Wonderful method by @sweiwang.bsky.social @cfcamerer.bsky.social et al on optimal design for pref parameter elicitation here: sites.pitt.edu/~swwang/pape...

For those who are interested in digging deeper into this, Ryan has just posted a very thoughtful and thorough reply here: bit.ly/4bQk0P0

The student sample was conducted online too, though with more oversight via zoom. I think this pushes the debate from online vs. lab to online sample vs. student sample

The models won’t be “complete” (every model is wrong etc), but they will be useful.

But this is a bet.

(Which is what makes the current period in BE so exciting)

Fin

Yes exactly, enter all the interesting work on description - experience gaps

Great thread. It’s possible that noise in cognitive representation of probabilities (or proportions) are driving some of these results. So to the extent that there’s more noise when probabilities aren’t explicitly presented in real world, perhaps you’d get broader scope of PT-like behavior

Your first point seems consistent with his conclusions…that these paradigms may have been measuring some information processing constraints rather than risk prefs all along?

2) all noisy coding models feature a prior which is first order for shaping choice, but irrelevant when perception isn’t noisy. To me, this is the key separating prediction and it’s testable at least experimentally

Agree that model fits are v useful but these new models do make some bold and testable predictions. (1) Enke & Graeber ‘23 show that empirical measures of cognitive uncertianty can explain choice bias and

This is great Dan—can you please add me? Thanks!