264 Chapter 7: Parameter Estimation
approximately .05 would require an approximate sample of size
n=
4(z.005)^2
(.05)^2
26
30
(
1 −
26
30
)
=
4(2.58)^2
(.05)^2
26
30
4
30
=1,231
Hence, we should now sample an additional 1,201 chips and if, for instance, 1,040 of
them are acceptable, then the final 99 percent confidence interval forpis
(
1,066
1,231
−
√
1,066
(
1 −
1,066
1,231
)
z.005
1,231
,
1,066
1,231
+
√
1,066
(
1 −
1,066
1,231
)
z.005
1,231
)
or
p∈(.84091, .89101) ■
REMARK
As shown, a 100(1−α) percent confidence interval forpwill be of approximate lengthb
when the sample size is
n=
(2zα/2)^2
b^2
p(1−p)
Now it is easily shown that the functiong(p)=p(1−p) attains its maximum value of^14 ,
in the interval 0≤p≤1, whenp=^12. Thus an upper bound onnis
n≤
(zα/2)^2
b^2
and so by choosing a sample whose size is at least as large as (zα/2)^2 /b^2 , one can be
assured of obtaining a confidence interval of length no greater thanbwithout need of any
additional sampling. ■
One-sided approximate confidence intervals forpare also easily obtained; Table 7.3
gives the results.
TABLE 7.3 Approximate100(1−α)Percent Confidence Intervals for p
XIs a Binomial (n,p) Random Variable
pˆ=X/n
Type of Interval Confidence Interval
Two-sided pˆ±zα/2
√
pˆ(1−pˆ)/n
One-sided lower
(
−∞,pˆ+zα
√
pˆ(1−pˆ)/n
)
One-sided upper
(
pˆ−zα
√
pˆ(1−pˆ)/n,∞
)