Process Model and Covariates

#1
Hi all,

I am having difficulty with Andrew Hayes' Process tool and my mediation model. Is it acceptable to run a mediation analysis with a continuous predictor, outcome and mediator variable, but then have dichotomous categorical variables as covariates? I am attempting to control for whether a person smokes or not.

Also, is it acceptable to include one further continuous variables as a covariate and a multicategorical variable with three levels. I have dummy coded this variable whereby a serves as the reference variable and b and c are examined. From my knowledge (limited) of multiple regression, I think it should be acceptable to include both dummy coded variables in addition to the other variables as covariates but I am not 100%.
 
#4
I'm using SPSS and Hayes' Process Macro. It's proving really difficult to find any good resources about the effects of covariates and what is acceptable or not. I've purchased his 2013 book but I'm still struggling.
 

hlsmith

Less is more. Stay pure. Stay poor.
#5
Hmm. I haven't used those but had thought that there were many resources related to it, since we get questions on it about at least every other week on this site.
 
#6
Hmm. I haven't used those but had thought that there were many resources related to it, since we get questions on it about at least every other week on this site.
I know, I've been really surprised so far. From what I can tell, everything seems in order though there really isn't much out there on covariates. Then there's the issue of reporting the results on top of whether the covariates are acceptable in the first place.
 

spunky

Doesn't actually exist
#7
Adding covariates in the PROCESS macro shouldn't create any problems with your model. I don't use PROCESS myself except for basic teaching stuff but from what I have heard from other colleagues/friends that use it, you just need to be very careful with what goes in which boxes and especially how you code categorical predictors. To the best of my understanding, you'll need to switch to something other than dummy coding (like effects coding) if you want to get the right partial regression coefficients you want.