The estimated marginal mean for a group is the estimated population mean and the predicted value for ALL people in that group, given that they score average on the covariate. So for someone who scores average on the covariate and is in group 1, the predicted score is equal to the estimated marginal mean in group 1.

If you you are interested in the predicted scores based on the actual score on the covariate of each subject, you should look at the predicted scores that you found via the 'save predicted values' option.

Given a certain value on the covariate and the group number, you will always find the same predicted score. However, the *observed* scores of two subjects in the same group with the same value for the covariate do not need to be the same. So even amongst subjects in the same group and with the same value for the covariate, there can be variability in the observed scores.

The reason for the 'variability in the estimated marginal mean' (do you mean the standard error?) is that the estimated marginal mean that you found in your sample is just an estimate based on a single sample. In a different sample you will probably find a different estimated marginal mean. The standard error of the estimated marginal mean is the (estimated) standard deviation of the estimated marginal means we would find in an infinite number of different samples.

Is this an answer to your question?