Created attachment 2072 [details]
nlme analysis of redundant data, with noise titrated to near zero
On certain pathological data sets, where the data is nearly redundant, lme returns nonsensical answers (which is understandable), but does so without producing any warnings (which is worrisome).
I would like to see warnings when I encounter singularities, so that I can be aware that results might not be reliable.
I attach a simple, complete, example. There is more discussion of the issue at https://htmlpreview.github.io/?https://github.com/dushoff/scratch/blob/master/nlme.html