I am trying to interpret my interaction effects, which are all negative. One example: Experience (variable A) x absolute size of the acquired knowledge base (variable B): B= -0.002, exp(B)= 0.998.

Can I interpret this interaction in the following way: As variable B decreases, the effect of A increases. How can this interaction be interpreted in terms of exp(B)?


When variable A increases by one unit the effect of variable B decreases by a factor .998 or $(.998-1)\times 100\% = -.2\%$.

There are many examples of this type of interaction effects, including one for a negative binomial model, here: http://www.maartenbuis.nl/publications/interactions.html

| cite | improve this answer | |
  • $\begingroup$ I tried to get my mind around this but I still have some questions :). I test the main effect of several IV. However, I also suspect that one of my IV (experience) has a moderates the relationship between the remaining IV and the DV. I have the following main effects for my IVs without interactions in the model: Experience=0.011*, IV B=0.016, IV C=0.106**, IV D=-2.693**, IV E-0.139. In a second model I now added the interactions between experience and the other IVs, leading to the following results: Experience=0.045**, IV A=0.092, IV B=0.146**, IV C=-2.506**, IV D=-0.156 $\endgroup$ – Schwadi Jun 16 '13 at 13:20
  • $\begingroup$ My interactions have to following coefficients: Experience x B=-0.009, Experience x C= -0.002*, Experience x D=-0.020 Experience x E=0.003. Do I understand it right that the coefficients of the main effects are now for when experience has a value of 0? The coefficient of the interaction should then show in which direction the coefficient of the main effect changes. Thus, a negative interaction means that a positive main effect will have an increased coefficient, which shows that there is moderation. A significant interaction coefficient then means that this moderation is significant? $\endgroup$ – Schwadi Jun 16 '13 at 13:26
  • $\begingroup$ In principle yes, except that I would not say something like "this moderation is significant". This often only distracts from the real message of your paper. I would stay a lot closer to the real meaning of the variables, something like "if someone has 0 years experience a year increase in education leads to an (exp(b)-1)*100% in/decrease in Y, and for every year increase in experience this effect of education decreases by (exp(a)-1)*100%." $\endgroup$ – Maarten Buis Jun 17 '13 at 7:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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