estimated marginal means in GLM

[<Raffaella.Milani@tvu.ac.uk>]

Can anyone help me to understand why the estimated marginal means in
Univariate General Linear Model (calculated by SPSS) change if I include
an interaction in the model? I understand that they adjust to
covariates, but why inserting an interaction should change the estimated
marginal means and if so which means should I use and on what basis?
Thanks a lot for your help. 
 
See the two sintaxes below
 
UNIANOVA

  gsi1  BY dependence stress  
  /METHOD = SSTYPE(3)
  /INTERCEPT = INCLUDE
  /PLOT = PROFILE( stress*dependence dependence*stress )
  /EMMEANS = TABLES(dependence) 
  /EMMEANS = TABLES(stress) 
  /EMMEANS = TABLES(dependence*stress) 
  /PRINT = DESCRIPTIVE PARAMETER
  /CRITERIA = ALPHA(.05)
  /DESIGN = depend1 stress dependence*stress .

 
UNIANOVA

  gsi1  BY dependence stress 
  /METHOD = SSTYPE(3)
  /INTERCEPT = INCLUDE
  /PLOT = PROFILE( stress*dependence dependence*stress )
  /EMMEANS = TABLES(dependence) 
  /EMMEANS = TABLES(stress) 
  /EMMEANS = TABLES(dependence*stress) 
  /PRINT = DESCRIPTIVE PARAMETER
  /CRITERIA = ALPHA(.05)
  /DESIGN = dependence stress .