Collinearity in mediation models

[DGrageda]

Hi everyone,

I was wondering if it's OK to run a mediation analysis on conceptually related variables with Pearsons correlations of .44 (between X to Y) and .65 (between X to M) .61 (between X to Y)? OR would collinearity be an issue that violates this mediation analysis? I found a significant mediation (as the 95% CI did not contain 0) and want to say that X causes Y in the presence of M. But am worried that collinearity is an issue, and if it is not an issue how do i justify this to someone who doesn't understand mediation analysis? Thank you, would really appreciate an answer or advise on where to read more about this so I can justify my model.

 

[spunky]

Not a problem. Multicollinearity is one of the biggest non-issues in the social sciences. Standard "rules of thumb" for when multicollinearity is a problem use the Variance Inflation Factor (VIF) as a guideline and you don't have "severe multicollinearity" until your VIF is above 5. But to get VIF greater than 5 you need correlations on the 0.8s or higher.

 

[DGrageda]

Not a problem. Multicollinearity is one of the biggest non-issues in the social sciences. Standard "rules of thumb" for when multicollinearity is a problem use the Variance Inflation Factor (VIF) as a guideline and you don't have "severe multicollinearity" until your VIF is above 5. But to get VIF greater than 5 you need correlations on the 0.8s or higher.

Thank you so much! Does the same apply to a moderation model/analysis?

 

[spunky]

Anything that uses Ordinary Least Squares (OLS) linear regression. And the engine behind those two methods is OLS regression so...yup.