1

I had a question regarding how to best perform "sensitivity analysis" on a Beta prior used for a estimating a proportion from a binomially distributed data problem (i.e., beta-binomial problem).

I understand that the shape parameters $\alpha$ and $\beta$ move the Beta to right and left. But how much and on what basis should I change these two parameters and compare their effects on the the posterior 95% credible Intervals (picture below)?

(Note: In the picture below, 'A' parameter denotes 'alpha')

enter image description here

rnorouzian
  • 3,986
  • You can use a uniform prior, which is the same as a Beta(1,1) distribution. Or you can do the empirical Bayes, like here. – dimitriy Dec 12 '17 at 02:38

0 Answers0