PARAMETRIC INFERENCE IN MULTIVARIATE MARKOV MODELS
WITH DEPENDENCE ON COVARIATES

Manuel
MOLINA FERNANDEZ^{1}

Miguel
GONZALEZ VELASCO[1]

Manuel BARRANTES LOPEZ[2]

§ ABSTRACT:
In this paper, the modelization of the probability distribution of a
nominal-scaled response vector, influenced by covariates, is studied. We assume
that the sequence of responses is well described by a multivariate Markov chain
and we model their transition probabilities by mean of the multiple-group
logistic regression model. The maximum likelihood estimation of the regression
parameters is considered and a recursive method to estimate the parameters when
some responses are missing is suggested. Finally, the likelihood ratio
criterion is used to test some hypotheses about the model.

§ KEYWORDS:
Multivariate Markov chain; multiple-group logistic regression model; maximum
likelihood inference.