5
$\begingroup$

Is there a way to specify two different random intercept terms for two sub-groups of subjects within a single mixed model? The reason why I'm interested is that it seems that the inter-personal variability of one group of subjects on my task is greater than the inter-personal variability of the other. I'm using lmer and currently specifying the subject factor as a random intercept, which pools all my participants into a single estimate, but I would like to have two separate estimates for the two sub-groups.

$\endgroup$
5
  • 2
    $\begingroup$ Would a random slope in the group indicator do the job? $\endgroup$
    – Macro
    Mar 13, 2013 at 14:54
  • $\begingroup$ It seems that it would not because I'm getting information about the average performance difference between the groups if I specify Group as a fixed factor. If I put Group in a random factor (1|Group), I don't seem to get an answer about the variability of group intercepts from the lmer summary. $\endgroup$ Mar 13, 2013 at 19:04
  • 5
    $\begingroup$ I'm assuming you have some grouping in the data (e.g. repeated measurements on people) other than the two subgroups you were talking about. Call that grouping ID and call the binary subgroup indicator G. Then, what I was suggesting was something like putting a (1+G|ID) term in your model. $\endgroup$
    – Macro
    Mar 13, 2013 at 21:38
  • 1
    $\begingroup$ And then the variance of the slope estimate would be indicative of the differences between the variances between the groups? $\endgroup$ Mar 14, 2013 at 2:02
  • $\begingroup$ Yes, that's right. $\endgroup$
    – Macro
    Mar 14, 2013 at 2:09

1 Answer 1

2
$\begingroup$

Using the "weights" argument in R's nlme you can allow for separate variances for each level of a grouping variable (heterogeneous variability).

For examples see:

https://quantdev.ssri.psu.edu/sites/qdev/files/ILD_Ch06_2017_MLMwithHeterogeneousVariance.html https://fukamilab.github.io/BIO202/03-C-heterogeneity.html

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.