I’m fitting a large regularized multiple regression model with a hierarchical prior on the standard deviation of the regression slopes (tau). When the model finishes running, I get a warning about low bulk ESS. All of the fixed effects have good (> 4,000) bulk ESS. However, using ShinyStan, I found that the tau has high autocorrelation, hence that’s probably the cause of the low ESS.

How should I go about fixing this? Will running the model for more iterations (currently I’m using the default 2,000) suffice?

That’s the brute force solution if we can’t figure out anything else.

If you’re using a non-centered parameterization here, you might also just try a centered parameterization for this one parameter. If the data is super informative of the coefficients for each group the centered parameterization works better.

Is this code put together in a way this is easy to just try out? If not we can make diagnostic plots. I think pairs plots of the group mean + individual effects is what you’d go for, but group mean in anything but the simplest models is a little sketchy to define I guess.