STUDY OF INFLUENCE IN LOGLINEAR MODELS
OF DIRECT ESTIMATION

Rosana CEPEDA[1]

Claudia MARINELLI1

Nélida WINZER[2]

§    ABSTRACT: In order to perform influence studies on contingency tables in which a loglinear model of direct estimation fits adequately, the diagnosis measurements proposed by Andersen (1992) for an independent model can be generalized in a simple way. These generalizations allow to determine how much influence each cell has in the formulated model. Nevertheless these measurements do not give specific information about the interaction terms of the model. In this work we present an analysis complementary to that of Andersen that permits to explain the influence on the interaction terms. More specifically, by means of the definition of influence curve in Canonical Correlation Analysis for categorical data, it is possible to decide which cell has layer influence on the inertia in each marginal table, hence, achieving in this way, a more exhaustive study. The advantages of the proposed methodology are discussed through its application to different data sets.

§    KEYWORDS: Influence curve; loglinear model; inertia; canonical correlation.

 



[1] Instituto Multidisciplinario de Ecosistemas y Desarrollo Sustentable, UNCPBA, Pinto 399, B7000GHG Tandil,Buenos Aires, Argentina.E-mail: rcepeda@exa.unicen.edu.ar / cmarine@exa.unicen.edu.ar

[2] Departamento de Matemática, UNS, Alem 1253, B8003JSM Bahía Blanca, Buenos Aires, Argentina, E-mail: nwinzer@criba.edu.ar