Estimating Markov Transition Probabilities from Micro-Unit Data

SUMMARY The standard method of deriving maximum-likelihood estimates for transition probability matrices is outlined. Statistical tests for determining the Markov property, the specific order of a chain and the homogeneity of a set of matrices are applied to data relating to manufacturing establishments. Refinements to the maximum-likelihood procedure are examined in the context of fractional disaggregation of long-period matrices and in the framework of fitting smoothing surfaces to observed transition probabilities.