Robust Method for Testing the Significance of Bivariate Correlation of Ordinal Data

Abstract

The Tau’s statistics were introduced to solve the problems of tied data but its effect on shape of table cannot be ascertained. This study compares the non parametric approaches for testing the significance of bivariate correlation for ordinal data based on table shape (square and rectangular table). The bootstrap method was used to compare the magnitude of the correlation values and the associated standard error of the values of  tau’s -b, c and Gamma over a square and rectangular table. The Gamma was found to be the most robust statistic for computing the correlation of ordinal data

Keywords: Tau statistics, Gamma, ordinal correlation, Bivariate correlation