Open tomtomme opened 3 years ago
Hello @tomtomme,
Thanks for the suggestions!
Finally bootstrapped CIs are missing for all correlation coefficients.
This has been implemented recently and will be part of the next release.
Otherwise great suggestions, I think we should definitely take a look into the ties and give that information to the user. I am not opposed to adding new coefficients, curious what @AlexanderLyNL and @EJWagenmakers think?
Best, Simon
@Kucharssim Is there any plan for this?
wow - you really added some nice stuff in 0.16.3 beta :D
Hi @tomtomme,
I am afraid I will have to let you down: The new coefficients Eta, Somers d, Kendalls Tau-c were not implemented, yet. The buttons of those coefficients are in the UI for some time as placeholders that indicate a future appearance of the functionality. However, they are hidden in a release mode, because the R code does not contain the functionality. The version you are trying is probably built in a debug mode so you now see the placeholders as well; however, they will do nothing as you were able to experience yourself.
I am sorry we have not got to this yet as we have a bigger fish to fry at the moment. I am wondering whether @arandolan would be available for implementing these coefficients at least for the contingency tables?
Also - wouldn't it also make sense to add all correlations to the correlation-tab and to split the correlation methods into nominal, ordinal, scale there too? Even Cramers V, Phi, Lambda etc. should be possible with raw data that is not yet in crosstabs / contingency tables. This would of course create a lot of duplication. Maybe it would be wise to merge crosstabs and correlations in the long run.
That's an interesting idea. We recently had a discussion about separating regression and correlation in the JASP ribbon (https://github.com/jasp-stats/INTERNAL-jasp/issues/1812). I would wait what we come up with there before ripping functionality from contingency tables and putting it into correlations. But let's keep an eye on it. Thanks for following up on this issue!
Very nice idea with the regression / correlation separation. This would indeed make it more feasible to maybe merge crosstabs with correlation! Thanks for the quick response and the info on the debug mode! Yes my build is marked as debug. And I know you are frying bigger fish. Do those first. This is not highest priority even on my personal list.
Hi all Just a quick note to say that I agree that additional correlation coefficients would be very useful indeed! Moreover, to suggest that a couple of these (Kendall's tau a and Stuart's/Kendall's tau c are implemented in desctools and an accurate reference for the latter took me a while to hunt down (so I think it would be very important to put in the help file!)
https://rdrr.io/cran/DescTools/man/KendallTauA.html
https://search.r-project.org/CRAN/refmans/DescTools/html/StuartTauC.html
Stuart, A. (1977). Spearman-like computation of Kendall’s tau. British Journal of Mathematical and Statistical Psychology, 30(1), 104–112. https://doi.org/10.1111/j.2044-8317.1977.tb00729.x
Others of those requested above are available in the correlation package:
https://easystats.github.io/correlation/articles/types.html
Makowski, D., Ben-Shachar, M. S., Patil, I., & Lüdecke, D. (2020). Methods and algorithms for correlation analysis in R. Journal of Open Source Software, 5(51), 2306. https://doi.org/10.21105/joss.02306
Hi @Kucharssim, I wanted to follow up on part of @tomtomme request. Specifically, is there a chance that this would be implemented: Phi & Fishers exact test for 2xX tables (R package "coin", from Fishers Exact Test Feature Request #99 ) and make sure to separate Fishers exact from Odds ratios You closed the other thread so that it could be tracked here, but I don't see any other mention of it except where it was originally asked. I have had to use the R Beta functionality to run R code for Fisher's so many times this year, so I was hoping that it would be integrated.
Thoughts?
Still valid for jasp 0.19.1
Missing
Papers for recomendations in docs
I have written an extensive review of the current literature on gamma vs spearman vs kendal vs pearson for my german textbook (work in progress). I would translate it, if needed for documentation when to use which.
Which Spearman Forumula is in use?
Also: spearman has a correction formula for data with lots of ties but jasp does not state if it uses that version:
or the shorter formula that ignores ties:
General
Eventually merge contingency tables and correlation modules and sort/split the correlation methods into nominal, ordinal, scale