Abstract by Cason Wight
Inference for semi-Markov models with panel data
Semi-Markov processes effectively model waiting times and probabilities for many multi-state scenarios. Observed data from such processes often lack details about transition times. For example, the exact transitions of cancer patients between differing stages of cancer are often not observed. In practice, data are collected at specific points in time, where the stage of the patients is observed at appointments. Intermittently-observed measurements, such as this, are known as panel data. Our purpose is to estimate the parameters of a semi-Markov model with panel data, which is needed in application areas such as medicine and engineering. The state-of-the-art technique uses the stochastic expectation-maximization (SEM) algorithm for inference. Our goal is to implement a similar technique using the standard expectation-maximization (EM) algorithm, which has better convergence properties. The EM algorithm will also shorten calculation time and increase accuracy.