Estimating prior sd for the parameter p of a beta binomial regression

That is what I am doing at the moment, since I am new to Bayesian modelling I was not sure it was a valid option because all the examples I have seen used values taken directly from the prior.