Two-Sample T 3 Plot: A Graphical Comparison of Two Distributions

Abstract

Consider two independent random variables x and y with means and standard deviations μ _x ,μ _y ,σ _x , and σ _y , respectively. Let F _x (t) = P[(x - μ, _x )/σ _x ≤ t] and F _y (t) = P[(y - μ _y )/σ _y ≤ t]. In this article we address the problem of testing the null hypothesis H ₀ : F _x ≡ F _y , against the alternative H ₁ : F _x ≡ F _y . A graphical tool called T ₃ plot for checking normality of independently and identically distributed univariate data was proposed in an earlier article by Ghosh. In the present article we develop a two-sample T ₃ plot where the basic statistic is the normalized difference between the T ₃ functions for the two samples. Significant departure of this difference function from the horizontal zero line is indicative of evidence against the null hypothesis. In contrast to the one-sample problem, the common distribution function under the null hypothesis is not specified in the two-sample case. Bootstrap is used to construct the acceptance region under H ₀, for the two-sample T ₃ plot.