Psychophysics has always been great.

i had experiments where double numbers were not sufficient for p values

Hurray we did! www.argmin.net/p/1000-hurts

I think the difference is that you view their paper as a theory commitment and I view it as methods development?

you think the paradigm (supported by their open source IP) is not asking the right questions.

but wait. How is this related to the original paper?

But you know, truth and epistemic clarity are not really things decided by popularity at SFN.

Now, the modeling I am talking about does not be "modeling modeling". It can simply be something along the lines "if this goes up then this other thing should go up" or "this perturbation will make this variable go down". I am not an equation purist.

If I say that harking is not a good thing then yes, I will not be popular among certain folks. It is so deeply engrained in the literature. You can do exploratory modeling (say implicitly plan to ask 1000s of questions) but then you need a 5 sigma kind of effect. Sorry.

But effect size is so big we use the slope itself as measurement. Still big effect sizes us. Much of perceptual psychophysics is like that. Take this medicine!

If you don't plan your science (try my planyourscience.com to make that easier) you are pretty much guaranteed to hark.

@gunnarblohm.bsky.social is the best!

Yeah that feels about right. IMHO it’s pretty safe and generally super disciplined in a deeply undisciplined way.

And yes, I personally quite like it. Which is weird given that I am rather a techno head.

I am interesting why (a) evolution made it so people like it. Maybe group bonding? I am very unsure. (b) what the mechanisms are for that.

indeed. But a happy gas particle!

In massively subsampled systems, directed predictability can be systematically uncorrelated with intervention effects because the dominant confounders are unobserved. Conditioning does not rescue this. (also see arxiv.org/abs/1812.03363)