I have two factors that are fully crossed, the levels of the factor are each coded 0 and 1. I am running a regression testing for one main effect and one interaction. The following is my logistic regression formula:

m1=glmer(y~1+A+A:B+(1|Participants)+(1|Word),data=data, family = "binomial")

I am wondering if this is acceptable (only testing for one main effect and an interaction), and also why I am getting two interaction terms in my output:

Fixed effects:
        Estimate Std. Error z value Pr(>|z|)   
(Intercept) -0.18740    0.21600  -0.868  0.38561   
A1           0.74546    0.28399   2.625  0.00867 **
A0:B1        0.01537    0.28244   0.054  0.95662   
A1:B1        0.15884    0.28650   0.554  0.57929   
  • 2
    $\begingroup$ Please migrate it to CV. $\endgroup$ – user227710 Jul 3 '15 at 0:17
  • $\begingroup$ Is there some reason why you are not including the main effect of B in your model? It's generally best to include main effects of variables for which you are examining interactions. $\endgroup$ – EdM Jul 3 '15 at 15:22

There are 4 possible states for the interaction of A and B. That is

  1. A0:B0, and not shown as it is absorbed into the intercept term for the regression.
  2. A1 (implicitly A1:B0).
  3. A0:B1.
  4. A1:B1.

Each of these states affects the response in a different way.


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.