I have more than one question, if you can help:
I have a regression model with 5 control variables each with 5 categories. I am wondering if there's any way, that can be also considered methodologicaly correct, to use them in a multiple regression analysis without transforming them into dummy coded variables? Problem is I forgotten to do this, already completed most of the analysis and now would have to take it from the beginning. After dummy coding and repeating the tests, there are differences in R change, they explain some extra 5% more and there is a small decrease in the proportion explained by the predictors.
Say I'm conducting a mediation analysis. From what I understand, reverse causality would mean my outcome predicts my mediator. How about the case when my mediator predicts my predictor, how would that be called? Theoretically, it is possible, I've tested both scenarios and they are significant. Should I report them both, would it be correct? (mediation tested in spss BK steps and Hayes macros).
I've used Hayes' macros to test the mediation. Problem is that using same set of variables but different macros gave me slighlty different results. I've read that it can be because of the sample size, because cases with missing values are deleted and won't be used in the analysis. But this doesn't seem the case. Why would the results be different if the macros are supposed to test the same mediation?
Any thoughts greatly welcomed.