I have three Groups (Experimental 1, Experimental 2, Control); a
dependent variable Y, and a covariate X and three Races (White, Black,
Hispanic) in my data file.
I am running GLM treating Group and Race as fixed effects and using X as
the covariate. (I have 9 cells here: 3 by 3, as you see). I see that the
group by race interaction is significant.
What I want to do is to document the means (on Y, adjusted for X) for
each one of the 9 cells and plot them for different values of X (at
X=x1, x2, x3...etc)
There are 2 possible approaches as far as I can see:
1. Run separate simply regression models for each one of the 9
cells (leaving Group and Race out and using only X in model). Obtain
prediction equation this way and plot predicted (adjusted) Y at X=x1,
x2, x3 etc... for each cell separately (using their corresponding
2. Run 1 single GLM with Group and Race as fixed effects and X as
the covariate. Save predicted Y as part of GLM analysis and plot these
predicted Y values across X for each cell separately.
Are these two approaches identical? If not, which one is a better idea
and why? (Cell sizes range from 14 to 100)
The second option might sound like the answer but do I include the group
by race interaction in the GLM model or not in this analysis?