Dear all,
I am doing Parametric Empirical Bayes (BMA = spm_dcm_peb_bmc(PEB)) according to Zeidman et al. (2019) for the first time. I have one between subject factor and two groups (HC and patients). Now I noticed that I get very different results connectivities after Bayesian Model Averaging depending on whether I have a
(1) first vector of 1s and a second, 'dummy' regressors of 0s and 1s OR
(2) again a first regressor of 1s and a second regressor of -1s and 1s.
I realise that the interpretation of the coefficient is somewhat different depending on whether I use dummy or effect coding (as in all GLMs). However, I get very different connections with sufficient evidence after Bayesian Model Averaging. I was wondering whether you could advice, which approach is most suitable in my case where I am basically looking for difference in connectivity between two groups?
Thank you very much in advance!
Best,
Susan
|