How to decide between individual versus group level random effect

Question

I have N participants each has inconsistent number of repeated measurements of an outcome Y. Each participant belongs to one of two groups.
My hypothesis is that the levels of Y differ significantly among groups and that the Y rate of change also varies significantly over time.
Should the group be set as fixed or random effect?

Robert Long · Answer

Since you only have 2 groups and you are interested in the "effect" of the group, then it should be a fixed effect. In R it would look something like:
Y ~ time + group + (1|subject)

You might also want to fit an interaction between time and group and possibly allow the fixed effects to vary by subject, with random slopes.

How to decide between individual versus group level random effect

One Answer

Add your own answers!

Ask a Question