to embarrass them further?
I guess you mean zenodo.org/records/6342... in which they in effect made no requirements of any sort on their two-spill hypothesis. Apparently a referee required some balance, but then they still left a couple pieces out. You really want to embarrass them further?
I'm curious what your take on this is.
michaelweissman.substack.com/p/bhattachar...
I see some mixture of being wrong, right, tendentious analysis, justified doubt, unjustified refusal to admit errors, and a quietly expressed intense ideology that's incompatible with public health.
My thoughts on the Covid origins group-think frying pan
econjwatch.org/articles/an-...
econjwatch.org/podcast/mich...
and the MAGA fire.
michaelweissman.substack.com/p/bhattachar...
🧪 Jay Bhattacharya will be a lead speaker in an upcoming NAS workshop on scientific integrity. Thats a bad choice: michaelweissman.substack.com/p/bhattachar...
JB's ideology drives not only his policy choices but his science. He's right about Covid origins but wrong about more immediate issues.
For a less-technical account, you can listen to this interview
econjwatch.org/podcast/mich...
with Jamie Robins www.rousseeuwprize.org/2022
