Confidence Intervals for Ratios of Multinomial Proportions

Abstract

In this thesis, we propose Methods for constructing five confidence intervals for ratio two multinomial distribution probabilities. The actual confidence interval coefficient (coverages) and mean or expected confidence interval lengths for all five Methods are compared via Mante Carlo simulation. It is found that a certain confidence interval based upon the logarithm of a sample ratio of properties has the best properties. Some other recommendations are made.