Student and Wilcoxon SR Bayesian tests in JASP (one sample). A bug?
Dear JASP keepers,
Thank you very much for your amazing tool.
I would like to draw your attention to one issue that I have recently face with JASP (v 0.13.1).
I attached my data. I am trying to perform a one-sample t-test to test whether variables PersonalM, RewardM and EmotionM are significantly different from zero. Variable EmotionM violates the assumption of normality and I used Wilcoxon signed rank test. BF(H1):BF(H0)=0.197:5.079. However, if I am using the Student t-test BF(H1):BF(H0)=425000:2.35e-06. I cannot understand why these two test provide so huge differences. Even with different calculation procedures for the Student and Wilcoxon tests, the difference in BF should not be so dramatic.
I would greatly appreciate your help/comments/ideas.
Thank you in advance,
alla
Comments
Yes, this is strange. I'll forward this to our expert.
E.J.
Hi Alla,
This is a known issue, unfortunately. The Wilcoxon signed rank test was recently implemented while we were still working out the statistical details. We had chosen to add some steps to the algorithm to make it more stable, but it turned out that this can lead to very conservative Bayes factors that should not be conservative. In the next release, we will have removed this step and use other means to make the algorithm more stable.
Kind regards,
Johnny
Hi Johnny,
Thank you very much for your prompt reply. This makes sense. Thank you for looking into this issue.
Overall, I cannot thanks enough to all of you for the amazing software. My students love it! The most important, teaching Bayesian analyses become so much easier!
Best wishes,
Alla