Dear all,

I have a causal chain X -> M -> Y and am interested in the indirect effect of X on Y. This can usually be easily estimated using SEM and nlcom.

My problem is that the X -> M part in my data requires a zero-inflated Poisson regression (whereas the M -> Y part does not require special treatment). How would you estimate the indirect effect of X on Y in this case?

I tried specifying the following SEM:

gsem (1: m <- , family(pointmass 0)) (2: m <- x, family(poisson)) (C <- x t)(y <- m x), lclass(C 2) lcinvariant(none) vce(cluster id)

However, I encounter two issues: First, the entire model is estimate in two latent classes, including the latter par (M -> Y). However, I would like to get one indirect effect at the end. Second, I don't think I am allowed to simply multiply the two coefficient (b[X->M] * b[M->Y]) because zero-inflated Poisson is logistic.

How would you proceed? I would be deeply grateful for some guidance.

Best,

Johannes