The Use of Models in the Analysis of Some Classes of Contingency Tables

Publisher Website
Google Scholar
Abstract
The use of linear models for the analysis of contin- gency tables is discussed specifically for the case where each cell is an independent binomial trial. However, the use of non-linear models for this situation is outlined and the extension of these ideas to contingency tables where margins are not held fixed is also made. The problem of central interest is the estimation of the unknown parameters of the models. For the linear models introduced in this paper, maximum likelihood methods produce estimates which are easily calculated and their variance-covariance matrix is readily obtainable in a systematic fashion. The suggested procedure lends itself to the Wald test and the Lagrange-multiplier test criteria for the testing of hypotheses with regard to subsets of the parameters. Estimation for the non-linear model is effectively handled by an unweighted least squares procedure. These estimates are unbiased if not fully efficient.

This publication has 1 reference indexed in Scilit: