Dear colleagues,
I am trying to conduct several moderation analyses with diverse dichotomous variables as DV and moderator variables using the macro PROCESS for SPSS. In some places, I have read that it is recommended to code every dichotomous variable of the pathway as -0.5 and 0.5. However, in other places, I have read they can be coded as dummy variables (0 and 1).
Does anyone know how dichotomous variables must be coded when working with moderation analysis in PROCESS?
Thank you in advance!
Hello Sergio,
you don't need the macro for moderatated regression.
1) if the DV is a binary variable, then use logistic regression but be aware that visualization needs a special approach:
Ai, C., & Norton, E. C. (2003). Interaction terms in logit and probit models. Economics Letters, 80, 123-129. https://doi.org/10.1016/S0165-1765(03)00032-6
2) If the moderator is binary, just multiply it as it is with your predictor. Of course, you can also recode it. That makes no difference but only determines the meaning of the first order effect of the predictor in the model (which is defined as "the effect of X when Z = 0"). Hence, a unchanged moderator (0/1) will cause the first order effect of X being "the effect of X in the reference group of the moderator". If find this intuitive.
Frazier, P. A., Barron, K. E., & Tix, A. P. (2004). Testing moderator and mediator effects in counseling psychology. Journal of Counseling Psychology, 51(1), 115-134
HTH
Holger
I agree that you do not need PROCESS, but it will detect binary DVs automatically and will calculate a logistic regression model, accordingly. And as Holger already said, the coding system does not matter for the statistical analysis itself, but only for the intepretation of the coefficients. Therefore, you should precisely know what you are doing and how to interpret it, e.g. dummy coding vs. effect coding.
When conducting moderation analyses using the macro PROCESS for SPSS, it is recommended to code dichotomous variables as dummy variables (0 and 1) rather than -0.5 and 0.5. This is because dummy coding allows for the estimation of an intercept, which is necessary for the interpretation of the moderation effect. Additionally, dummy coding is more commonly used in statistical analyses and is more easily interpretable by others who may be reviewing your work.
Hi Dinesh,
centering only matters for the interpretation of the "first-order effects"--that is the parameter estimates of x in presence of the product term. These are the effects of X when the moderator = 0. In case of a binary moderator D, using the raw variables has nice implications, as the intercept is the mean of Y for the reference group (D=0) and X = 0 (which, if you center x, will be the mean of the reference group with average X). The first order effect of x will be the effect of X in the reference group (instead of the effect of X for mean D, which makes no sense).
In contrast, the test of the moderation effect (i.e., the coefficient of the product term and its significance) does not depend on centering or scaling which I proofed with a small simulation here:
https://www.researchgate.net/post/Can-I-combine-two-moderating-models-into-1-in-AMOS#view=5f9bc96570ea116d6a091ca7
HTH
Holger
- Badges
- Science topic
- Similar topics
- Social Science
- Media
- Social Media