search

jasp-stats / jasp-issues

This repository is solely meant for reporting of bugs, feature requests and other issues in JASP.
58 stars 29 forks source link

Implement Smarter Starting Value for Newton-Rhapson in B. Informed T-Tests #85

Open quentingronau opened 7 years ago

quentingronau commented 7 years ago

Suggestion by Richard: Maybe use quantile value based on conjugate analysis?

quentingronau commented 9 months ago

So this is about the function that computes the quantiles of the posterior. Currently it uses the posterior mode as starting value, but there might be a better way of choosing this starting value as suggested by Richard.

quentingronau commented 9 months ago

We have a paper about the informed t-test, but it does not give details about how the quantiles are computed in JASP. The best approach would be to simply look at the JASP code (lines 131-156).

shun2wang commented 9 months ago

Hi @tomtomme just another topic for issue labels: please note waiting for requester label will auto colse the issue somedays ago. So we should add it carefully for feature request issue😜 Thank you for your work!

tomtomme commented 9 months ago

@shun2wang Yes, the tag was not appropriate here. Thx for the hint!

quentingronau commented 9 months ago

I am not sure if you are aware of this, but I have written this code in my time as a JASP team member. This issue was meant as a reminder for myself, but I never saw the need to change it because it is fast enough. Also, knowing the Bayesian experts on the JASP team in person, I am more hopeful than you that they know what "use quantile value based on conjugate analysis" means :). But again, personally I think what we have at the moment is perfectly fine.

tomtomme commented 9 months ago

Thx for the info! I were not aware. So we leave it as is and you can close this if you feel the time has come or have time to dig into the code again. I am just someone from the translation team who tries to clean up all those old issues, so I really appreciate your fast responses.

vandenman commented 9 months ago

@tomtomme this issue is indeed something we shouldn't forget. At the same time, the priority is not that high (it's an improvement, not a bug). It would be useful to reorganize these kinds of issues into a project at some point because right now it's very easy to lose track of them.