Scribd
Upload
41 views8 pages

Lecture11 - Nonparametric Two - Sample Tests PDF

1) Nonparametric methods, such as the Wilcoxon rank sum test and Wilcoxon signed rank test, can be used for two-sample comparisons when the underlying distributions are non-normal. 2) The Wilcoxon rank sum test compares the ranks of two independent samples and tests if their population medians are the same. The Wilcoxon signed rank test compares the ranks of differences between paired samples and tests if the population medians of the differences are equal to zero. 3) Both tests calculate a test statistic and use its expected value and variance under the null hypothesis to derive a Z-score and p-value for hypothesis testing.

Uploaded by

nurul
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
41 views8 pages

Lecture11 - Nonparametric Two - Sample Tests PDF

1) Nonparametric methods, such as the Wilcoxon rank sum test and Wilcoxon signed rank test, can be used for two-sample comparisons when the underlying distributions are non-normal. 2) The Wilcoxon rank sum test compares the ranks of two independent samples and tests if their population medians are the same. The Wilcoxon signed rank test compares the ranks of differences between paired samples and tests if the population medians of the differences are equal to zero. 3) Both tests calculate a test statistic and use its expected value and variance under the null hypothesis to derive a Z-score and p-value for hypothesis testing.

Uploaded by

nurul
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 8

Lecture 11.

Two-sample Comparison (II):


Nonparametric method
兩個母群體中位數之比較: 無母數之方法

Nonparametric method
非參數化之方法；無母數之方法

對應前述之參數化方法：t-test, paired t-test (assuming

normal distribution(s))
When the underlying distributions are very similar, but both

are far from normal (Gaussian) [e.g., right-skewed],

nonparametric methods can be applied.

Population

m1 m2

Data (sample)

××× × ×× × × × × × × ×

× ×× ×× × × × ××
One-tailed and Two-tailed test
˙ One-tailed
H0: m1≦m2 (null) vs. Ha: m1> m2 (alternative)
˙ Two-tailed
H0: m1＝m2 vs. Ha: m1 ≠m2

Note: comparison between parametric and


nonparametric methods for two-sample problem.

Independent Paired samples

samples

parametric T test Paired t-test

nonparametric Wilcoxon Wilcoxon

rank sum test signed rank test


˙
Procedure：
▪ To construct the test statistic:
▪ Sampling distribution of the test statistic under Ho
▪ Significance level (1-α)
▪ Type I error (α) and the p-value
Independent samples

Data:

▪ X1,X2,…,Xn ; Y1,Y2,…,Ym
▪ Pooled sample: Z1＜Z2＜Z3＜Z4＜…＜Zn+m
▪ Rank: 1,2,3,4,…,(n+m)
(Wilcoxon’s) rank sum test:
If the smaller rank sum (of X-group or Y-group)=W, and the

sample size of the corresponding group is nS (the other is nL).

If nS=n, then nL=m; if nS=m, then nL=n.

Under null (H0)


E(W)=[( nS＋nL) ×( nS＋nL+1)/2] ×[nS/(nS＋nL)]

=nS×( nS＋nL+1)/2；

Var(W)=nSnL(nS＋nL+1)/12 (Proof: see Appendix11.1)

SE(W)=√[nSnL(nS＋nL+1)/12] (SE=standard error)

Approximate Z-score:

[W-E(W)]÷SE(W)~N(0,1), approximately, when n and m are

both large.

Example: (Ref. to Table 13.3)


W=313
E(W)=18×(18+21+1)÷2=360
Var(W)=18×21×(18+21+1)÷12=1260
SE(W)= √(1260)=35.5
ZW=(313-360)÷35.5=-1.32;
P value=2×0.093=0.186 ……….
Paired samples

Data: Table 13.2

X1,X2,…,Xn ;

Y1,Y2,…,Yn

Difference: d1,d2,…,dn ; di = Xi－Y i

Rank r1,r2,…,rn (for |di|)

Sign + + - -…+ (+: for di positive;

- : for di negative)
(Wilcoxon’s) signed rank test
If the smaller |rank sum|=T, (absolute value of rank sum)
Under null (H0)
E(T)=[n(n+1)/2]÷2=n(n+1)/4;
Var(T)=n(n+1)(2n+1)/24 (Homework) (Proof: see below*)
SE(T)=√[n(n+1)(2n+1)/24] (SE=standard error)
Approximate Z-score:
[T-E(T)]÷SE(T)~N(0,1), approximately, when n is large.

Example: (Table 13.2)


T=19
E(T)=14×15÷4=52.5
Var(T)=14×15×29÷24=253.75
SE(T)=15.93
ZT=(19-52.5)÷15.93=-2.10;
P value=2×0.018=0.036 ……….

Proof(★): Let W=ΣUi, where Ui=0 (with probability=1/2), and=ri (with


probability=1/2). It is then interesting to note that: the W defined in this
way have the same distribution with the statistic T. Also, {Ui} are
independent of each other because the outcome of Ui is independent of
that of Uj for i≠j. So,
E(W)=Σ(EUi)=Σ(0+i/2)=(1/2)Σi i=(1/2)(n(n+1)/2)= n(n+1)/4;
Var(W)= Σ(VarUi)= Σ(EUi2-( EUi) 2)= Σ(i2/2-( i/2) 2)= Σ(i2/4)
=(1/4)( n(n+1)(2n+1)/6) QED
SAS code

output
Appendix 11.1 (★)

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy