What do I do with BF10 = 4.698e+13, error%=100?

eniseg2

Hello everyone,

I'm looking at a model predicting protein load in the hippocampus (post-mortem). This variable is ordinal and probably shouldn't be used in a Bayesian ANCOVA (compare to null model, BF10) but I tried anyway. I'm trying to find the best model which combined the dementia status and the result of four memory tests (four variables) plus the interactions.

JASP has given me larger BF10 than I have ever seen, my analysis of effects for inclusion contains three infinities. The error rate for all models, however, is really high. What do I do if the model error rate reaches 100%? Surely that means the model is garbage? Our sample size is <30.

Kind regards and thank you for any input,

Anna

EJ

Hi Anna,

If you have a massive BF you can tolerate a high error rate. Does it matter for your conclusions whether the BF is one trillion or two trillion in favor of the alternative hypothesis? Probably not. BFs this high will have a lot of action in the tails of the distributions, and this lowers the numerical accuracy. So nothing to worry about. If you want to lower the error rate you can go to "Advanced options" and increase the number of samples.

Cheers,

E.J.