Variance structures for residuals - brms

Please also provide the following information in addition to your question:

  • Operating System: debian 9
  • brms Version: brms_2.8.0

Hi all,

Just trying brms for the first time so sorry if I’m asking something ignorant.

I wanted to know if there was a way to specify the variance structure of the residuals. I work on clonal organisms that I measure in different environments and treat the response variable as two separate responses (one for each environment in a 2 environment case).

At the group level I’m able to estimate the covariance between responses but as no individual organism is measured in both environments I can’t estimate the residual covariance.

When using programs such as asreml-R or MCMCglmm, I can use the idh() variance structure that estimates seperate residual variances for each response but not the covariance.

Is there an equivalent in brms?


In brms, you can either fit the full residual correlation matrix (available for a few families only currently) or assume residual correlations to be zero. You can control this by using the set_rescor function.

Thanks for the tip Paul.