PAIRWISE t-TEST AND ANALYSIS OF VARIANCE Emese Vágó, Sándor Kemény Budapest University of Technology and Economics Outline I. Summary of paired t-test and analysis of variance II. Similarities and differences in the two procedures III. Comparison of the models IV. Comment IV. Conclusions Example 1 2 3 After diet 65 45 78 Before diet 70 49 86 … … … People The effect of a diet on weight was investigated using ten people. Results: 10 79 85 1 2 3 y 2n H 0 : E (d ) 0 n d d … y 1n … n G ro u p d i= y 1 i- y 2 i 1 2 y 11 y 21 d1 y 12 y 22 d2 y 13 y 23 d3 … … Pairs Paired t-test i 1 n i n s 2 d dn d E (d ) sd / n d (n 1)n n (d i 1 i d) ~ t n -1 (d i 1 i d) n 1 Analysis of variance 1 2 3 2 y 11 y 12 y 13 y 21 y 22 y 23 … n 1 … … Random effect (B) F ix e d e ffe c t (A ) y 1n y 2n yij: weight of jth men before(i=1)/after(i=2) diet r: levels of A (2) n: levels of B (10) r MS A MS AB y i 1 y r n i 1 j 1 y.. 2 i. yi. y. j y.. 2 ij ( n 1)n ~ F1,n-1 Similarities and differences t 2 n 1 F1,n1 ? Conditions of the two analysis differ in spite of the same statistics. Why is it so? ? Same statistic Same model Assumptions - paired t-test d j ~ N d , 2 d 1 2 2 d 2 1 2 2 Product moment correlation, can not be estimated ~ N , y1 j ~ N 1 , y2 j 2 2 1 2 2 y1 j 1 1 j 1 j ~ N (0, ) y2 j 2 2 j 2 j ~ N (0, ) 2 1 Assumptions of independence for d j , y1 j , y2 j 2 2 Assumptions - analysis of variance yij i j ij ij ij ~ N (0, e2 ) j ~ N (0, b2 ) 2 yij ~ N ( i , b2 ab e2 ) Assumptions of independence for - yij - levels of B 2 ij ~ N (0, ab ) Comparison of assumptions Paired t-test Analysis of variance yij ~ N (i , ) yij ~ N ( i , ) 2 b 2 i 2 ab i i If If 2 1 2 2 12 22 2 2 2 b2 ab e2 the models do not correspond 2 e Comparison of assumptions ANOVA method may be used with weaker assumptions about homogeneity of variances for: - one random, and - one fixed effect (number of levels is two) design - sample sizes are one - testing only the effect of fixed factor Comment - distribution of the random effect d ~ N d , d2 yij ~ N 1 , 2 i yij ~ ? d2 12 22 1 2 y1 j 1 1 j y2 j 2 2 j 1 j 1 j y1 j 2 j 2 j y2 j ij ~ N (0, i2 ) ij ~ N (0, i2 ) d j 1 2 1 j 2 j Comment - sample sizes larger then one Sample size: 1 Sample size: q y1 jk 1 j 1 jk y2 jk 2 j 2 jk 1 j 1 j y1 j 2 j 2 j y2 j y1 j 1 j 1 j . y2 j . 2 j 2 j . ij ~ N (0, i ) ij . ~ N (0, i2 / q) d j 1 2 1 j 2 j d j 1 2 1 j . 2 j . 2 Consequence ANOVA method may be used with weaker assumptions about homogeneity of variances for testing the effect of a 2-level fixed factor, if the other factor is - random, or fixed and - sample sizes are one, or larger then one. THANK YOU FOR YOUR ATTENTION !
© Copyright 2026 Paperzz