Confidence intervals for a shift. [Maritz (1979)] (i) Let X1,..., Xm; Y1,..., Yn be independently distributed according
Question:
Confidence intervals for a shift. [Maritz (1979)]
(i) Let X1,..., Xm; Y1,..., Yn be independently distributed according to continuous distributions F(x) and G(y) = F(y − ) respectively. Without any further assumptions concerning F, confidence intervals for can be obtained from permutation tests of the hypotheses H(0) : = 0. Specifically, consider the point (z1,...,zm+n) = (x1,..., xm, y1 − ,... , yn − ) and the m+n m
permutations i1 < ··· < im; im+1 < ··· < im+n of the integers 1,..., m + n. Suppose that the hypothesis H() is accepted for the k of these permutations which lead to the smallest values of
m
+n j=m+1 zi j /n −m j=1 zi j /m
, where k = (1 − α)
m + n m
.
Then the totality of values for which H() is accepted constitute an interval, and these intervals are confidence intervals for at confidence level 1 − α.
(ii) Let Z1,..., ZN be independently distributed, symmetric about θ, with distribution F(z − θ), where F(z) is continuous and symmetric about 0. Without any further assumptions about F, confidence intervals for θ can be obtained by considering the 2N points Z
1,..., Z
N where Z
i = ±(Zi − θ0), and accepting H(θ0) : θ = θ0 for the k of these points which lead to the smallest values of
|
Z
i|, where k = (1 − α)2N .
[(i): A point is in the acceptance region for H() if (yj − )
n −
xi m =|¯y − ¯x − |
is exceeded by at least m+n n
− k of the quantities | ¯y − ¯x − γ|, where (x
1,..., x
m, y
1,..., y
n) is a permutation of (x1,..., xm, y1,..., yn), the quantity γ is determined by this permutation, and |γ| ≤ 1. The desired result now follows from the following facts (for an alternative proof, see Section 14):
(a) The set of ’s for which (y¯ − ¯x − )2 ≤ (y¯ − ¯x − γ)2 is, with probability one, an interval containing y¯ − ¯x.
(b) The set of ’s for which (y¯ − ¯x − )2 is exceeded by a particular set of at least m+n m
− k of the quantities (y¯ − ¯x − γ)2 is the intersection of the corresponding intervals
(a) and hence is an interval containing y¯ − ¯x.
(c) The set of ’s of interest is the union of the intervals
(b) and, since they have a nonempty intersection, also an interval.]
Section 5.10
Step by Step Answer:
Testing Statistical Hypotheses Volume I
ISBN: 9783030705770
4th Edition
Authors: E.L. Lehmann, Joseph P. Romano