I have discovered that standardized coefficients are not provided for moderation models, such as model 59 in the PROCESS macro. This is because the effects of each variable are conditional and cannot be adequately explained by a single standardized coefficient. PROCESS does not provide this functionality because it could lead to confusion. However, some research is providing standardized coefficients despite indicating that they have utilized the PROCESS macro’s model 59 (moderation). This raises the question of how this is possible. Should I also include β values in my research?
Cohen, Cohen, West, and Aiken (2003, pp. 282-283) describe the proper steps for calculating the standardized coefficients in moderated regression:
1) Calculate z-scores for Y, X1 and X2, i.e., zY, zX1 and zX2
2) Compute the product zX1*zX2
3) Run the moderated regression with zX1, zX2, and zX1*zX2 as predictors and zY as dependent variable
4) The unstandardized solution from this regression with z-scores gives you the proper standardized solution.
The default "standardized solution" in regression software should be ignored for moderated regression as it does not provide the proper standardized coefficient for the product term.
Reference
Cohen, J., Cohen, P., West, S. G. & Aiken, L. S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences. Mahwah, NJ: Erlbaum.
Christian Geiser Thank you for helpful answer. In studies that use moderated PROCESS model 59, do authors usually report standardized coefficients (β) or is it more common to present only unstandardized coefficients (B) along with simple‑slope plots?
I would think that unstandardized coefficients are more common in moderated regression and I would definitely report those. Whether or not standardized coefficients should also be reported depends on whether they add anything useful to the interpretation (i.e., whether they make it easier to interpret the effect(s)).
- Badges
- Science topic
- Similar topics
- Publishing