Chapter 12--Multisample Inference

# That means high type ii error rate and hence large

This preview shows page 1. Sign up to view the full content.

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: : Biostatistics Analysis of Variance (ANOVA) Multiple Comparisons The Kruskal-Wallis Test Two-Way ANOVA Bonferroni’s Method Benferroni Inequality says, αE ≤ mαC . So to control EWER at level α, use Fisher’s method with α/m. Bonferroni’s procedure declares ˆi signiﬁcant if | ˆi | ≥ t1−α/(2m),n−k ˆ V ( ˆi ). This procedure gets more conservative as m gets large. That means high type II error rate and, hence, large number of false negatives. A 100(1 − α)% simultaneous conﬁdence interval ˆi ± t1−α/(2m),n−k ˆ V ( ˆi ). To apply Bonferroni’s method, we need to know m in advance. Chapter 12: Multisample Inference Stat 491: Biostatistics Analysis of Variance (ANOVA) Multiple Comparisons The Kruskal-Wallis Test Two-Way ANOVA The Scheﬀe’s Method For testing the signiﬁcance of ˆi , , Scheﬀe’s Method declares ˆi ; i = 1, . . . , m, signiﬁcant if | ˆi | ≥ ˆ V ( ˆi ). (k − 1)F1−α,k −1,n−k Scheﬀe’s method controls EWER at a level α. A 100(1 − α)% simultaneous conﬁdence interval for . . ., m is ˆi ± (k − 1)F1−α,k −1,n−k 1, 2, ˆ V ( ˆi ). Notice that Scheﬀe Method does not depend on m. That is, it is designed for arbitrarily large number of contrasts. In general it tends to be too conservative unless m is large. This method has to be used if you have post-hoc comparisons as it accounts for the after the fact nature of the hypothesis. Chapter 12: Multisample Inference Stat 491: Biostatistics Analysis of Variance (ANOVA) Multiple Comparisons The Kruskal-Wallis Test Two-Way ANOVA Example Consider the blood pressure example, with the three contrasts deﬁned before. ˆ i | ˆi | V ( ˆi ) Fisher Bonferroni Scheﬀe Cut Oﬀ Cut Oﬀ Cut Oﬀ 1 4.25 1.80 3.80 4.85 6.30 2 1.25 1.80 3.80 4.85 6.30 3 4.75 1.27 2.71 3.42 4.44 Note that Fisher’s Method declares that ˆ1 and ˆ3 signiﬁcant. Both Bonferroni and Scheﬀe declare only ˆ3 signiﬁcant. Chapter 12: Multisample Inference Stat 491: Biostatistics Analysis of Variance (ANOVA) Multiple Comparisons The Kruskal-Wallis Test Two-Way ANOVA Example Cont’d Since Fisher’s does not control EWER, we base our conclusions on Scheﬀe or Bonferroni. In general, the choice between Scheﬀe and Bonferoni is made based on which one declares larger number of contrasts signiﬁcant (has lower cut oﬀ point), that procedure would be more powerful or commits less type II error. Chapter 12: Multisample Inference Stat 491: Biostatistics Analysis of Variance (ANOVA) Multiple Comparisons The Kruskal-Wallis Test Two-Way ANOVA Example Cont’d 95% Conﬁdence Interval ˆi i Fisher 1 4.25 0.43 , 8.07 2 1.25 -2.60 , 5.07 3 -4.75 -7.45 , -2.50 Notice that Bonferroni and Fisher’s is individual. Bonferroni Scheﬀe -0.58 , 9.08 -2.02 , 10.52 -3.58 , 6.08 -5.02 , 7.52 -8.17 , -1.33 -9.18 , -0.32 Scheﬀe are simultaneous where as We can also construct one sided CI or one sided test by ﬁnding the one sided cutoﬀ values. Chapter 12: Multisample Inference Stat 491: Biostatistics Analysis of Variance (ANO...
View Full Document

## This note was uploaded on 02/03/2014 for the course STAT 491 taught by Professor Solomonharrar during the Fall '12 term at Montana.

Ask a homework question - tutors are online