Interpreting Main Effects across alternative interactions in moderated multiple regression (MMR)

RPM

New Member
#1
I am conducting regressions with two sets of categorical predictors (and also several covariates, but these are not relevant to my question). My predictors are: politics (Left, Center, Right) and attitude prime (endorse, neutral, oppose). Politics is transformed into two dummy variables (Left vs. Center; Right vs. Center) and attitude is also transformed into two dummy variables (endorse vs. neutral; oppose vs. neutral). I am running each dummy variable combination separately, e.g. [Left vs. Neutral] X [Endorse vs. Neutral], [Left vs Neutral] X [Oppose vs. Neutral], etc. yielding 4 separate regressions. The regressions are structured hierarchically, with covariates first, then politics dummy, then attitude prime dummy.

Here's my question: I find that the main effects of one dummy variable shift slightly across alternating values of the other dummy variable. For example, politics dummy1 has slightly different Bs and p values when run with attitude dummy 1 (endorse vs. neutral) compared to attitude dummy 2 (oppose vs. neutral).

I could "fix" this by running all dummy variables and all interactions simultaneously in a single regression, but I want to keep each analysis distinct.

Advice would be greatly appreciated.
 
Last edited: