More of a Bayesian that a Stan question, but it’s something I think about when using Stan.
In a lot of cases where I have a large dataset, scaling the predictors (using the 2 sd method as prescribed by Gelman) often makes sampling go much faster. When I do this, I tend to stick with a “weakly-informative” prior such as normal(0,1) for my parameters. This typically ends up working fine, but try not to default to basic priors whenever possible. I was wondering if there was some sort of guide similar to the Steven Miller article linked below on how to incorporate prior information for scaled parameters. Hopefully this isn’t too basic of a question, just something I think about a lot and haven’t found a great answer on.
Thank you.
Miller blog post: How Should You Think About Your Priors for a Bayesian Analysis? | Steven V. Miller