BF changes
Hi everyone,
soooo, I have run a super easy Bayesian Wilcoxon test on the version 0.13 of Jasp, and comes out that everytime I run it, the result changes. How is this possible?? something I don't understand?
I attach the result of three runs made one after the other.
Comments
Hi,
I'm just a JASP novice but I would suspect that the program does not use an analytical solution to calculate the posterior. Instead it perhaps approximates the posterior using sampling. This changes the results somewhat every time you run your analysis.
If this is true, one would expect that, if you increase the number of samples you draw from the posterior, this should reduce the sampling error, which in turn should reduce the BF changes that you observe every time you run the analysis.
Cheers,
Michael
Correct. But see also https://forum.cogsci.nl/discussion/6354/student-and-wilcoxon-sr-bayesian-tests-in-jasp-one-sample-a-bug#latest
!! We will have addressed this in the next version.
Cheers,
E.J.
thanks to both of you!!
PS so it's best if I do not report the test in a paper I am writing...I'd rather wait next version...(the most scary part is that on some comparisons, I had BF varying from anecdotal to strong evidence in favour of H1, so that's quite a change...)
have a great day!