Good day! I am trying to see if Variable C moderates the relationship between my Variable A and B. I understand that this model falls under Model 1 of the Process Macro available for SPSS and so I have used that to analyze my data so far.
What I've found particularly striking is that the breakdown of results from that analysis doesn't seem to match the preliminary linear regression results I conducted between Variable A and B. According to the linear regression, A predicts B but according to the breakdown in the model, the effects of A on B are insignificant.
Attached are some photos of the results. What am I not understanding right about this? Does the linear regression actually not measure the same thing as that part of the moderation breakdown?
Obviously, everything (difference between the two models) goes back to your C Variable.
In fact, the numerical measurements of C are also involved in estimating the regression coefficient parameter for A.
Even if this was not a moderated regression the effect of A with C in the model could be different from the effect of A on its own (because C could be correlated with A). Indeed it could be uncorrelated with A and this statistical significance of A could change when C is in the model (because C reduces the error).
This is even more so with A and C and the interaction (A*C) in the model because A and C are likely to be correlated with A*C (eg., if both are largely positive or both largely negative as is common). A common practical solution is to centre A and C (subtract the mean of A from A so it has mean zero etc.) before running the moderator analysis (I don't know if PROCESS does this automatically as I don't use it). This makes the effect estimated in the interaction model evaluated at the mean of A and C and this should look similar to the effect of A and C in the model without the interaction. Note the interpretation of A in the two models is still different. In the model without interaction the A effect of A is constant over all values of C and vice versa. The presence of an interaction means that the effect of A depends on C and vice versa. So you can calculate the effect of A for a particular value of C but not describe its effects with a single number (though the centred model gives its effects when C is average).
Based on the lack of significance of Variable C, I would say that it does not moderate B in a significant way.
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology
- Research Papers