1
$\begingroup$

I have trainied a mixed model that is overfitting. We used lme4 and got this warning boundary (singular) fit: see ?isSingular and sure enough running isSingular as per the lme4 manual retunrs TRUE. My model is fitting 45 variabels as fixed effects plus two random effects and a nested effect. So Obviously the model is to complex and there is colinearity. My boss doesn't want me to use regularization for reasons of interpretability, so that leaves me with variabel removal I guess. the lme4 manual recomended I run rePCA on the model so Isdid but I'm not sure how this may help me understand the problem. Also how could I get the PCs for the fixed effects and use those to help?

> rePCA(fml)
$`mri_info_deviceserialnumber:rel_family_id`
Standard deviations (1, .., p=1):
[1] 0.2939328

Rotation (n x k) = (1 x 1):
     [,1]
[1,]    1

$mri_info_deviceserialnumber
Standard deviations (1, .., p=1):
[1] 0

Rotation (n x k) = (1 x 1):
     [,1]
[1,]    1

attr(,"class")
[1] "prcomplist"
$\endgroup$
0

0

Browse other questions tagged or ask your own question.