Mixed models and estimated marginal means

[Stephen Cox-4]

Folks,
 
I have run a model looking at the difference in reported work hours for employees working in 9 different Positions. The data was collected across 51 different work sites, so I have clustered data. I ran a Mixed model, specifying work site as a 2nd level variable on the Random command line.
 
I asked for estimated marginal means for each group, plus the overall mean. These are different from the descriptive means provided. I thought that taking such clustering into account would change variances and standard errors, but not the point estimates.
 
I have not been able to find out how SPSS calculates the EMMs in this context of correlated data. Obviously these site level correlations are being taken into account, but I am not sure how they are calculated, nor how I should interpret them.
 
Can anyone else assist, or tell me where to go for information (preferably that isn't just matrix algebra).
 
Cheers,
 
Stephen Cox
QUT.
 
 

[jmills]

It seems this question has gone unanswered here before... but can anyone explain how the estimated marginal means are calculated in the "Mixed" procedure?  

I'm trying to examine differences in the means for four different groups-- but I have longitudinal data (with some missing values), so I believe the linear mixed model is the correct approach.  However, the EMM are exhibiting an opposite trend from the observed means.  Without knowing how the EMM are calculated, I can't decide whether this is something I'm doing wrong or the real result of taking time into account.

Thoughts?