FAQ/reffs - CBU statistics Wiki

Revision 1 as of 2012-12-06 12:55:42

Clear message
location: FAQ / reffs

Effect size for multilevel models

$$\rho$$ = variance_{between subjects} / (variance{between subjects} + variance{within subjects})

I read in the wikipedia that this "design effect" is used with cluster observations. If we fit a mixed model with a random subject-specific intercept, the clusters are the observations within a participant e. g. the 7 times the participant chose a fruit or a snack. The "design effect" is

D_{eff} = 1 + (m-1) $$\rho$$

where m is the number of observations in each cluster (e.g. number of repeated measures per subject).