I am modelling the activity of animals (subjects). Each subject is measured for activity repeatedly (which requires a random intercept associated with subject). Subjects have also been measured in two environments (a and b). I specifically want to see if the activity level of a given subject in environment a correlates with its activity level in environment b.

In line with this really useful answer (How to specify uncorrelated random slopes in lmer() syntax?), I tried doing a log-likelyhood ratio test on these models:

full.model <- 
    lmer(activity ~ (1+environment|subject), family='binomial')
no.correlation.model <- 
    lmer(activity ~ (1|subject) + (environment-1|subject), family='binomial')

It appears that this code works nicely for gaussian data, but that it does not for binomial data. In this case, no.correlation.model has two random intercepts, a 'slope' associated with environment and a correlation.

Is there a way to test whether the correlation estimated in full.model is significant?

Here is a reproducible dataset.

activity <-rbinom(n=1000, size=1, prob=0.5)
subject <- rep(1:100,each=10)
environment <- rep(c('a', 'b'), 500)

Many thanks in advance!


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.