I am testing the effectiveness of a treatment on accuracy of decisions. I believe that the treatment will help people with low iq make more accurate decisions, and for people with high iq it won't make a difference because their decision making is already optimised.

What i do is i split the sample in high and low iq. Then i compare accuracy between control and treatment within low iq and control and treatment within high iq.

In this scenario, is my model

  1. testing the moderating effect of treatment on the relationship between iq and accuracy, or
  2. testing the moderating effect of iq on the relationship between control/treatment and accuracy?


  • $\begingroup$ It can be both. It depends on your story. $\endgroup$ Jun 14, 2022 at 10:53
  • 1
    $\begingroup$ Does this answer your question? "Moderation" versus "interaction"? Statistically, moderation is represented as an interaction term in a regression model. There is no distinction between the two variables involved in the interaction. Your interpretation of the results, based on your understanding of the subject matter, might put more emphasis on one over the other. $\endgroup$
    – EdM
    Jun 14, 2022 at 13:17


Browse other questions tagged or ask your own question.