|
|
 |
| Discrete-time model with mediation |
 
|
| Message/Author |
|
|
|
|
| I am assisting a postdoctoral fellow who is interested in estimating indirect effects for a model of the form X -> M1 -> M2 -> Y, where Y is a discrete-time variable. I was wondering if there would be any contraindications to using the WLSMV estimator and MODEL INDIRECT to fit this model using a data structure analogous to how Y would be treated using a pooled logistic regression approach for estimating discrete time survival models as described, for instance, in Paul Allison's Survival Analysis using SAS book. Thanks. |
|
|
|
|
I guess that is possible. So I think you refer to considering the latent response variable Y* as the outcome as you would if you did regular logistic/probit effect estimation without going into the counterfactual effects for the binary Y.
Otherwise, perhaps you can set up the data as in the DTSA of UG ex 6.19 and consider effects for f. |
|
|
|
|
Thank you, Bengt.
Yes, in this analysis we would consider y* as the outcome so that we could use MODEL INDIRECT with the traditional (non-counterfactual) effects because my understanding is that with two mediators in a chain or sequence X->M1->M2->Y, counterfactual causal indirect effects are not available.
I assume this would also be the case with the DTSA framework of UG ex 6.19 - if we had X->M1->M2->F, we would still need to use WLSMV to obtain the indirect effect of X on F through M1 and M2, right? |
|
|
|
|
| Right on both. |
|
|
| Back to top |
|
|
