Skip to content

lme4 - Mixed Model question

2 messages · Axel Urbiz, Ben Bolker

Original
Axel Urbiz
# 
Dear List, 

I?d appreciate any guidance on the following. 

I?m using a mixed effects logistic regression model, to allow coefficients to vary by a group variable. However, my case is not typical in the sense that I need to specify a different set of covariates for each level of the group variable. Say I have 3 covariates {x1, x2, x3} and 2 groups {g1, g2}. I want to specify a model for g1 that only depends on x1 and x2, and a model for g2 that only depends on x2 and x3. 

Note that the 2 groups is just for illustration. I actually have many more than that. 

Is this possible with lme4?

Thanks,
Axel.
Ben Bolker
# 
?? I *think* that if you simply set the values of the covariates you 
want to disregard for each group to 0 for that group, you'll get what 
you want.? I don't think there's a way to do that in the formula 
specification; you can obviously write code to do it, but doing it 
elegantly could be tricky.? If I were doing it I might try something 
*approximately* along the lines of

 ?? vars <- list(g1=c("x1","x2"), g2=c("x2","x3"))

 ?? all_vars <- unique(unlist(vars))

 ?? for (i in seq_along(models)) {

 ????? my_data[group==names(models)[i], setdiff(all_vars, vars[[i]])] <- 0

 ?? }
On 7/13/20 12:06 PM, Axel Urbiz wrote: