You are on page 1of 4

Confidence Interval for Population Mean

STAT 400
July 15, 2016
Remark Recall that the CLT implies that for X1 , . . . , Xn with a common finite mean and variance
that,

X

N (0, 1) , n

(1)

using
If n is large (e.g., n > 30), we can approximate probabilities concerning the sample X
the normal distribution.

P z 2

z 2 1

(2)

where z 2 is a quantile with /2 probability in the right tail.


Rearranging terms implies,
z X
+ z
P X
2
2
n
n

(3)

z .
We construct a (1 )% confidence interval (CI) for using X
n
2
Note this approach for constructing CIs for assumes we know the population variance, 2 .
= z 2 n is the margin of error. We can assess how big n must be to yield an interval of a
certain size, i.e., .
If we have a guess for 2 and a desired interval size, , we can compute the required sample
size as, n =

z 2 2
2

STAT 400

CI for

Example 1 Suppose the lifetime of a particular brand of light bulbs is normally distributed with
standard deviation of = 75 hours and unknown mean.
is within
a. What is the probability that in a random sample of 49 bulbs, the average lifetime X
21 hours of the overall average lifetime?
( 21)
( + 21)

Z
75/ 49
75/ 49

+ 21) = P
P ( 21 X

= P (1.96 < Z < 1.96) = 0.95

(4)

= 843 hours. Construct a 95%


b. Suppose the sample average lifetime of the 49 bulbs is X
confidence interval for the overall average lifetime for light bulbs of this brand.
z/2
= 0.05, z/2 = z0.025 = 1.96, so X

= 843 21 = (822, 864).

c. Construct a 90% confidence interval for the overall average lifetime for light bulbs of this
brand.
z/2
= 0.10, z/2 = z0.05 = 1.64, so X

= 843 17.62 = (825.38, 860.62).

d. Construct a 92% confidence interval for the overall average lifetime for light bulbs of this
brand.
z/2
= 0.08, z/2 = z0.04 = 1.75, so X

= 843 18.76 = (824.24, 861.76).

e. What is the minimum sample size required if we wish to estimate the overall average lifetime
for light bulbs to within 10 hours with 90% confidence?
= 0.10, z/2 = z0.05 = 1.64. We want 10, which implies n

2 2
Z/2

1.64575
10

i2

152.2139. So we employ a sample size of n 153.

Stepanov, Culpepper

STAT 400

CI for

Example 2 How many test runs of an automobile are required for determining its average milesper-gallon rating on the highway to within 0.5 miles per gallon with 95% confidence, if a guess is
that the variance of the population of miles per gallon is about 6.25?
= 0.05, z/2 = z0.025 = 1.96. We want 0.5, which implies n

2 2
Z/2

1.962.5
0.5

i2

96.04. So we employ a sample size of n 97.


Remark: It is unrealistic to always know 2 . An alternative is to use the sample estimate,
s2 =

1
n1

i=1 (Xi

Pn

2 . If Xi N (, 2 ) for i = 1, . . . , n are mutually independent then X


and
X)

S 2 are independent and,


(n 1) S 2
2 (n 1)
2

(5)

Theorem: (Students t Distribution). Let Z N (0, 1), U 2 (r), and Z and U are independent.
Define T by the following ratio,
Z
T =q

(6)

U
r

T has a t distribution, T (r) with pdf,




r+1
1
2
 
 , < t <
f (t) =
r+1
r
r 2 1 + t2 2

(7)

As the degrees of freedom r increases, T Z (there is very little difference between the two
when r = 100).
If 2 is unavailable and n is large we can approximate probabilities regarding the sample
mean as,

X
s
n

=r

/ n
2
(n1) s 2

T (n 1)

(8)

n1

t ,n1 s .
If 2 is unknown, we construct a two-sided (1 )% CI for using X
n
2
+ t ,n1 s or X
t ,n1 s .
We can also construct 1-sided CIs using either X
n
n
2
2
3

Stepanov, Culpepper

STAT 400

CI for

Example 3 A manufacturer of TV sets wants to find the average selling price of a particular model.
A random sample of 25 different stores gives the mean selling price as $342 with a sample standard
deviation of $14. Assume the prices are normally distributed. Construct a 95% confidence interval
for the mean selling price of the TV model.
t24,0.025 S =
s = 14 and n = 25 (meaning we have 24 degrees of freedom), so X
n
342 2.063 1425 = (336.22, 347.78).
Example 4 A random sample of size n = 9 from a normal distribution is obtained: 4.4, 3.7, 5.1,
4.3, 4.7, 3.7, 3.5, 4.6, and 4.7.
a. Compute the sample mean x and the sample standard deviation s. (Hint:
P 2
x

= 168.83)

=
X

38.7
9

= 4.3, s =

r 
1
8

168.83

38.72
9

xi = 38.7 and

= 0.55.

b. Construct a 90% confidence interval for the overall (population) mean.


t8,0.05 S = 4.3 1.860 0.55
= (3.96, 4.64).
X
n
9
c. Construct a 90% one-sided confidence interval for that provides an upper bound for .
Since we are looking for an upper bound for , that means we want to have 90%
confidence that is not greater than some point, and we must find that point.
+ t8,0.1 S = 4.3 + 1.397 0.55
= 4.56
X
n
9
So 4.56 is our upper bound, and our confidence interval is (, 4.56).
d. Construct a 95% one-sided confidence interval for that provides a lower bound for .
Since we are looking for a LOWER bound this time for , that means we want to
have 95% confidence that is not less than some point, and we must find that point.
t8,0.05 S = 4.3 + 1.860 0.55
= 3.96
X
n
9
So 3.96 is our lower bound, and our confidence interval is (3.96, )
4

Stepanov, Culpepper