I'm a bit confused about the various notations that are used to describe effects and coefficients in mixed models. I have 2 main effects and included the 2-way interaction effect in my model. The Wilkinson notation of that model is given with

$$ R \sim 1 + G + V + G{:}V + (1|I_{ID}) $$

with my two main effects $G \in \{G_1, G_2, ... G_N\}$ and $V \in \{V_1, V_2, ... V_M\}$.

Now I understand, the $:$ operator only includes the interactions, i.e. $G:M \in \{G_1V_1, G_1V_2, ... G_NV_M\}$ whereas the $*$ operator would also include the lower order effects. So $G*M$ would be equivalent to $G + M + G{:}M$.

Using Jamovi and GAMLj, I get omnibus test result that then include test-results for these effects: $G$, $V$, $G*V$. The last entry seems redundant because following the definition above, it should be $G{:}M$ instead.

The same is true for coefficients table. These then include entries like $G_1-G_2 * V_1-V_2$ which is the combination of the $G_1$ versus its reference $G_2$ and $V_1$ versus its reference $V_2$. Again I'm a bit confused why the $*$ operator is used here.

Does anybody know why the $*$ operator is used in these cases? Or is there a standard on how to use the two operators and what they actually denote?

  • $\begingroup$ I suspect this is a software "infelicity" (i.e. someone might argue it's not actually a bug, and it's not worth getting hung up on whether it is or not). It would be good to provide a link: gamlj.github.io . I would suggest submitting an issue at github.com/gamlj/gamlj/issues $\endgroup$ – Ben Bolker Feb 1 at 18:07

I don't know what Jamovi and GAMLj are, but I agree with you that, when you specify the main effects $G$ and $V$, it is superflous to write also write $G*V$ since $G*V$ literally means $G + V + G:V$, so that $G + V + G*V$ actually means $G + V + G + V + G:V$. All software that I am aware of would just ignore the extra main effects.

On that basis, I can't see any reason that the $∗$ operator is used in your case.

However, we should always remember that an interaction term is literally the multiplication of one variable by another, so if we were to define $*$ as the interaction, then it would make sense.


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.