When a Variable Is Not Perfectly Correlated with Itself: Time to Retire Kendall's Tau_a Rank Correlation

摘要

Kendall's tau_a rank correlation coefficient is commonly presented in research methods texts and is readily available in statistical software packages. Tau_a is not defined to accommodate ties in the ranked data it analyzes. Kendall's tau_b correlation, though, is defined to accommodate ties in the ranked data. Where there are no ties in the ranked data, the tau_b formula is the same as tau_a, i.e., tau_b completely supplants tau_a. The present research suggests that tau_a not be presented in textbooks or statistical software packages.