# Treatment mean comparisons in a mixed model with random and fixed effects

I have run a mixed model with one random effect, one fixed effect and a second order covariate effect. The random effect models the individual experimental units in my study. The LSMeans calculated for each experimental unit from the full model are quite different from the effects calculated running an individual model for each experimental unit by itself using only the fixed effect and second order covariate in the model.

Is this to be expected? If so, why are the full model effects so different from the effects calculated with one experimental unit at a time?

• Conclusions drawn from the mixed model are intended to be generalised to the population from which your experimental units come from. Conclusions drawn from the "individual model" are only valid for one specific unit. It is a little bit like comparing some average $\bar{x}$ with one single observation $x_i$. – ocram Dec 11 '12 at 18:52