Easy Bayesian Statistics

hlsmith

Omega Contributor
#1
Bayesian Statistics Easy?

Well the thread title was to get your attention. About a month ago I attended a workshop on Bayesian statistics. It introduced the basic principles and gave regression examples and then loosely mentioned its use in missing data and sensitivity analysis for bias corrections.


I am going to focus on the regression part here. After playing around in the Bayesian area for a few weeks, I came to the conclusion it was a good time in my life to learn these approaches, given I had most of the mainstream stats figured out.


So now I am really surprised by how easy the Bayesian process seems: give priors; calculate likelihoods; then get posteriors.


So I am taking this time to confirm my understanding on the following. I can just use this approach all of the time now with informative and/or uninformative priors. So if I am running a regression and I don't want to state informative priors, I can just use flat priors and get estimates like before. THOUGH NOW, I can interpret p-values and say that their is a blank probability of result given null is true and say I am 95% confident the true values is in the credible interval. Obviously given correct model specification, etc. After taking so long to memorize the frequentist interpretations, I just wanted to make sure I had the Bayesian interpretation right.
 

hlsmith

Omega Contributor
#2
So does anyone do descriptive statistics differently when using Bayesian inferential approaches? So, when initially reporting measures of central tendency and dispersion for variables do you just present mean or median still as with frequentist approaches? I am not sure how you would do them differently unless you used bootstrapping or another approach. I suppose as in Bayesian the data is fixed and the parameters are random so how would that effect reporting a mean with standard deviation in the initial descriptive statistics?


It seems like it is another example of how Bayes approaches are still coupled with frequentisti approaches, say like when doing Bayesian model diagnostics (e.g., Geweke diagnostics, Raftery-Lewis,Gelman-Rubin, etc.).
 
Last edited: