Opher Donchin

The Life Sciences Department at BGU (@LsBgu) has an open Tenure Track Position in Systems Neuroscience! Your lab can be somewhere in this building :-) For more details: bgu-academic-recruitment.my.site.com/Recruiters/VF_…

We're hiring! Tenure-track faculty position in Systems Neuroscience @bengurionu (BME, all ranks). Areas: neuroimaging, neural prosthetics, electrophysiology, translational neuroengineering & more. Apply: bgu-academic-recruitment.my.site.com/Recruiters/VF_… #Neuroengineering #AcademicJobs #Neuroscience

New article published in Communications Psychology! Large reaching datasets quantify the impact of age, sex/gender, and experience on motor control. nature.com/articles/s4427…

excellent new paper breaking a long-standing impasse by the identification of specific behavioral conditions under which CINs are recruited to augment DA release in vivo : Cholinergic modulation of dopamine release drives effortful behaviour nature.com/articles/s4158…

@KordingLab @vineettiruvadi @ehudahissar That's ok. It may have run its course.

@opherdonchin @vineettiruvadi @ehudahissar I can’t parse this thread, sorry.

This Gershman and Ullman paper is just so cool in showing that humans really can't not view correlation without assuming causality. Summary of my quixotic life in neuroscience: gershmanlab.com/pubs/GershmanU…

@KordingLab @vineettiruvadi @ehudahissar Because proving that there are many conditions in which what everyone assumes is actually true isn't nearly as interesting as identifying the places where the common wisdom is wrong.

@opherdonchin @vineettiruvadi @ehudahissar Wait how is a geometrical statement about the covariance structure ideology or salesmanship? In a 1/f world we can answer exactly the how correlated are correlation and causation question!

@KordingLab @vineettiruvadi @ehudahissar That's just ideology and salesmanship. The interesting question is how much we know about where the boundary lies. Separately: how many correlations of what strength would be sufficient to be evidence for causality? IOW: does causality require agency?

@opherdonchin @vineettiruvadi @ehudahissar I think as it stands it makes the "correlation does not suggest causality" argument under a wide range of assumptions. But you are right that I could probably sharpen that.

@KordingLab @vineettiruvadi @ehudahissar I think the ms frames the question wrong. Correlation <> Causation is old hat. Correlation suggests there may be causation under the following conditions is much more productive and interesting.

@opherdonchin @vineettiruvadi @ehudahissar So I should expand the ms into a longer paper?

@KordingLab @vineettiruvadi @ehudahissar I'm curious about this being written up quantitatively.

@vineettiruvadi @opherdonchin @ehudahissar Its a good predictor if there is no or little collinearity or if the system is low-d

@vineettiruvadi @KordingLab @ehudahissar Isn't that what a predictor is? How about: there may be a correlation between causation and correlation, under appropriate priors about world structure?

@opherdonchin @KordingLab @ehudahissar *never said predictor, just that it was a good starting point.

@KordingLab @vineettiruvadi @ehudahissar This is fun. McEleath's Statistical Rethinking has a couple of great chapters on this. It's not what I meant, though. I want the conditions under which correlation is a good predictor for causation as @vineettiruvadi suggests.

@opherdonchin @vineettiruvadi @ehudahissar I unfortunately don't think there are papers. I quickly (and dirty) wrote something up a few weeks ago: dropbox.com/scl/fi/lg7r257… -- the problem is that everyone in stats knows so its unpublishable.

@KordingLab @vineettiruvadi @EhudAhissar (probably not here) says that causality is not a coherent concept for systems in steady state. His talk on this in the sensory domain is really fascinating.

@KordingLab @vineettiruvadi I'm not sure I've seen any sort of quantitative study of how these things all interact, especially given priors on the natural distribution of prevalence and magnitude of causal effects. Is there something out there?

Let’s bring science back to X. Not influencers. Not grifters. Just scientists. Who’s worth following? 👇

@_alice_evans 1000 miles - The Proclaimers

What are the most romantic songs from world history? NOT songs about unrequited love NOR short-run intimacy ('Baby it's Cold Outside') But songs that celebrate mutual devotion, where each party's utility is weighted equally? I have a research idea & want your help!! 🙏

@CVakalopoulos @KordingLab Possibly. My question ("how reliable should the literature be, ideally?") may also be the wrong question. What is the right one? I love good questions!

That’s entirely the wrong question. It’s a wholesale change in academic culture that’s required and much of that is a need to dispense with the current vetting system of default markers that creates an indefinite number of paper mill themes, ones that simple correlational studies engender

People here believe that the majority of neuroscience papers transparently do not contribute to science. If so, what does that mean for our field? How can we change things?

@KordingLab There's always tradeoffs. It doesn't have to be black and white or universal. It would be good to encourage them and offer excellent opportunities. Compulsory classes aren't bad but will not create 100% change.

@opherdonchin You hold that the methods geniuses who can't produce a logical argument can't learn how to tighten their argument? Or that its not worth their time?

@KordingLab Logical clarity seems like an arbitrary standard. I know scientists who consistently identify field changing experiments and can't put together a coherent argument. I know methodological geniuses, too. That also moves the field.

@opherdonchin I agree. But not that papers from median institutions lack logic or clarity than those from top institutions. On average probably lower impact, maybe more incremental, certainly less expensive setups. But maybe less logically inconsistent as no need to pretend "first".

@neurosutras @KordingLab I've been at low tier and high tier institutions. There's a lot of luck involved but in aggregate the scientists at the higher tier institutions worked harder and smarter as grad students and the grad students work harder and smarter today.

@KordingLab 100% is rarely optimal. You must trade off false positives with false negatives and strong constraints have their own costs. That said, your current qualified formulation may not contradict your original survey.

@opherdonchin I think that 100% of highly read papers should be done in a way where the logic is clear to anyone with deep understanding of the field.