Can your manipulation condition be a predictor in a multiple regression? I recently ran an experiment which manipulated participants in one of two conditions, by either promoting similarities between two groups, or emphasizing differences between two groups.
I then ran a multiple regression to find out what predicts the amount of tax that a participant would be willing to pay for another group. 
As the analysis was exploratory, I added all of the variables including my nominal manipulation condition (group 1 = promote similarities, group 2 = emphasize differences).
The analysis determined that Symbolic threat (threat felt from cultural values of others) and condition (which manipulation was assigned to participant, i.e which group individual belonged to) predicted the DV (amount will pay for other group in tax).
I was just wondering, is it weird and unusual to add your manipulation condition as a predictor in a multiple regression? I made a scale which was supposed to measure the manipulation and it differed significantly between groups. However, when doing a MR this is not included as a predictor... When using the manipulation condition (group 1 or 2) it is included. I don't really understand why this is. Any advice would be really appreciated.
Thank you for taking the time to read this, and for any helpful comments.
Richard
 A: It would be weird and unusual not to include the manipulation condition. But you might want to include it's interaction with the other predictors as well. Your current model holds the slopes constant across the manipulation while an interaction would allow them to vary separately.
It sounds like you're used to using regression to assess relationships among measured variables so maybe this will make it easier.  What if, in collecting your data, you had discovered that one group came from Unityland and culturally tended to promote similarities while the other came from Dissensionland and promoted differences. Wouldn't you include the country of origin as a categorical predictor? The main difference between that and what you're doing is that now your manipulation is a true experimental predictor (assuming a few other things) and substantially better than just happening upon it.  The same could be said of your other predictors in your experiment that you haven't manipulated but in reverse. What if you manipulated them?
