This question relates to the same analysis that I posted about yesterday, but the question is different, so I have started a new entry.

I have conducted a moderation analysis by entering continuous independent variable X, continuous moderator M, both centered to their means, and the product of the two (X.M, interaction factor) into a simultaneous regression model, which was significant. The regression coefficient for interaction term X*M is significant (but has an opposite sign to the main effects). The dependent variable is Y.

The following simple slopes were plotted, regressing Y on X at high (+1SD) and low (-1SD) levels of M (please ignore the "medium" slope):

enter image description here

The unstandardized slopes for high M=0.027 (SE=0.2) and for low M=0.135 (SE=0.901) (neither is statistically significant, but that is the topic of the other question).

My question now is this:

From the figure, it is obvious that for the plotted values, the value of Y for a given value of X is higher at high M than it is at low M. However, the slope of low M is greater than for high M. How should this be interpreted? "X was predictive of higher levels of Y at high M than at low M"? Or "The effect of M on the relationship between X and Y was greater at lower levels of M than lower levels of M?"

Thanks in advance!

  • $\begingroup$ In the regression, were coefficients for all of M, X and the M:X interaction significant? $\endgroup$
    – EdM
    Commented Jun 9, 2015 at 15:41
  • $\begingroup$ Yes, they were all significant at p less than 0.05. $\endgroup$
    – B B Jinx
    Commented Jun 9, 2015 at 16:00

1 Answer 1


There are several ways to present this. You should make it clear that both higher X and higher M were associated with higher Y. For the "opposite sign" of the interaction effect, this means both that the influence of X on Y decreased as M increased and that the influence of M on Y decreased as X increased.

So one way to express all of this would be: "Increases in both independent variables X and M were associated with increased values of Y, but the influence of either independent variable on Y decreased as the other independent variable increased."

The statements you propose in the question have some limitations. For example, "X was predictive of higher levels of Y at high M than at low M" would be true even if there wasn't a significant X:M interaction, provided that there were positive slopes for both X and M individually.

"The effect of M on the relationship between X and Y was greater at [higher, I assume you meant] levels of M than lower levels of M" is ambiguous, because the absolute magnitude of the effect of M on the Y/X slope is greater at higher M but the sign of the effect is negative, so that the Y/X slope becomes close to 0 at high M. In fact, the statement is so ambiguous that I wasn't quite sure which of the two "lower"s in that sentence you had intended to be "higher".

  • $\begingroup$ Thank you! That was very helpful. In particular, the explanation of how the opposite sign for the interaction had a role to provided me with the piece of the puzzle that was missing. $\endgroup$
    – B B Jinx
    Commented Jun 9, 2015 at 19:01
  • $\begingroup$ As a follow up question, for future reference, if M were to be a dichotomous categorical variable instead of continuous (say, two possible groups, A and B, dummy coded with Group A being reference=0, and Group B=1), and the regression coefficients for the direct effects of X and M were positive but the interaction effect was negative (and all significant, of course) could one say that "both increases in X and belonging to Group B were associated with higher values of Y, but the influence of X was lower in Group B than Group A?" $\endgroup$
    – B B Jinx
    Commented Jun 9, 2015 at 19:11
  • $\begingroup$ Yes, but it might be clearer if you said "... the influence of X on Y was lower ..." (Don't forget that the relation between group B and Y also decreases with increasing X in this case.) And although I used the words "influence of" myself, that might not be the best terminology to use in a formal presentation, as some would take that to imply a causal influence rather than a statistical association. $\endgroup$
    – EdM
    Commented Jun 9, 2015 at 19:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.