Zero order correlation negative, Direct effect Positive

#1
Hi,

I am performing mediation analysis, and have first of all looked at the correlation between variables.

The zero order correlation between X and Y is positive and significant. However in the mediation analysis, the direct effect of X on Y (after holding the mediator constant) has a negative beta and is significant.

I am just trying to understand this - could the fact that the correlation between these variables is positive but the regression shows a negative beta value be because in the zero order correlation, you are not controlling for the mediator or other variables, and so on the surface it may seem that there is a positive relationship between these variables. However, the direct effect does control for the mediator, so reveals that a negative relationship between X and Y, suggesting that the mediator is very positive, and may 'mask' the negative relationship that actually exists between X and Y in the zero order correlations?

Your help is greatly appreciated!
 

Karabiner

TS Contributor
#2
Maybe you should do search for suppressor effect in multiple regression analysis. Seemingly this is what you are dealing with.

With kind regards

Karabiner