Test treatment effect differences in repeatedly measured symptoms with binary values: The matched correspondence analysis approach

When a continuous variable is measured twice, paired t test can be used to examine the statistical difference between two time points. However, when several related but dichotomously scored (0, 1) variables are measured twice, it would not be reasonable to use paired t test (or chi-squared test) to examine the related binary variable differences. Therefore, the present study introduces a novel statistical approach, called matched correspondence analysis (matched CA), which tests the related binary value differences between two time points. Matched CA was originally designed to study between-group comparisons (e.g., gender) in two contingency tables of the same size, with the same row and column quantities. However, unlike the original matched CA, the present study applies matched CA to the analysis of within-group matched matrices (e.g., at admission and at discharge) and examines the related binary value differences between two time points. To test the stability of parameter estimates, permutation and bootstrapping methods are used, and the pros and cons of within-group matched CA are discussed.