MATH 524, Fall 2024
Nonparametric Statistics
Second assignment, due Monday, October 21, 2024, noon
1. In a study of the comparative tensile strength of tape-closed and su- tured wounds, the following results were obtained on 10 rats, 40 days after incisions on their backs had been closed by suture or by surgical tape. [These data are from a paper by Ury and Forrester published in The American Statistician, vol. 24 (1970), pp. 25–26].
Rat number: 1 2 3 4 5 6 7 8 9 10
Tape:
|
659
|
984
|
397
|
574
|
447
|
479
|
676
|
761
|
647
|
577
|
Suture:
|
452
|
587
|
460
|
787
|
351
|
277
|
234
|
516
|
577
|
513
|
Test the hypothesis of no effect against the alternative that the tape- closed wounds are stronger using the sign test and Wilcoxon’s signed- rank test. Without showing detailed calculations, state the results of the same tests when the tensile strength of each of the taped wounds
is decreased by a) 5 units; b) 10 units. Comment.
2. In the above study of the effect of tape closing on wounds, use the Nor- mal approximation to determine for what values of Wilcoxon’s signed- rank test statistic the null hypothesis of no effect should be rejected at
the 5% significance level when a) N = 20; b) N = 40; c) N = 60.
3. Suppose that in the comparison of a new headache remedy with a standard one, the expressions of preference for the new drug by nine subjects are as follows:
The new remedy is...
much more efficient
|
1
|
somewhat more efficient
|
4
|
no better nor worse
|
2
|
somewhat less efficient
|
1
|
much less efficient
|
1
|
... than the standard remedy
Show how Wilcoxon’s signed-rank test statistic, Vs* , can be used in this case, and find its p-value.
4. The following data report the weight (in lbs) that 12 first-graders were able to lift before and after an 8-week muscle-training program. [These data are from a paper by Schweid, Vignos, and Archibald in the Amer- ican Journal of Physical Medicine, vol. 41 (1962), pp. 189–197.]
Before:
|
14.4
|
15.9
|
14.4
|
13.9
|
16.6
|
17.4
|
After:
|
20.4
|
22.9
|
19.4
|
24.4
|
25.1
|
20.9
|
Before:
|
18.6
|
20.4
|
20.4
|
15.4
|
15.4
|
14.1
|
After:
|
24.6
|
24.4
|
24.9
|
19.9
|
21.4
|
21.4
|
Determine the values of the estimators θ(¯), θ(˜), and θ(ˆ) of θ, under the
assumption that Pr(D ≤ x) = L(x−θ) can be expressed in terms of the cumulative distribution function L of a distribution that is symmetric with respect to the origin.
5. Prove that the power function Π(∆) = Pr(Vs ≥ v | ∆) of Wilcoxon’s signed-rank test is non-decreasing in ∆ and such that if the nominal level of the test is comprised between 2 −N and 1 − 2−N , Π(∆) → 0 or
1 as ∆ → −∞ or +∞, respectively.
6. The shift model for paired data (X, Y) consists in assuming that there exists a constant ∆ ∈ [0, ∞) for which the distribution of Y − ∆ is the same as the distribution of X . In this context, it can be shown that the Pitman efficiency of the sign test with respect to Wilcoxon’s signed-rank test is given by
where Z = Y − X has distribution L under H0 : ∆ = 0 and ℓ denotes the corresponding density.
a) Compute eS,V (L) when L is Cauchy, Normal, and Uniform.
b) Show that eS,V (L) ≥ 1/3 when L is unimodal.
c) By considering densities defined, for all z ∈ R and α ∈ (0, ∞), by
show that eS,V (L) can be arbitrarily large, and interpret the result.