Log-linear models : an assessment of three approximations for computing power

Author

John F. Schneider

Date of Award

12-1986

Degree Type

Thesis

Degree Name

Master of Science

Major

Statistics

Major Professor

Ralph G. O'Brien

Abstract

This thesis investigates three methods of computing power for a log-linear model analysis. The goal of this research is to determine which of the three is the "best," defined here as the method whose nominal powers are closest to the powers obtained from a Monte Carlo study. Two Monte Carlo studies were carried out. The first one examined cases where each of the three methods yielded approximately the same nominal powers. The second study examined cases where the three methods yielded widely different nominal powers.

Simulations were carried out to help answer the question at hand. A data set would be generated, and a hypothesis would be tested. A tally was made of the number of times the hypothesis was rejected. The proportion of times this happened provides an estimate of the "true" power. This would then be compared with the nominal value.

Of the three methods studied, the likelihood ratio, the Wald, and the Oler noncentrality approximations, the likelihood ratio noncentrality approximation comes out the best of the three.

Recommended Citation

Schneider, John F., "Log-linear models : an assessment of three approximations for computing power. " Master's Thesis, University of Tennessee, 1986.
https://trace.tennessee.edu/utk_gradthes/13801