Bayesian correlation - betaA note
Hi,
I came across this message underneath the output when running a Bayesian correlation "betaA > 0 is not TRUE"
I couldn't find any information about what this means. Should I be worried?
Thanks
Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
Categories
Supported by
Comments
The default setting for the Bayesian correlation is a prior width of 1. This is based on a stretched beta distribution that is symmetric around 0.5, so beta(a,a). The prior width is then defined as 1/a. It is clear that "a" should be higher than 0. How did you set the options so as to receive this feedback? We might use this to produce a more informative error message.
Cheers,
E.J.
Hi agata_,
The error message is not informative, as it implies that the Bayes factor could not be computed due to possibly a data error or an overflow. We should improve this. If possible, could you please open an issue with the jasp file on our github page? Or send me the jasp file by email (firstletter dot surname at jasp-stats dot org). Then I can have a closer look at it.
Cheers,
Alexander
It only happens for some variables in a large correlation table, and for some it goes away when I set the exclusion to listwise, but not always.
I left all the settings at default, so the beta prior width says 1.
Maybe it is a matter of missing values or unequal variable sizes. I will try to email the file, thanks a lot.
I also think that this happens where the BF value is an infinity symbol, which in previous versions didn't produce any error messages (I just updated to the latest one), so I just took it as "really large value". Does that mean I shouldn't rely on these results then?