I have a dataset with two duration outcome variables and many explanatory variables. The outcome variables can be censored and uncensored. So, I am using parametric survival analysis to model them.
outcome variable 1: duration of keeping a phone
outcome variable 2: duration of keeping a laptop
I know how I can model each of them separately with parametric survival analysis. But, I would like to model their joint distribution (considering the correlation between the two).
I could not find any related resources on this topic. I was wondering if any resource (web page, article, etc) is available for modelling such a problem?
Is this something similar to structural equation modeling? Both Stan and brms (by extension) support this. Here is a walkthrough RPubs - BRMS and Bayesian SEM and Gelman talks a bit about it here Structural equation modeling and Stan « Statistical Modeling, Causal Inference, and Social Science . A longer paper here https://arxiv.org/pdf/2008.07733.pdf
Thank you for your reply and for the material.
I do not think SEM can help in this situation as I have censored outcome variables. Survival analysis can handle censored outcome variables. That is why I am looking for an approach based on survival analysis.
Thank you anyway!
If you’re writing your own Stan code, you can code each one as a standard survival analysis and model their intercepts as a multivariate normal.
If you’re not writing your own code any multi state model package could work (eg- R’s msm package)… but I don’t recall if there’s a Stan-based msm package…
Thank you for the suggestion. It was informative. However, as I need to customize my likelihood function, I need to write my own stan code.
And regarding modeling intercepts as a multivariate normal, would you know of any article on that so that I can read more about it?