Dear List,
I am trying to implement Bayesian analysis in my research due to potential advantages such as direct inference with posterior estimates and possibility to incorpore prior knowledge when available.
In my frequentist background when dealing with multiple covariates for explaining my dependent variable we used at least in my area of animal science:
-> univariable analysis with screening of potentially interesting variables retaining variables with a potential interest after adjusting for other (let say P less than a threshold commonly 0.2...0
-> then do a manual stepwise procedure with all the covariates kept at the 1st step and reanalysing the model until all the remaining are <0.05 or any adjusted P<value.
I know this approach has many pitfalls (multiple comparisons and risk of overfitting).
My question was on the corresponding approach in a Bayesian framework.
is there any advices or SAS tutorial on how to perform multiple regression analysis with Proc MCMC and especially for dealing with variables that are crossing the no-effect cut-off and final variable selection?
