[R-sig-ME] Computing pair-wise associations of fixed effects in gLMM
Phillip Alday
ph||||p@@|d@y @end|ng |rom mp|@n|
Mon May 25 00:58:54 CEST 2020
I'm a bit confused by your question -- you "suddenly" introduce multiple
response variables but don't describe what they represent. This is just
as important as describing your predictors! Otherwise we have no way of
knowing where e.g. the Poisson distribution is a reasonable assumption.
Also, note that really shouldn't test normality of your response
variable for two reasons. First, as the size of your data increases, it
becomes easier and easier to reject the null hypothesis of normality for
trivial reasons. No real data is perfectly normally distributed, even
data generated from a normal distribution and so the test rejects more
and more data that really would be fine. Second, it's not the "absolute"
(or more precisely, marginal) distribution of your data that matters,
but rather the distribution of the residuals (or equivalently, the
conditional distribution).
I'm also not clear what you mean with pairwise correlation of
categorical predictors -- do you mean the correlation of fixed-effects
estimates? lme4 will give you that information, but I don't know if
that's what you're looking for. Are you looking for how much the effects
of the different (levels of the different) factors correlate with each
other?
That doesn't yet help you that much, but if you clarify a bit, maybe we
can help you more! :)
Best,
Phillip
On 5/5/20 1:47 pm, Julian Gaviria Lopez wrote:
> Dear list members.
>
>
> I have a nested data comprised by 2 factors (conditions: A and B). Each factor has 8 levels: (clusters: c1,c2,c3,c4,c5,c6,c7,c8). N=33. Aim: To assess the pairwise association between the factors (i.e. correlation between Ac1 and Bc1, etc.). Although an LMM will count for the dependent nature of the data (repeated measures of the 33 participants observed in condition A, and consecutively in B), some of the dependent variables are not normally distributed (7 out of 16) according to the shapiro test. For this reason, I think a gLMM might be a good option:
>
>
> M <-glmer(observation~condition+cluster+(1|subject),data=mDATA,family="poisson")
>
>
> Questions:
>
>
> 1) Would anyone is aware of a better option regarding the modelling method?
>
> 2) In case gLMM is the "right" way to go, I wonder how could I compute the pairwise correlations of the "fixed effects" (e.g. Ac1-Bc1; Ac1-Bc2; ... Ac1-Bc8), with "glmer" function, or maybe with the glmmTMB?
>
>
>
> Thanks in advance
>
>
>
> Julian Gaviria
> Neurology and Imaging of cognition lab (Labnic)
> University of Geneva. Campus Biotech.
> 9 Chemin des Mines, 1202 Geneva, CH
> Tel: +41 22 379 0380
> Email: Julian.GaviriaLopez using unige.ch
>
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
[[alternative HTML version deleted]]
More information about the R-sig-mixed-models
mailing list