[R-sig-ME] Error in mer_finalize(ans, verbose): q = xxx > n = yyy
David_Hinds at perlegen.com
Tue Sep 9 19:31:17 CEST 2008
Could this error be changed to just a warning, or could we have an option to skip this test? I'm using lme4 to fit binomial mixed effects models that do have reasonable solutions but are tripped up by this test. I'm working around the problem for now by commenting it out.
In my case part of the problem is use of a "trick" I saw on this list, to allow for subgroup-specific random effects, i.e.:
y ~ x + (0+x.1|z) + (0+x.2|z) + ...
where x.1, x.2, etc are indicators for levels of x. Here, most of the random effects are fixed at 0 but they still count when evaluating whether the number of random effects exceeds the number of observations.
-- David Hinds
More information about the R-sig-mixed-models