12.4The Two-Sample Problem 531
=2(1−.9498)
=.1004
which can be compared with the exact value, as given in Example 12.4c, of .1225.
In Example 12.4d,n=9,m=13, and so
n(n+m+1)
2
=103.5
nm(n+m+1)
12
=224.25
SinceT=72, we have that
d=| 72 −103.5|=31.5
Thus, the approximatep-value is
p-value≈ 2 P
{
Z>
31.5
√
224.25
}
= 2 P{Z>2.103509}
=2(1−.9823)=.0354
which is quite close to the exactp-value (as given in Example 12.4d) of .0364.
Thus, in the two examples considered, the normal approximation worked quite well in
the second example — where the guideline that both sample sizes should exceed 7 held —
and not so well in the first example — where the guideline did not hold. ■
(b)Simulation If the observed value of the test statistic isT=t, then thep-value is
given by
p-value=2 min
{
PH 0 {T≥t},PH 0 {T≤t}
}
We can approximate this value by continually simulating a random selection ofn
of the values 1, 2,...,n+m— noting on each occasion the sum of thenvalues.
The value ofPH 0 {T≥t}can be approximated by the proportion of time that the
sum obtained is greater than or equal tot, andPH 0 {T≤t}by the proportion of
time that it is less than or equal tot.
A Chapter 12 text disk program approximates thep-value by performing the
preceding simulation. The program will run most efficiently when the sample
of smallest size is designated as the first sample.