Dear Dr. Bolker
I appreciate your kind help.
I can identify the version of packages at that time.
lme4 was 1.1-23 and lmerTest was 3.1-2.
For reference, are there any other packages that might affect the results?
I have tried, but it seems that other packages would affect them.
Very sorry to bother you while you are busy, but I appreciate your
cooperation.
2021?8?27?(?) 6:58 Ben Bolker <bbolker at gmail.com>:
[Please keep r-sig-mixed-models at r-project.org in the Cc: list!]
The warning message is *probably* irrelevant; to make it go away you
would need to install the Matrix package from source on your machine
(the binary packages on CRAN are built with the latest R version of a
particular release series).
A preliminary investigation on my side had the same general
conclusions (fitting with 1.1-23 and 1.1-26 gave identical answers). I'm
afraid that if you have tried with both versions of the package and
can't replicate your former results exactly then, logically, *the
lme4/lmerTest package versions cannot be the source of the problem*;
some other aspect of your setup must have changed. Numerical
computations are, unfortunately, subject to a great deal of
(small-scale) instability due to changes in underlying linear algebra
packages, compilers, etc..
I'm not sure what else we can do to help.
sincerely
Ben Bolker