Suppose you have panel data (many observations for each person) and you have a discrete ordered outcome for worker status (entry level, assistant, manager, exit labor force) that you want to model transitions between as a function of time-fixed and time-varying covariates. For example based on covariates, what is probability of an assistant becoming a manager at the next time step. I wish to model this type of data in R but am not sure if there is such a way to do it to account for the repeated observations. I have seen many different packages including nnet pglm and vgam but they all give different result and some of the probabilities simply don't make sense. For example in some cases the probability of going from "exit labor force" to "assistant" is more than 20% even though there are no transitions like this in the model. Finally- some people tell me this is a markov problem but that assumes the time in each worker status does not matter but it does so i am wondering how to account for these different aspects.

Any advice would be very useful at this time.

  • $\begingroup$ I think "some people" were right. Try using continuous time Markov process in your favourite search engine and see whether that helps. $\endgroup$ – mdewey Dec 23 '20 at 15:03
  • $\begingroup$ That the transition matrix varies over time only means this Markov chain is time-inhomogeneous; but a Markov chain nevertheless. $\endgroup$ – ColorStatistics Dec 23 '20 at 17:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.