I am analyzing a repeated measures experiment with a multilevel growth curve in SPSS. I want to control for a potential confound (baseline social support) when looking at differences in depressive symptoms between my groups. My measure of social support is a level-2, time-invariant variable (only measured at baseline). To control for this variable, is it sufficient to just enter it as a main effect in the fixed effects portion of the model? Or do I also need to enter a time*social support interaction to control for its influence on depressive symptoms across all timepoints?
You can control for social support by including it as a predictor in the fixed effect part of your model. Then all other coefficients are telling you about outcome differences when comparing two individuals with the same value of social support.
The issue of a time by social support interaction addresses a different issue, which you alluded to. Such an interaction would indicate whether changes in your outcome are at all different for individuals with higher/lower social support. This could serve a similar covariate role in the case that you were interacting other level 2 variables with time. This is consistent with Raudenbush & Bryk's notion of a "slopes as outcome" model.
Most folks would likely do the first (social support as predictor) and only do the second (interaction) if they had a hypothesis about the moderating role of social support on outcome changes.