[R-sig-ME] test_terms for lme model with crossed random effects and weights

[Guillaume Adeux]

Hi everyone,

I wanted to allow for different variances across treatment levels so I
headed towards nlme:lme in order to use varIdent().

However, I also have crossed random effects (year and plot) which needs a
little hack in nlme:lme to specify.

Hence the final model is:

dat$Grp=1 ### fake grouping in order to create grouped data object
gr=groupedData(response~time|Grp,data=dat) ### grouped data object
mod4 <- lme(response~ block+treatment*time, random =
pdBlocked(list(pdIdent(~ year - 1),pdIdent(~ plot -
1))),weights=varIdent(form=~1|treatment), data=gr,method="ML")

The question now is: is it possible to supply such a model to
monet::test_terms() ?
I cannot find such examples in
https://rdrr.io/github/singmann/monet/src/examples/examples.test_terms.R

Thank you all once again for you help.

Sincerely,

Guillaume ADEUX

Nb: if I understood correctly from previous messages, this type of model
(crossed random effects, specific variance structure) could also be fitted
with glmmTMB

[[alternative HTML version deleted]]

[Maarten Jung]

I don't know if monet::test_terms() can do this.
But maybe the emmeans package can help here. The joint_tests() function
computes Type-III test for each effect in a model [1].
Also (what I find quite interesting), "[t]here is an experimental mode =
"satterthwaite" option that determines degrees of freedom approximately: It
estimates a needed gradient in the covariance matrix experimentally by
randomly perturbing the response values. Thus, the degrees of freedom will
vary slightly (or possibly even a lot) if the reference grid is
re-calculated" [2].

Cheers, Maarten

[1]
https://cran.r-project.org/web/packages/emmeans/vignettes/interactions.html
[2] https://cran.r-project.org/web/packages/emmeans/vignettes/models.html#K


On Tue, Oct 23, 2018 at 12:06 PM Guillaume Adeux <

Post by Guillaume Adeux
Hi everyone,
I wanted to allow for different variances across treatment levels so I
headed towards nlme:lme in order to use varIdent().
However, I also have crossed random effects (year and plot) which needs a
little hack in nlme:lme to specify.
dat$Grp=1 ### fake grouping in order to create grouped data object
gr=groupedData(response~time|Grp,data=dat) ### grouped data object
mod4 <- lme(response~ block+treatment*time, random =
pdBlocked(list(pdIdent(~ year - 1),pdIdent(~ plot -
1))),weights=varIdent(form=~1|treatment), data=gr,method="ML")
The question now is: is it possible to supply such a model to
monet::test_terms() ?
I cannot find such examples in
https://rdrr.io/github/singmann/monet/src/examples/examples.test_terms.R
Thank you all once again for you help.
Sincerely,
Guillaume ADEUX
Nb: if I understood correctly from previous messages, this type of model
(crossed random effects, specific variance structure) could also be fitted
with glmmTMB
[[alternative HTML version deleted]]
_______________________________________________
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

[[alternative HTML version deleted]]