Discrete time survival analysis with time-varying (potential endogenous) discrete covariates question

[Daniella Brals]

Dear all,

Does somebody know whether it is possible to model discrete survival data (daily data, but data was observed once in 24 hours) with time-constant and time-varying (possibly endogenous) discrete covariates? I have been looking at many packages for survival analysis and joint modelling but I can't seem to find one that fits everything, i.e. discrete time survival + longitudinal covariates + discrete covariates + potential endogenous covariates.

Any ideas?

Thank you very much in advance!

Greetings!

[Maarten Buis]

A discrete time survival model has no commands on its own, because it does not need one. It is just your regular logit, probit, cloglog, or some other model for binary dependent variable on properly prepared data. The data has to be a person/period file (assuming your unit of analysis are persons, e.g. when they are firms then you would have a firm period file), which you can create with stsplit. What you are doing is trick Stata into maximizing the likelihood of a discrete time survival model, so you don't need to (and thus should not) add any "corrections" for the fact that the same person appears multiple times in you dataset. The datastructure makes it trivial to include time varying variables (either discrete or continuous). You can also make the effects differ over time if you want by adding the appropriate interaction terms.

I don't know whether this trick also works with endogenous models like an erm. You will just need to write down the likelihood for the model you want and check whether you can use the same trick to estimate the model.