The Retention Model: A Markov Chain with Variable Transition Probabilities

Abstract

When considering a nonhomogeneous Markov chain one often desires an intermediate hypothesis between constancy and free variation of transition probabilities. It is shown that some interesting and easily tested hypotheses can be formulated by considering the diagonal elements of the transition matrices independently of the off-diagonal elements. After deriving the maximum likelihood estimates, a reanalysis of panel data on voting intentions Is presented, In which the nonhomogeneity is seen to be entirely due to variation in the retention probabilities piit.