I'm not 100% certain either. Thus this discussion. Very helpful
-
-
Replying to @GunnarBlohm @tyrell_turing and
I don't know all the details but maybe the LIP ramps vs steps debate would be a good case study for this. You could argue the two "sides" had motivation to find that their model fit better. So preregistration may have been appropriate.
2 replies 0 retweets 3 likes -
Replying to @andpru @neurograce and
This is a different point! Super important and very straight forward. No questions here. But that's not what I'm talking about
0 replies 0 retweets 2 likes -
Replying to @andpru @GunnarBlohm and
Right. Also, consider: what does prereg accomplish for experiments? It prevents ppl from hiding negative results. But, what are negative results in a model? What do we gain from knowing that it never worked other than the fact that the authors' math was wrong?
1 reply 0 retweets 1 like -
Replying to @tyrell_turing @andpru and
Say you hypothesized that servo control can explain the minimum intervention principle. You implement that hypothesis and find no. So you turn to optimal control and it works. Wouldn't it be good to know about the initial null result?
1 reply 0 retweets 3 likes -
Replying to @GunnarBlohm @andpru and
So, I guess my question would be: do you have mathematical analyses to show that servo control can explain it? If so, then the fact that you struggled with your simulations tells us either (a) the math was wrong (in which case, not publishing is okay) or (2) your coding is weak.
2 replies 0 retweets 1 like -
Replying to @tyrell_turing @GunnarBlohm and
Q: should mathematicians preregister? I suppose that my perspective on models is that they're a bit closer to math than experiments, and my feeling is that there is no need to publish negative results in math because failure to prove something is not super informative.
1 reply 0 retweets 1 like -
Replying to @tyrell_turing @GunnarBlohm and
But, I recognize that my perspective does not encompass many models in neuroscience, where it really is closer to experiments than it is to mathematics. In that case, yes, I can see value to prereg (or at least very strict rules about fitting procs and counter model testing).
1 reply 0 retweets 2 likes
What's not covered by crossval here?
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.