Is there any way to constrain eta in NLME or does that violate some aspect of the diff solver? I sometimes get a subject with an eta that is magnitudes too large.
Hi William, you can always use a transformation that constrains the parameter space to say [a-b] instead of [-inf,+inf].
But you need also to see why the individual subject data is giving you this very far ETA.
Having constraints on the ETA distribution is a departure from the normality or lognormality of your parameter distribution
In simulation mode you can reject an eta and keep simulating but this require elaborate simulate statement
e.g. https://github.com/metrumresearchgroup/mrgsolve/issues/53mrgsolve has this functionality.
https://stats.stackexchange.com/questions/73885/mcmc-on-a-bounded-parameter-space