Classification errors in contingency tables analyzed with hierarchical log-linear models. Technical report No. 20 [electronic resource].
- Washington, D.C. : United States. Dept. of Energy, 1978. and Oak Ridge, Tenn. : Distributed by the Office of Scientific and Technical Information, U.S. Dept. of Energy.
- Physical Description:
- Pages: 111 : digital, PDF file
- Additional Creators:
- Stanford University. Department of Statistics, United States. Department of Energy, and United States. Department of Energy. Office of Scientific and Technical Information
- Restrictions on Access:
- Free-to-read Unrestricted online access
- This thesis is concerned with the effect of classification error on contingency tables being analyzed with hierarchical log-linear models (independence in an I x J table is a particular hierarchical log-linear model). Hierarchical log-linear models provide a concise way of describing independence and partial independences between the different dimensions of a contingency table. The structure of classification errors on contingency tables that will be used throughout is defined. This structure is a generalization of Bross' model, but here attention is paid to the different possible ways a contingency table can be sampled. Hierarchical log-linear models and the effect of misclassification on them are described. Some models, such as independence in an I x J table, are preserved by misclassification, i.e., the presence of classification error will not change the fact that a specific table belongs to that model. Other models are not preserved by misclassification; this implies that the usual tests to see if a sampled table belong to that model will not be of the right significance level. A simple criterion will be given to determine which hierarchical log-linear models are preserved by misclassification. Maximum likelihood theory is used to perform log-linear model analysis in the presence of known misclassification probabilities. It will be shown that the Pitman asymptotic power of tests between different hierarchical log-linear models is reduced because of the misclassification. A general expression will be given for the increase in sample size necessary to compensate for this loss of power and some specific cases will be examined.
- Published through SciTech Connect., 08/01/1978., "coo-2874-40", and Korn, E L.
- Funding Information:
View MARC record | catkey: 14148543