Paul-Christian Bürkner

The `1*variable` syntax does not work in brms that way (it just works in lavaan). You have to soft-fix the coefficient with a very strong prior. There is currently no...

You are absolutely right. In any case, specifying these kinds of models in the current brms syntax is cumbersome and we definitely need a special, user-friendly syntax for it.

I generally like your ideas! Thanks for this! A few comments: * `=~` is no valid R operator but rather translates to `y = ` which is reasonable, but nothing...

The general brms approach is to use R formulas for the model specification and differentiate their meanings by the functions to which they are supplied. That is, your second proposal...

Thanks again! Much of this is in accordance with the structure I have in mind, but just realized I haven't written it down somewhere. Below is the my suggestion on...

The only thing I don't understand in your proposal is the latent interaction part. Could you perhaps write down in a set of equations how such an interaction model you...

I don't think the measurement model above makes a lot of sense. The interaction between two latent variables should not have a measurement model itself but arises simply by multiplying...

Let's put the interaction discussion aside for a second. We can come back to this later on once there is a working implementation of SEMs in brms. I agree with...

I think users should be able to identify loadings both by name and by position, that is both `fix(x1 = 1, x3 = "a")` and `fix(1, NA, "a")` should work....

It could indeed be worth it to add lavaan syntax (or parts of it) as syntactical sugar on top. However, my plan is to support a lot of SEM models...