Gelman’s paper on EP says there are 2(D+1) shared parameters for a model with D explanatory variables. By looking at sample code m1a.py, apparently the two extra shared parameters are sigma (std dev for y), and sigma_a (std dev for alpha, which is the intercept). I am not sure why then is sigma_b, the std dev for beta, not a shared param. If someone could share some wisdom on this, I would appreciate it.
The inclusion of the intercept term does introduce a huge inconvenience. In my case the model involves two sets of betas (and corresponding predictors), which means the estimate of 2(D + 1) is broken. I now have two intercept terms to introduce and will need to make adjustments to the mu_phi, omega_phi and dphi. Would someone confirm my understanding is correct.