High bayes factor but no effect

Feldkamp27

Hey,

in my analysis of effects output table, for one variable "lvisC" I got a bayes factor of 175.625. However, when I looked at the model averaged posterior summary, I could not find any effect. Is this possible, considering that the there is 175 times (very strong) more evidence for the alternative hypothesis? By the way, the output is very strange , as "lvisC" (M=6.20; SD=0.7) and "mvisC" (M=5.86; SD=1.25) almost have the same mean only differing by 0.3 and for mvisC I found a bayes factor <1 and no effect as well.

Kind regards,

Niklas

Feldkamp27

EJ

Dear Niklas,

This is strange behavior! Apparently lVisC is in almost all models with appreciable posterior probability, but its model-averaged posterior includes zero. Under usual circumstances, this would not happen, so I'm going to guess that the circumstances are not usual; perhaps the posterior distribution for lVisC is positive when it is included in some models, but negative when it is included in others? You could look at individual models to find out! If you plot the posterior I would predict a bimodal shape. If this hunch is true, it would not be appropriate to summary it with a single mean an standard deviation...

Interesting case.

Cheers,

E.J.