[R-sig-ME] glmer conditional deviance DECREASING after removal of fixed effect??

[Juho Kristian Ruohonen]

Hi,

I thought it was impossible for deviance to decrease when a term is
removed!?!? Yet I'm seeing it happen with this pair of relatively simple
Bernoulli GLMMs fit using lme4:glmer():

> full <- glmer(y ~ (1|id) + x1 + x2 + x3 + x4, family = binomial, nAGQ =
> 6, data = anon)

> reduced <- update(full, ~. -x1)

> c(full = deviance(full), reduced = deviance(reduced))


*     full  reduced*
*2808.671 2807.374 *

What on earth going on? FYI, I am deliberately comparing conditional
deviances rather than marginal ones, because quite a few of the clusters
are of inherent interest and likely to recur in future data.

My anonymized datafile is attached.

Best,

Juho

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: anon.txt
URL: <https://stat.ethz.ch/pipermail/r-sig-mixed-models/attachments/20230831/bca7eb87/attachment-0001.txt>