I would like to set up a brms formula that includes two variables, say
A is boolean (true/false) and
B continuous. The formula should consider
A by default, and
B only if
A is true (since
B has no physical meaning if
A is false). Is this possible to express in a brms formula? I had a hard time finding related examples so my problem might be ill-posed.
One approach that gives reasonable parameter estimates is to convert
A from TRUE/FALSE to 1/0 and then include
B with an interaction:
formula = A + B:A
Is this a valid solution? It feels like this is not a “standard” way to include an interaction term as
B is not included as a main effect.
- Operating System: Windows 10
- brms Version: 2.14.4
This is how I would do it. Sure, it’s not standard. But you don’t have a standard model and your justification seemed sensible.
Thanks again for your view on this.
Sorry, I got a late follow-up question. I can also open a new thread but I thought this fits better here. Since I have other continuous variables in the model that I would like to compare parameter coefficients across after fitting, I standardize all variables before fitting in brms. Do you have any idea on how to do that for variable
B in the example above? Since
B only makes sense when
A is true (or equal to 1), I should actually replace all values of
B with NA’s wherever
A is false. But that would cause the interaction
B:A to become NA as well, while it should remain zero, possibly causing issues for the fitting.