Technical Reports


Estimation of chain-dependent survival distributions from censored observations

Dyer, Justin S.; Tang, Hsin-Khuern
HP Laboratories


Abstract: We present a survival model for data with multiple recurrent events of several interrelated types, such as might arise from customer transactions on a website. The model is based on a semi-Markov chain that drives several renewal processes. We develop a computationally tractable expectation-maximization (EM) algorithm for fitting the model to observed data. Through this model, we are further able to estimated quantities related to an absorbing, or "churn", state that is never actually observed in the data and which corresponds to subjects leaving a service in our application. Furthermore, we provide a closed-form solution for the fully nonparametric case which leads to consistency results for this parameters. Finally, we demonstrate the approach through numerical simulations.

23 Pages

External Posting Date: March 6, 2010 [Abstract Only]. Approved for External Publication - External Copyright Consideration
Internal Posting Date: March 6, 2010 [Fulltext]

Back to Index