Covergence Problem

[cyhsieh-psy]

I'm not sure if this weill be covered. Is there a guideline to resolving the convergence problem when fitting models?

[cyhsieh-psy]

I think the answer to this question is partly covered by the discussion "calc.derivs = FALSE and REML = FALSE". The warning messages of nonconvergence that we enocunter may be trivial (e.g. Hessian matrix). The software and optimisers nowadays are pretty good at reaching convergence. Instead of paying too much attention to this matter, maybe we should be more conscious of issues like singularity (?) and model (mis)specification.