I have a question regarding the output for my mediation analysis. I am new to the zero-inflated models and mediation analysis. I apologize if my questions are elementary.

My mediation model is zero-inflated poisson (hereafter, zip) and my outcome model is logistic regression. The output is attached here.

Since the mediation model is zip, there are two sets of estimates generated: poisson (gamma 1 from the orange path) and logit (gamma 2 from the blue path). Then, in the outcome model, both the poisson and logit parts predict Y. Hence, there should be beta 1 from the orange path and beta 2 from the blue path. Please correct me if I am wrong.

From the brms output, there is only one estimate for Y ~ M. My question is how to interpret the relationship between M and Y in brms. It seems that M is not decomposed into 2 parts (poisson and logit) when predicting Y. Is that so? Thank you very much for your help!

Can you post the brm call that does the calculations? I’m not familiar with these kinds of models. It’s always fair to ask these kinds of questions. It’s not always obvious what the interfaces do without using them a bunch.

Other than that, you can replace your ‘brm’ call with make_stancode to generate the underlying Stan model brms uses (make_standata for the data) and verify brms is doing the thing you want.

Either of those things help? If not we can summon someone else to the thread who’ll know.

I am so sorry for my late reply. I am actually stuck on the output summary. I was surveying relevant literature. To my knowledge, there is only one paper talking about zero-inflated mediators. Based on the paper, there should be 2 estimates from the mediator to the outcome variable as indicated in my drawing in the first post (i.e., beta 1 and beta 2). In the brms output, there is only one estimate for Y ~ M (i.e., letterVoccur_disc_sFreq = 0.47). Therefore, I am just wondering if I am missing anything or misunderstanding the output. Thank you very much!! zero inflated mediator.pdf (1.5 MB)