We are working on a concurrent validity study in which we want to calculate the validity of 2 different language measures taken from young children, and compare this validity between groups of children (autistic and non-autistic), while accounting for some other demographic variables and the nested structure of the data.
Having read the blue-tit vignette in brms documentation, we see parallels between that setup and our question: in the blue-tit vignette the model gives the contribution of two factors, genetic and environmental, averaged over the different levels of those factors (ie individual mother birds and foster nests) on the two outcomes, and we are interested rather in the contribution of the two levels of one factor (factor: group, levels: autistic and non-autistic).
We were hoping some forum users might be able to give us some feedback on whether our model structure makes sense. For the moment we are thinking of using ICC/Generalisability Theory to derive random intercepts per participant, as an indicator of their true language ability, as well as calculate the amount of variance apportioned to different sources in this model
model1 <- lmer(score ~ (1|P) + (1|test) + (1|group:test))
and then use the individual participants’ intercepts in this brms model
model2 <- brm (mvbind(indivInterceptsTest1, indivInterceptsTest2) ~ socioeconomicStatus + bilingualStatus + (1|p|group), data = data, chains = 2, cores = 2)
but we’re not sure that this will give us the desired outcome, that is, the difference between the two groups in the strength of the relationship between the language ability metric derived from test 1 and test 2.
We appreciate your time and thanks for any feedback!