Listy Biometryczne – Biometrical Letters vol. 39(1), 2002 pp. 29-40
|
MAXIMUM LIKELIHOOD ESTIMATION FOR MULTIVARIATE CATEGORICAL DATA Ewa Bakinowska, Radosław Kala Department of Mathematical and Statistical Methods, Agricultural University of Poznań, Wojska Polskiego 28, 60-637 Poznań, Poland ewabak@owl.au.poznan.pl, kalar@owl.au.poznan.pl |
The generalized linear models were developed for a broad class of practical problems related with the sets of multivariate categorical data. The analysis of such models can be based on the maximum likelihood principle. This approach has been discussed and the large-sample distributions, as well as an iterative fitting algorithm, were presented. In this paper we give a direct method of deriving the maximum likelihood equations. This approach is free from the superfluous assumptions and, in result, can be applied to the experiments with a priori empty cells. The theory is illustrated by four simple examples.
multivariate categorical data, logistic transform, linear model, multinomial distribution
