Hi!
I am working together with a colleague who has done a study with an amount of non significant findings. The journal she wants to publish in requests in these cases Bayes factors to evaluate these non-significant findings.

After a bit of reading I got to following which seams to do the work on these exercise data, where the hypothesis is that drug has an effect:
Code:
sysuse cancer, replace
logistic died drug age /*normal regression"
set seed 123567
bayes, saving(H1): logistic died drug age
est sto H1
bayes, saving(H0): logistic died age
est sto H0
bayesstats ic H0 H1, bayesfactor
First question: Is this a reasonable approach?

Yet when my colleague works on her data, she gets several very very large factors (like e^17) and our concern is that this is because the models don't converge and/or are strongly auto correlated. Basically we add
Code:
bayes:
in front of the logistic regressions, save the results and test them just as in the example above. For some of these models we get warnings about auto correlation which adds to our concern.
It is very difficult to find out whether we should adjust our models in the bayesian context, although all other results in the paper reported are from the "normal" logistic models. Can anyone shed some light on this if the goal is to find Bayes factor?
I am not at all familiar with bayesian statistics, so I apologize beforehand if I got it all mixed up.
Thanks!