Unfortunately, this is not an answer. But I am also interested in a marginal PRESS statistic for mixed models and also a margnial RSquared. If the random effect comes from a block effect, it is really interesting to see how much variance can be explained by the actual input parameters. I have models where the RSquared equals almost one but the variance explained by the random effect is very large and therefore my "actual" input parameters do not influence the output very much. And I want to quantify how much to investigate whether my model is suitable for predictions.
I think something like that could be implemented in JSL, but not really easy. @lazzybug I would be glad to hear if you found an answer to your question!
