Back to the future!

the obvious answer is that i am incorrect, uninformed, obstructionist, because otherwise these 10 prior papers from high impact journals would never have been published...

try explaining immortal time bias to a bunch of ICU researchers. their response has invariably been to hand me a stack of papers studying the same condition where the clock started at ICU admission, regardless of if the pt had been in the hospital for 5 min or 5 days before going to the ICU.

i'm sure it depends on the situation. but i agree that there is not a lot of critical thinking. and while sometimes overly compelx methods are not the answer (or even a smokescreen for junk data), sometimes they are needed, since--as you said--human health is complex.

to wit, i am currently trying to convince a clinician that even tho a prior paper used "avg marginal effects", the quantity that they really want is is avg marginal effect at the mean. 🫠

medical journal reviewers will scoff at an unfamiliar method. but if there is a ref to other papers that used similar methods, then everything is fine. in my own experience, this phenomenon permeates medical research field.

2nd, and more importantly, is that they don't understand enough stats/methods/etc to really understand what they have, and whether or not they can even get what they (think) they want.

to me, the reasoning is twofold. first, they see something that was published using method X and, ergo, another paper using method X is also likely to get published.

agreed--great for teaching/learning. flat/default priors can nicely highlight how the likelihood/data dominates the posterior. it's also illustrative, imo, when a frequentist and bayesian model give similar results--really drives home the need to understand priors and how they influence posterior

I have a serious love-hate relationship with linear algebra.

This has become my default approach because the search function is so horrid.

"linear in the predictors" is something i probably heard 500 times before someone explained it to me. and a few years passed before i truly understood it.

similar to "holding other predictors constant". a critical stats concept that sounds nice, but understanding it is maybe not so intuitive...

‘Promiscuous dichotomization’ is a nice turn of phrase.

What time is dinner? And also, what’s your address? I can bring a nice bottle of wine.

I’m trying no to go full chicken little, but given how often I’ve gotten garbage results from LLM queries…

Chat gpt fails the Turing test. Numerical integration is magic!