Group random effects for two between-subject factors

Interfaces brms
,
1

Hello!

I’m aware of the usage of gr() to estimate seperate variance-covariance matrices for each level of a between-subject factor as per here. Using the third example there,

fit3 <- brm(count ~ Trt + (1|gr(patient, by = Trt)), data = epilepsy)

What if I have another between-subject variable (say, COND with 2 levels) that I want to include in the ‘by’ argument so that I can estimate four unique variance-covariance matrices (i.e., Trt0+COND1, Trt0+COND2, Trt1+COND1, and Trt1+COND2). Would it just be the below?

fit3 <- brm(count ~ Trt + (1|gr(patient, by = Trt + COND)), data = epilepsy)

Thank you!

2

I suggest you add a new variable with four levels to your dataset, combining the levels of Trt and COND, which you can then pass to the by argument.

3

Thanks Paul, was thinking along those lines so good to know that is what you’d suggest. Leading on from that, I was curious to know your thoughts on whether it would be fine to do the model call below,

fit3 <- brm(count ~ Trt * COND + (1|gr(patient, by = Trt_COND)), data = epilepsy)

Or does it need to be,

fit3 <- brm(count ~ Trt_COND + (1|gr(patient, by = Trt_COND)), data = epilepsy)