Bayesian modeling scheme

[shansfolder]

Just a remark for the future work (not part of this pull request):

• for a normal distribution, the returned effect size is already normalized
• for a Poisson distribution, we have to normalize it by hand

Maybe we should stick to one particular scheme and return a non-normalized effect size from the normal model (or the normalized from the Poisson model).