Categorical Data Analysis: Some Reflections on the Log Linear Model and Logistic Regression. Part II: Data Analysis

Imrey, Koch, Stokes and collaborators (1981) have reviewed the literature of log linear and logistic categorical data modelling, and presented a matrix formulation of log linear models parallel to the well-known representation of general linear models for continuous data. Several related methods of...

Full description

Saved in:
Bibliographic Details
Published in:International statistical review 1982-04, Vol.50 (1), p.35-63
Main Authors: Imrey, Peter B., Koch, Gary G., Stokes, Maura E., Darroch, John N., Freeman, Daniel H., Tolley, H. Dennis
Format: Article
Language:English
Subjects:
Analytical estimating
Covariance matrices
Estimation methods
Least squares
Linear models
Logistics
Maximum likelihood estimation
Modeling
Parametric models
Statistics
Citations: Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Imrey, Koch, Stokes and collaborators (1981) have reviewed the literature of log linear and logistic categorical data modelling, and presented a matrix formulation of log linear models parallel to the well-known representation of general linear models for continuous data. Several related methods of estimation, maximum likelihood, weighted least squares and a hybrid, functional asymptotic regression methodology, were discussed. In this sequel, seven data sets are interpreted with the purpose of clarifying the relationships between log linear and logistic modelling, and between the several estimation approaches. Asymptotic covariance matrices of estimated parameters are also developed in two 'nonstandard' situations. /// Imrey, Koch, Stokes et collaborateurs (1981) ont examiné la littérature portant sur les modèles log linéaires et logistiques, et ont offert une formulation matricièlle de la modèle log linéaire générale qui est comparable à la representation bien connue de la modèle générale linéaire pour les données continuelles. Ils ont discuté quelques approches à l'estimation des paramètres: le maximum de vraisemblance, les moindres carrés pondérés, et la régression asymptotique fonctionelle. Ici, sept ensembles des données sont interpretés afin de clarifier quelques articulations entre les modèles log linéaires et logistiques, et entre les méthodes d'estimation des paramètres. Aussi, les matrices des covariances asymptotiques des paramétres estimées sont décrites pour deux situations pas 'classiques', concernant les distributions non centrales et les plans de sondages complexes.
ISSN:0306-7734
1751-5823
DOI:10.2307/1402458