In the one-way∗ layout to compare the means of k normally distributed populatons, it may not be valid in some cases to assume homogeneous variances.
Hence the ANOVA∗F-test∗ is not applicable, and the Welch [19] test was proposed to fill this void. An important special case (k = 2) is the famous Behrens–Fisher∗ problem.
This special case was solved by Welch [18] several years earlier than the general case.
His solution for k =2 was refined and tabled by Aspin [1,2] and has become known as the Aspin–Welch test(AWT). Further tables were later provided by Trickett et al. [17].
Competing solutons to the Behrens–Fisher problem have been suggested by Fisher [8], Lee and Gurland [11](denoted LG), Cochran [6], and Welch himself[18; 2, Appendix].
All these tests depend on normality, and Yuen [21] and Tiku and Singh [16] attempt more robust solutions.
Some competing procedures for general k are due to Brown and Forsythe [5], James [9],and Bishop and Dudewicz [3].
TheWelch and Brown-Forsythe tests have been extended by Roth [13] to the case where the k populations have a natural ordering (e.g., different dosages of the same drug) and a trend test∗ is desired to detect differences in the means that are monotone as a function of this ordering.
In the one-way∗ layout to compare the means of k normally distributed populatons, it may not be valid in some cases to assume homogeneous variances.Hence the ANOVA∗F-test∗ is not applicable, and the Welch [19] test was proposed to fill this void. An important special case (k = 2) is the famous Behrens–Fisher∗ problem.This special case was solved by Welch [18] several years earlier than the general case.His solution for k =2 was refined and tabled by Aspin [1,2] and has become known as the Aspin–Welch test(AWT). Further tables were later provided by Trickett et al. [17].Competing solutons to the Behrens–Fisher problem have been suggested by Fisher [8], Lee and Gurland [11](denoted LG), Cochran [6], and Welch himself[18; 2, Appendix]. All these tests depend on normality, and Yuen [21] and Tiku and Singh [16] attempt more robust solutions.Some competing procedures for general k are due to Brown and Forsythe [5], James [9],and Bishop and Dudewicz [3]. TheWelch and Brown-Forsythe tests have been extended by Roth [13] to the case where the k populations have a natural ordering (e.g., different dosages of the same drug) and a trend test∗ is desired to detect differences in the means that are monotone as a function of this ordering.
การแปล กรุณารอสักครู่..