Potential scale reduction (Rhat, imputation phase) - What does psi implies?
Closed this issue · 2 comments
Hello,
I am not sure if this is the right place to asks questions. If not, please excuse me.
I am currently using the package mitml to impute multilevel data with missings on two levels and I have a question about the summary. I was wondering if one can only try to increase convergence by increasing n.burn or if it can also cohere with the imputation model itself. However, I am not able to grasp the meaning of the values of the potential scale reduction, especially psi. Can someone maybe help me with this?
Many thanks
The different rows in the summary() output refer to different types of parameters in the imputation model, where Psi includes the variances and covariances of the random effects. Convergence tends to improve when the number of iterations is increased (i.e., by increasing n.burn or n.iter or both). However, poor convergence (i.e., large values for "R-hat") can also be a sign of model misspecification (e.g., when level-2 variables are included in the level-1 model). So if increasing the number of iterations does not help, the issue is likely with the model or data.
No further response, so I'm marking this as closed (but can open again if needed).