I have relatively complex model with a multicategorical IV, a mediator and three moderators.
I want to estimate the model in SPSS Process.
My question concerns which covariates to include in the model:
I have a list of collected data on variables with all kinds of scales (metric, ordinal, nominal).
Which kind of covariates can I include in the Process model without transforming anything or taking any other preparations? Only metric covariates?
What about ordinal and nominal covariates? Would I need to take any other measures before including them as covariates (Dummy Coding etc.)? Would that complicate the model substantially?
I'm somewhat of a beginner in statistics and thankful for every feedback!
Hello Marcus,
That's a lot of questions! Let's unpack a few of them.
0. Though you didn't ask, you might be better served using SEM software, given the model complexity and variety of data types. Mplus is frequently mentioned as an option when you have non-metric data, though other packages can also be used (such as lavaan in R).
1. According to Andrew Hayes' text (Ch 13), Introduction to mediation, moderation, and conditional process analysis: A regression-based approach, he specifies that categorical antecedent (your IV) variables are to be dummy coded.
2. Only metric covariates can be used without transforming/recasting into dummy variate sets (or equivalent). The lone exception to this would be dichotomous variables (e.g., sex, coded as 0, 1 or something similar).
3. Yes, of course; recasting non-metric variables into variate sets will make the model appear to be "busier," and will require more effort to explain once the analysis is complete.
Good luck with your work.
