Nature headline: “Timing matters for COVID vaccine effectiveness: Younger and older people gained greater protection if they had their jabs in the middle of the day.” Good joke: “Went from ‘95% effective’ to ‘only works if you get it at lunch time’”.
This is a (yet another) good example of how statistical modelling (and therefor science) is often totally off base. Any researcher and reviewer working with statistical models should read: Leo Breiman's - Statistical Modeling; The Two Cultures and Gerd Gigerenzer - Mindless Statistics.
Following the advice in these two paper's should go a long way in redeeming us of such idiotic statistical modelling articles.
'We fed all of the data that we had into a sorting algorithm and these were the most significant numbers that it came up with. We can't suggest any kind of mechanism or plausible reason at all. We didn't begin with a hypothesis and either disprove it or fail to disprove it.'
I like the post Briggs but since your stack is 'Science is not the Answer', I have to point out what you already know, this study ain't science. Not even close. It is just spare cycles on somebody's laptop, turned into fodder for a grant application(pay me i'm studying covids!!). Would have been more productive mining bitcoin or doing SETI@home or whatever people do with spare cycles these days.
Maybe rearranging our whole society to worship at the altar of statistics wasn't our brightest idea?
What's not so funny is that there will be a lot of people believing their conclusions, and also there will be many "Peer Reviewed" papers issued in the future that will use this one as a source.
I noticed that Cox regression is all over the place in medical academic publishing, probably because it is part of every statistics software, because its input interface is easy, and because it never complains (i.e., it always produces results, and never asks "are you sure that you want to run Cox regression with control for 24 variables on a sample in which only a few hundred events were observed?"). Two examples I dealt with: bivalent booster effectiveness (https://cm27874.substack.com/p/bivalent-intransparent-ambivalent) and effect of Covid vaccination on fecundity (https://cm27874.substack.com/p/fun-with-fecundity).
A ridiculous deflection for a ridiculous injection and ridiculous virus.
...for not-so-ridiculous funds and not-so-ridiculous status points.
Hyperbolic Hersterical Harridans is a good band name!
Lots to laugh at.
"Laugh laugh, I thought I'd die, it seemed so funny to me."
Much more to cry about.
"Cry the Beloved Country."
Cry wolf.
Crimea river.
"Just cryin in the rain."
"Don't it make my brown eyes blue."
BTW: I had my bad-ass booster during a lunch break 18 months ago and have suffered heartburn ever since.
This is a (yet another) good example of how statistical modelling (and therefor science) is often totally off base. Any researcher and reviewer working with statistical models should read: Leo Breiman's - Statistical Modeling; The Two Cultures and Gerd Gigerenzer - Mindless Statistics.
Following the advice in these two paper's should go a long way in redeeming us of such idiotic statistical modelling articles.
'We fed all of the data that we had into a sorting algorithm and these were the most significant numbers that it came up with. We can't suggest any kind of mechanism or plausible reason at all. We didn't begin with a hypothesis and either disprove it or fail to disprove it.'
I like the post Briggs but since your stack is 'Science is not the Answer', I have to point out what you already know, this study ain't science. Not even close. It is just spare cycles on somebody's laptop, turned into fodder for a grant application(pay me i'm studying covids!!). Would have been more productive mining bitcoin or doing SETI@home or whatever people do with spare cycles these days.
Maybe rearranging our whole society to worship at the altar of statistics wasn't our brightest idea?
Obligatory XKCD reference:
https://xkcd.com/882/
I mean *really*? This is *obvious* p-hacking.
I'm laughing so hard, I'm crying.
What's not so funny is that there will be a lot of people believing their conclusions, and also there will be many "Peer Reviewed" papers issued in the future that will use this one as a source.
This is exactly precisely definitely, and depressingly, how it works.
...culminating in meta-analysis of systematic review—which will ultimately seal the sientific™ consensu$ for posterity to marvel at.
I noticed that Cox regression is all over the place in medical academic publishing, probably because it is part of every statistics software, because its input interface is easy, and because it never complains (i.e., it always produces results, and never asks "are you sure that you want to run Cox regression with control for 24 variables on a sample in which only a few hundred events were observed?"). Two examples I dealt with: bivalent booster effectiveness (https://cm27874.substack.com/p/bivalent-intransparent-ambivalent) and effect of Covid vaccination on fecundity (https://cm27874.substack.com/p/fun-with-fecundity).
The ease of software has not made things better. Good grief. Thanks.