Summary
Sometimes it would be useful to be able to construct confidence intervals for e.g. visit specific variance estimates. In principle, this is possible today already, by combining the diag(fit$cov) point estimates with standard errors that could be derived from fit$theta_vcov and the known mapping from theta to the covariance matrix entries. However, that is quite cumbersome, and the user would need to do it from scratch.
Therefore it would be great to return an (approximate) covariance matrix for the the covariance matrix estimates themselves as part of the model fit object.
Additional Information
- We could check how
glmmTMB is doing it (or not)
- In principle we should be able to use the automatic differentiation also for this purpose, would need to write down the details
Summary
Sometimes it would be useful to be able to construct confidence intervals for e.g. visit specific variance estimates. In principle, this is possible today already, by combining the
diag(fit$cov)point estimates with standard errors that could be derived fromfit$theta_vcovand the known mapping fromthetato the covariance matrix entries. However, that is quite cumbersome, and the user would need to do it from scratch.Therefore it would be great to return an (approximate) covariance matrix for the the covariance matrix estimates themselves as part of the model fit object.
Additional Information
glmmTMBis doing it (or not)