6

I have three variables. I know that variable B has a correlation with variable C. I'm interested in whether variable A impacts variable B's relationship with variable C, and not on whether variable A directly impacts variable C.

Multiple regression seems not appropriate for this question as this would be measuring each of variable A and B impact on C, which is not what I want.

My other thought was to simply multiple A by B and then correlate with C in order to determine whether this has increased / decreased the correlation.

Any other ideas to determine whether A impacts B's relationship with C?

Or thoughts on why either of the above is or isn't useful?

DoJ
  • 61
  • 2

1 Answers1

4

The best approach is to combine both any "direct" effect of A along with the interaction represented by the product of A times B in a multiple regression. The general form for such a model, as it's represented in R syntax, would be:

C ~ B + A + A:B

That model (if you use the standard treatment coding) will provide:

  • an intercept--the value of C when A and B are both at their reference values (0 for a continuous predictor),

  • individual coefficients for A and B, representing how much C changes with respect to each of them when the other predictor is at its reference level, and

  • a coefficient for the interaction A:B (which is just the product of A times B for continuous predictors), representing how much the individual associations of each of A and B change as the other one of the pair changes.

For this to be most interpretable, you should include A in your model in addition to the interaction coefficient even if you don't care specifically about its individual association with C.

EdM
  • 57,766
  • 7
  • 66
  • 187