Analysis of Opinion Swing: Comparison of Two Correlated Proportions

An important problem that arises in introductory courses of applied statistics and categorical data analysis is the evaluation of opinion swing. Suppose that a group of individuals is surveyed on their support for the President. After the State of the Union Address the individuals are consulted again. The objective is to analyze whether or not there has been a change-a swing-in their opinion. Due to the longitudinal nature of data, only tests of hypothesis are presented to the students-for instance the McNemar test. The estimation of the parameters of interest is left to the advanced literature. In this article we suggest simple solutions for the estimation problem, to be presented in introductory courses of applied statistics.