SURGERY RESIDENT BLOG

And review books to boot!

Biostatistics: Wilcoxon Rank Sum Test

Biostatistics: Wilcoxon Rank Sum Test

 

1. CHAPTER 13, PROBLEM 11

#11.  The characteristics of low birth weight children dying of sudden infant death syndrome were examined for both females and males. The ages at time of death for samples of 11 girls and 16 boys are shown below.

Age (days)

Females Rank Males Rank
53 3 46 1
56 4 52 2
60 7.5 58 5
60 7.5 59 6
78 10.5 77 9
87 15 78 10.5
102 16 80 12
117 20 81 13
134 22.5 84 14
160 24.0 103 17
277 27 114 18
115 19
133 21
134 22.5
167 25
175 26
157 221

a) Test the null hypothesis that the median ages at death are identical for the two populations.  What do you conclude?

HYPOTHESIS
– Hypotheses in terms of the medians of the two distributions (populations)
– Ho: Medianpop1 = Medianpop2
– Ha: Medianpop1 ≠ Medianpop2

SAMPLE INFORMATION
– Independent sample #1:
– Size n1 = 11
– (Let size n1, or sample 1 be the one with the samller sample size)

– Independent sample #2:
– Size n2 = 16

COMBINE THE DATA
THE SUMS
– W = sum of ranks in the smaller group (sum of n1 ranks)
– What is W? W = 157

TO TO TABLE A.7
– n2 = 16; – Find the table for your n2.
– n1 = 11; – Find the column for n1.
– What is your W? 157

– Pr(W≤W) = the area you see.
– What is Pr(W≤W)?

MEAN = E(W) = Uw = [n1(n1+n2+1)]/2
Uw = [11(11+16+1)]/2
Uw = 154

VARIANCE = Var(W) = σ2w = [n1n2(n1+n2+1)]/12
Var(W) = σ2w = [11X16(11+16+1)]/12
Var(W) = σ2w = [11X16(11+16+1)]/12
Var(W) = σ2w = 410.6667
SD= σw = 20.264

COMPUTE THE Z STATISTIC
–  z = [W – Uw]/ σw

z = {W – [n1(n1+n2+1)/2]} / sqrt{[n1n2(n1+n2+1)]/12}
z = [157 – 154]/ 20.264
z = 0.148 = 0.15

Find Pr(Z<z) using Table A.3 (N(0,1) distribution)
Pr(Z<z) = 0.440

CALCULATE THE P VALUE
– For a two sided test, double that probability
p < 0.880

I conclude that there is no difference between the median ages at death for males and females.

b) Do you feel that it would have been appropriate to use the two-sample t-test to analyze these data? Why or why not?

I do not feel that a two-sample t-test would be appropriate.  The data for males and females seems to be skewed to the right in terms of age of death, making it inappropriate to use the two sample t-test.

Share

Categorised as: Uncategorized


Comments are closed.

Powered by Google Talk Widget