Chapter 9: Sampling Distributions

9-1 Sampling Distribution of the Mean

A sampling distribution is created by sampling
a population.
There are two methods to create a sampling
distribution:
1. Theoretical method: apply the rules of
probability and the laws of expected value
and variance to derive the sampling
distribution from sample statistics.
2. Empirical method: draw several samples of
the same size from a population, calculate
the statistic of interest, and then use
descriptive techniques to build the sampling
distribution.
In this chapter we describe the former method
and then briefly demonstrate the latter one.
9-1a Sampling Distribution of the Mean of
Two Dice
Population Distribution of One Die
 ! ! !
𝜇 = ∑ 𝑥𝑝(𝑥 ) = 1 + 2 + ⋯ 6 = 3.5
" " "

!
𝜎 # = ∑(𝑥 − 𝜇)# 𝑝(𝑥) = (1 − 3.5)# + (2 −
"
! !
3.5)# + ⋯ (6 − 3.5)# = 2.92
" "

𝜎 = 3𝜎 # = √2.92 = 1.71
Sampling Distribution of the Mean of Two
Dice

! # "
𝜇$̅ = ∑ 𝑥̅ 𝑝(𝑥̅ ) = 1.0 + 1.5 + ⋯ 3.5 +
&" &" &"
!
⋯ 6.0 = 3.5
&"
 !
𝜎$̅# = ∑(𝑥̅ − 𝜇$̅ )# 𝑝(𝑥̅ ) = (1.0 − 3.5)# +
&"
! "
(1.5 − 3.5)# + ⋯ (3.5 − 3.5)# + ⋯ (6.0 −
&" &"
!
3.5)# = 1.46
&"

𝜎$̅ = 9𝜎$̅# = √1.46 = 1.21

9-1a Central Limit Theorem

From the sampling distribution of the mean of
two dice, we learned:
• The distribution of 𝑋 is different from the
distribution of 𝑋;.
• The mean of 𝑋 is the same as the mean of
𝑋;:

𝜇$̅ = 𝜇
• The variance of 𝑋; is half the variance of 𝑋:
'!
𝜎$̅# =
#

It can be further demonstrated that, for a

sample of size n:
 • as the sample size increases, the
distribution of 𝑋; narrows.
• the relation between the variances
becomes:

𝜎#
𝜎$̅# =
𝑛
These findings can be generalized in the
Central Limit Theorem:
The sampling distribution of the mean of a
random sample drawn from any population is
approximately normal for a sufficiently large
sample size. The larger the sample size, the
more closely the sampling distribution of 𝑋; will
resemble a normal distribution
9-1a Sampling Distribution of the Sample
Mean
The sampling distribution of the sample mean
for large populations can be summarized as:

1. 𝜇$̅ = 𝜇
2. If the population is infinite:
 𝜎
𝜎$̅ =
√𝑛
3. If the population is finite, then the finite
population correction factor* is applied:

𝜎 𝑁−𝑛 𝑁D < 20
𝜎$̅ = = when 𝑛
√𝑛 𝑁 − 1

4. If 𝑋 is normal, then 𝑋; is normal. If 𝑋 is

nonnormal, then 𝑋; is approximately
normal for sufficiently large sample sizes.
The definition of “sufficiently large” depends
on the extent of nonnormality of 𝑋.
*See notes for additional details on the finite
population correction factor.
The source of the correction factor is provided
in the online appendix Hypergeometric
Distribution. An analysis reveals that if the
population size is large relative to the sample
size, then the finite population correction factor
is close to 1 and can be ignored. As a rule of
thumb, we will treat any population that is at
least 20 times larger than the sample size as
large. In practice, most applications involve
populations that qualify as large because if the
population is small, it may be possible to
investigate each member of the population, and
in so doing, calculate the parameters precisely.
As a consequence, the finite population
correction factor is usually omitted.
9-1b Creating the Sampling Distribution
Empirically
We can create the sampling distribution of the
mean of two dice empirically by:
1. Tossing two fair dice repeatedly.
2. Calculating the sample mean for each
sample.
3. Counting the number of times each value of
X occurs.
4. Computing the relative frequencies to
estimate theoretical probabilities.
Obviously, such empirical approach is not
practical because of the excessive amount of
time required to toss the dice enough times to
make the relative frequencies good
approximations for the theoretical probabilities.
However, we can use Excel to simulate samples
from a variety of different populations making it
easy for us to create many kinds of sampling
distributions (see Exercises 9.1-9.8 for a few
applications.)
*See notes for Excel instructions.
Instructions:
We can use Excel to approximate the theoretical
sampling distribution. We’ll start with the
sampling distribution of the mean of the toss of
two dice.
1. Set up the distribution of the toss of one
die. In Column A, store the numbers 1, 2, 3,
4, 5, 6 and in Cell B1 type:
= 1/6
Do not type .1667 or any other rounded
version since the sum of the probabilities will
not equal 1 causing Excel to issue an error
warning at step 4.
Drag to fill cells B2–B6.
2. Click Data, Data Analysis, and Random
Number Generation.
3. Type 2 to specify the Number of Variables
and type 10000 to specify the Number of
Random Variables.
4. Click Discrete distribution and in the
Parameters box type A1:B6 to specify the
Value and Probability Input Range.
5. Specify New Worksheet Ply and click OK.
Columns A and B of the new work-sheet ply
will fill with the random numbers.
6. In column C row 1, type
=AVERAGE(A1:B1)
7. Drag to fill the rest of Column C. Column C
will now contain the values of the sample
means.
Example 9.1 – Contents of a 32-Ounce Bottle
The amount of soda in each 32-ounce bottle is
a normally distributed random variable, with
mean of 32.2 oz and standard deviation of 0.3
oz.
What is the probability that:
1. A single bottle contains more than 32 oz?
2. The mean amount of four bottles is greater
than 32 oz?
We define:
X = amount of soda per bottle (normal)
μ = 32.2 oz, σ = 0.3 oz
Solution
1. We need to find 𝑃(𝑋 > 32):
𝑋 − 𝜇 32 − 32.2
𝑃(𝑋 > 32) = 𝑃 H > I
𝜎 0.3
= 𝑃 (𝑍 ≻ −0.67) = 1 − 𝑃(−0.67)
= 1 − .2514 = .7486
1. We need to find 𝑃(𝑋; > 32) with n = 4:

𝜎 0.3
𝜇$̅ = 𝜇 = 32.2 and 𝜎$̅ = = = 0.15
√𝑛 √4
𝑋; − 𝜇$̅ 32 − 32.2
;
𝑃 (𝑋 > 32) = 𝑃 O > P
𝜎$̅ 0.15

= 𝑃 (𝑍 ≻ −1.33) = 1 − 𝑃(−1.33)
= 1 − .0918 = .9082
9-1c Salaries of a Business School’s
Graduates
Chapter Opener – Example
In the advertisements for a large university, the
dean of the School of Business claims that the
average salary of the school’s graduates one
year after graduation is $800 per week with a
standard deviation of $100.
A student surveys 25 graduates and calculates a
mean salary of $750 per week.
Solution: To verify the dean’s claim, we need to
calculate 𝑃(𝑋; < 750). The distribution of 𝑋, the
weekly income, is likely to be positively skewed,
but not sufficiently so to make the distribution
of 𝑋; nonnormal. Thus, we may assume that 𝑋 is
normal with:

𝜇$̅ = 𝜇 = 800, and 𝜎$̅ = 𝜎⁄√𝑛 =

100⁄√25 = 20
𝑋; − 𝜇$̅ 750 − 800
𝑃(𝑋; < 750) = 𝑃 O < P
𝜎$̅ 20
= 𝑃(𝑍 < −2.5) = .0062
The probability is very small. The dean’s claim is
not justified.
9-1c Using the Sampling Distribution for
Inference
In preparation for next chapter on interval
estimation, we present another way of
expressing the probability associated with the
sampling distribution.
In Section 8-2, we defined z.025 to be the value of
Z such that the area to the right of z.025 under
the standard normal curve is equal to .025. We
calculated z.025 = 1.96.
Because the standard normal distribution is
symmetric about 0, the area to the left of −1.96
is also .025.
Thus, the area between −1.96 and 1.96 is .95,
that is:

𝑃(−1.96 < 𝑍 < 1.96) = .95

Because the sampling distribution is normally
distributed, we can substitute the equation for Z
into the previous probability statement.

𝑋; − 𝜇$̅
𝑃 O−1.96 < < 1.96P = .95
𝜎 ⁄√𝑛
Using algebra, we produce:

𝑃R𝜇$̅ − 1.96 𝜎⁄√𝑛 < 𝑋; < 𝜇$̅ + 1.96 𝜎 ⁄√𝑛S

= .95
If we now plug in mean and standard deviation:

100 100
𝑃 H800 − 1.96 < 𝑋; < 800 + 1.96 I
√25 √25
= .95
𝑃(760.8 < 𝑋; < 839.2) = .95
We can conclude that there is a 95% probability
that a sample mean of size 25 will fall between
$760.8 and $839.2, which does not include the
sample mean of $750. Thus, the dean’s claim is
not supported.
9-2 Sampling Distribution of a Proportion
In Section 7-4, we introduced the binomial
distribution, described by the parameter p, the
probability of success in any trial.
When p is unknown, we estimate it with the
sample proportion (read as p hat):
𝑋
𝑃T =
𝑛
Where X is the count of successes and n the
sample size. The random variable X is binomially
distributed.
However, a binomial random variable is
discrete, and it does not lend well to the
calculations needed for inference. Next, we
show how to approximate a discrete binomial
random variable with a continuous normal
random variable.
9-2a Binomial Distribution Example
Consider a binomial distribution with n = 20,
and p = .5.
The probability that X = 10, can be written as:

𝑛!
𝑃(𝑋 = 10) = 𝑝 ( (1 − 𝑝)(*+()
𝑛! (𝑛 − 𝑋)!
20!
= 𝑝!- (1 − 𝑝)(#-+!-)
10! (20 − 10)!
= .1762
From Section 7-4, we also learned that the
binomial distribution parameters are:

𝜇 = 𝑛𝑝 = 20(. 5) = 10

𝜎 = 3𝑛𝑝(1 − 𝑝 = 320(. 5)(.5) = √5 = 2.236

9-2a Normal Approximation to the Binomial
Distribution
To approximate the probability that X = 10
using the normal distribution requires that we
find the area under the normal curve between
9.5 and 10.5 (*see notes):

𝑃(𝑋 = 10) ≈ 𝑃(9.5 < 𝑌 < 10.5)

where Y is a normal random variable that
approximates the binomial random variable X.
We standardize Y to find:
 𝑃(9.5 < 𝑌 < 10.5)
9.5 − 10 𝑌 − 𝜇 10.5 − 10
= 𝑃H < < I
2.236 𝜎 2.236
= 𝑃(−0.2236 < 𝑍 < 0.2236)
= 𝑃(𝑍 < 0.2236) − 𝑃(𝑍 < −0.2236)
= .5885 − .4115 = .1770
Which is an excellent approximation.

Continuity correction factor:

To draw a binomial distribution, which is
discrete, it was necessary to draw rectangles
whose bases were constructed by adding and
subtracting .5 to the values of X. The .5 is called
the continuity correction factor.
9-2b Omitting the Correction Factor for
Continuity
When calculating the probability of individual
values of X as we did when we computed the
probability that X equals 10 earlier, the
continuity correction factor must be used.
When computing the probability of a range of
values of X, we can omit the correction factor.
However, the omission of the correction factor
will considerably decrease the accuracy of the
approximation when the value of X is near the
center of the distribution, but it can be tolerated
for large sample sizes or when the calculations
involve the tails of the distribution.
Example:
Consider the calculations (using Excel) of P(X ≤
8) applied to the previous example.
Binomial distribution: 𝑃 (𝑋 ≤ 8) = .2517
Normal approximation
with the correction factor: 𝑃(𝑌 < 8.5) =
𝑃 (𝑍 < −0.894) = .1857
without the correction factor: 𝑃 (𝑌 < 8) =
𝑃 (𝑍 < −0.671) = .2511
9-2c Approximate Sampling Distribution of a
Sample Proportion
Using the laws of expected value and variance
(see the online appendix Using the Laws of
Expected Value and Variance to Derive the
Parameters of Sampling Distributions), we can
determine the mean, variance, and standard
deviation of 𝑃T:
Sampling Distribution of a Sample
Proportion
𝑃T is approximately normally distributed
provided that 𝑛𝑝 ≥ 5 and 𝑛(1 − 𝑝) ≥ 5
The expected value: ER𝑃TS = 𝑝
 0(!+0)
The variance: VR𝑃T S = 𝜎.#/ =
*

0(!+0)
The standard deviation: 𝜎./ = 9
*

The standard deviation of 𝑃T is called the

standard error of the proportion.
Example 9.2 – Political Survey
In the last election, a state representative
received 52% of the votes cast. One year after
the election, the representative organized a
survey that asked a random sample of 300
people how they would vote in the next
election. If we assume that the representative’s
popularity has not changed, what is the
probability that more than half of the sample
would vote to re-elect?
Solution:
The number of respondents who would vote for
the representative is a random binomial variable
with n = 300, and p = .52.
We want to find: 𝑃R𝑃T > .50S.
𝑃T is approximately normal with ER𝑃TS = 𝑝 = .52
0(!+0) .(2#)(.34)
and 𝜎./ = 9 =9 = .0288
* &--

𝑃T − 𝑝 . 50 − .52
𝑃R𝑃T > .50S = 𝑃 O > P
𝜎./ . 0288
= 𝑃(𝑍 ≻ 0.69) = 1 − 𝑃(𝑍 < 0.69)
= 1 − .2451 = .7549
If the level of support has not changed, the
probability of a favorable sample exceeds 75%.
9-3 From Here to Inference
Probability Distribution
In Chapters 7 and 8, we used knowledge of the
population and the parameter(s) of a probability
distribution to make probability statements
about individual members of the population.
Population & Parameters ⇒ Probability
Distribution ⇒ Individual
Sampling Distribution
In this chapter, we developed the sampling
distribution, wherein knowledge of the
parameter(s) and some information about the
distribution allow us to make probability
statements about a sample statistic.
Population & Parameters ⇒ Sampling
Distribution ⇒ Statistic
Statistical Inference
Starting in Chapter 10, we will assume that most
population parameters are unknown. A sample
drawn from the population will provide the
required statistic, and the sampling distribution
of such statistic will enable us to draw inferences
about the parameter.
Statistic ⇒ Sampling Distribution ⇒ Parameter
Chapter Summary
• The sampling distribution of a statistic is
created by repeated sampling from one
population.
• In this chapter, we introduced the sampling
distribution of the mean and the
proportion.
 • We described how these distributions are
created theoretically and empirically.
Chapter 10: Introduction to Estimation
10-1a Point and Interval Estimators
Point Estimator
A point estimator draws inferences about a
population by estimating the value of an
unknown parameter using a single value or
point.
Drawbacks:
1. Because the probability that a continuous
random variable will equal a value is 0, it is
virtually certain the estimate will be wrong.
2. A point estimator does not reveal how close
the estimate is to the unknown population
parameter.
3. A point estimator does not have the
capacity to reflect the effects of larger
sample sizes.
Interval Estimator
An interval estimator draws inferences about a
population by estimating the value of an
unknown parameter using a range of values.
Properties:
1. An interval estimator is an unbiased
estimator because its expected value
equals the unknown population parameter.
2. An unbiased estimator is consistent
because the difference between the
estimator and the parameter grows smaller
as the sample size grows larger.
3. An unbiased estimator also has relative
efficiency when it has a variance smaller
than other unbiased estimators.
10-2a Identify-Compute-Interpret System
The approach Identify-Compute-Interpret
System (ICI) illustrates the process of solving
statistical problems. The process is divided into
three stages:
1. To begin, we identify the appropriate
 statistical method to employ.
2. Next, we compute the statistics using one
of the four approaches to Microsoft Excel as
described in Chapter 1:
• Statistical functions
• Analysis ToolPak
• Spreadsheets
• Do It Yourself
3. Finally, we interpret the results.
10-2a Estimating the Population Mean when
the Population Standard Deviation is Known
Consider the task of estimating the unknown
mean μ of a population that is described by a
continuous random variable X and has standard
deviation σ.
The first step in the estimation procedure is to
draw a random sample of size n and calculate
the sample mean 𝑥̅ .
Because of the central limit theorem, if X is
normally distributed or n is sufficiently large,
then 𝑋; is normally distributed. Thus, as shown
in Section 9-1, we can build the following
probability statement associated with the
sampling distribution of the mean:

𝑃R𝜇 − 𝑍5⁄# 𝜎⁄√𝑛 < 𝑋; < 𝜇 + 𝑍5⁄# 𝜎⁄√𝑛S

=1−𝛼
Simple algebra manipulation yields:

𝑃R𝑋; − 𝑍5⁄# 𝜎⁄√𝑛 < 𝜇 < 𝑋; + 𝑍5⁄# 𝜎⁄√𝑛S

=1−𝛼
The equation says that, with repeated sampling
from this population, the proportion of values of
𝑋; for which the interval includes the population
mean μ is equal to 1 − α.
10-2a Confidence Interval Estimator of μ
when σ Is Known
' '
𝑥̅ − 𝑧5⁄# , 𝑥̅ + 𝑧5 ⁄#
√* √*

The probability 1 − 𝛼 is called the confidence

level.
 '
𝑥̅ − 𝑧5⁄# is called the lower confidence
√*

limit (LCL).
'
𝑥̅ + 𝑧5⁄# is called the upper confidence
√*
limit (UCL).
We often represent the confidence interval
estimator of μ as:
𝜎
𝑥̅ ± 𝑧5 ⁄#
√𝑛
where the minus sign defines LCL, and the plus
sign defines UCL.
The table to the right shows four commonly
used confidence levels and 𝑧5⁄# .

1− α α/ zα / 2
α 2

.90 .10 .05 1.645

.95 .05 .025 1.960

.98 .02 .01 2.326

.99 .01 .005 2.576

Example 10.1 – Doll Computer Company
To lower inventory cost, Doll’s operations
manager wants to estimate the mean computer
demand during lead time at a 95% confidence
level.
A sample of 25 lead times is collected and
demand during each period recorded. Assume
that the manager knows that the standard
deviation of demand is 75 computers.
DATA file: Xm10-01
Identify
The manager should build a confidence interval
estimator of μ when σ is known:
𝜎
𝑥̅ ± 𝑧5 ⁄#
√𝑛
We have: 𝑛 = 25, and 𝜎 = 75.
Compute

𝑥̅ = b 𝑥8 D𝑛 = b 9,254⁄25 = 370.16

1 − 𝛼 = .95 ⇒ 𝛼 = .05 ⇒ 𝛼 ⁄2 = .025

From Table 3, Appendix B: 𝑧5⁄# = 𝑧.-#2 = 1.96
Thus, the confidence interval estimator is:

𝜎 75
𝑥̅ ± 𝑧5 ⁄# = 370.16 ± 1.96
√𝑛 √25
= 370.16 ± 29.40
LCL = 340.76, UCL = 399.56
Interpret
The estimate of (340.76, 399.56) computers can
be used as an input in developing an inventory
policy.
*See notes for Excel instructions.
Instructions:
1. Type or import the data into one column.
(Open Xm10-01.) In any empty cell,
calculate the sample mean
(=AVERAGE(A1:A26).
2. Open the Estimators Workbook and click
the z-Estimate_Mean tab. In cell B3, type
or copy the value of the sample mean. If
you use Copy also use Paste Special and
 Values. In cells B4–B6, type the value of s
(75), the value of n (25), and the confidence
level (.95), respectively.
10-2b Interpreting the Confidence Interval
Estimate
Incorrect interpretation
“There is a (1 − α)% probability that the
population parameter lies between LCL and
UCL.”
This interpretation is wrong because it implies
that we can make probability statements about
a population parameter, which is not a variable
but a fixed and unknown quantity.
Correct interpretation
"If we repeatedly draw samples of the same size
from a population, (1 − α)% of the interval
estimates will include the unknown population
parameter and α% will not."
10-2c Information and the Width of the
Interval
Interval estimation is as useful as it is efficient,
and a narrower interval provides more
information than a wider one.
The width of the interval estimate is a function
of the population standard deviation σ, the
confidence level (1 − α)%, and the sample size n.
𝜎
𝑥̅ ± 𝑧5 ⁄#
√𝑛
How can we reduce the interval’s width?
• The sample mean 𝑥̅ only affects the point
estimate, not the width.
• σ is a parameter of the population, and we
have no control over it.
• We could decrease 1 − α, and therefore
𝑧5⁄# , but then we would have less
confidence on the inference.
• We can always increase the sample size n,
but there may be practical limitations to it.
10-2e Computer Simulation of a Confidence
Interval Estimator of μ Using Excel
Simulation of a 95% confidence interval estimate
of the population mean. Simulation
characteristics: 1,000 samples, n = 9, normal
population with μ = 5, and σ = 1.
1. Click Data, Data Analysis, and Random
Number Generation.
2. Type 9 to specify the Number of Variables
and type 1000 to specify the Number of
Random Numbers.
3. Click Normal in the Distribution box and in
the Parameters box type Mean 5 and
Standard Deviation 1.
4. Specify New Worksheet Ply and click OK.
Columns A to I of the new worksheet ply
will fill with the random numbers.
5. In Cell J1, type =AVERAGE(A1:I1), and drag
to fill the rest of Column J. Column J will
now contain the values of the sample
means.
6. In Cell K1, calculate the lower limit of the
 95% confidence interval estimate. Drag to fill
the rest of Column K. These are the LCLs.
7. In Cell L1, calculate the upper limit of the
95% confidence interval estimate. Drag to fill
the rest of Column L. These are the UCLs.
8. To instruct Excel to determine which
intervals contain μ in Cell M1, type
=AND(K1<=5,L1>=5). Drag to complete
Column M. When TRUE, μ is included.
9. To count the number of intervals that
exclude the population mean 5, in Cell N1,
type =IF(M1=TRUE,0,1) and drag to fill
Column N.
10. Sum Column N and you will have the
total number of 95% confidence interval
estimates that are “wrong.”
10-3a Error of Estimation
We define the error of estimation as the
difference between an estimator and the
parameter. In this chapter, the error of
estimation is expressed as the difference
between 𝑋; and μ:
With a simple algebraic manipulation, we
express the confidence interval as:
𝜎
|𝑋; − 𝜇| = 𝑧5⁄#
√𝑛
We interpret this to mean that 𝑧5⁄# 𝜎⁄√𝑛 is the
maximum error of estimation we are willing to
tolerate. We label it B, the bound error of
estimation, that is:
𝜎
𝐵 = 𝑧5⁄#
√𝑛
10-3b Determining the Sample Size
If the population standard deviation σ, the
confidence level (1 − α)%, and the bound on the
error of estimation B are known, we can solve
for n and produce the required sample size to
estimate a mean:
𝑧5⁄# 𝜎 #
𝑛=e f
𝐵
Consider that, in Example 10.1, before collecting
the data, the manager needs to estimate the
mean demand during lead time within 16 units.
Remember also that 1 − α = .95, and σ = 75.
Thus:

𝑧5⁄# 𝜎 # 1.96 ∙ 75 #
𝑛=e f =H I = 84.41
𝐵 16
Because the sample size must be an integer and
we want the bound on the error of estimation
to be no more than 16, any non-integer value
must be rounded up. Thus, we round n to 85.
Appendix 10.A – XLSTAT Output and
Instructions
Example 10.1 – Confidence Interval Estimate
of a Mean
1. Type or import the data into one column.
(Open Xm10-01.)
2. Click XLSTAT, Parametric tests, and One-
sample t-test and z-test.
3. Check One sample under Data format: In
 the
Data dialog box, type the input range (A1:A26).
4. Check Column labels if the first row of the
data contains the name of the variable.
Choose Range:, Sheet, or Workbook
depending on where you wish the results to
appear. Under Tests click z-test.
5. Click Options and choose Mean ≠
Theoretical mean in the Alternative
hypothesis box. Type any number since
you’re not conducting a test here. Under
Variance for the z-test:, check User
2
defined and type the 56265 for σ . Specify
the value of α in percent (5) in the
Significance level (%) box.
6. Click Outputs and check Descriptive
statistics, Detailed results, and Confidence
interval.
Appendix 10.B – Stata Output and
Instructions
Example 10.1 – Estimating a Population
Mean: Standard Deviation Known
1. Import the data into one column. (Click
File/Import/Excel
spreadsheet(*xls,*xlsx)/Chapter10/Xm10-
01.) Check Import first row as variable
names.
2. Click Statistics, Summaries, tables and
tests, Classical tests of hypotheses, and z-
test (mean comparison test known
variance).
3. Select One-sample, select Demand in the
Variable name: box and type any number
 in the Hypothesized mean: box. Type 95
for the Confidence level, and 75 for the
Standard deviation.

Chapter 11: Introduction to Hypothesis Testing

11-1 Concepts of Hypothesis Testing
The critical concepts in hypothesis testing may
be summarized as follows:
1. There are two hypotheses
• The null hypothesis (H0 – pronounced “H
nought”)
• The alternative (or research) hypothesis
(H1).
2. The test begins with the assumption that H0
is true.
 3. The goal of the test is to determine whether
there is enough evidence to infer that H1 is
true.
4. There are two possible decisions
• Reject H0. Conclude that there is
enough evidence to support H1.
• Do not reject H0. Conclude that there is
not enough evidence to support H1.
11-1 A Nonstatistical Application of
Hypothesis Testing
In a criminal trial, a defendant stands accused of
a crime in front of a jury.
It can be said that the jury conducts a test of
hypotheses based on the evidence presented by
both the prosecution and the defense, that is:
• H0: The defendant is innocent.
• H1: The defendant is guilty.
The jury can only make two decisions (i.e.,
render two verdicts):
 • Reject H0. There is enough evidence to
conclude that the defendant is guilty: a
conviction.
• Do not reject H0. There is not enough
evidence to conclude that the defendant is
guilty: an acquittal.
Under no circumstances, the decision that H0 is
true, and the defendant is innocent, is ever
made.
11-1 Error Types
Two possible errors can be made in a test:
• A Type I error occurs when we reject a
true null hypothesis.
P(Type I error) = α, also called the significance
level.
• A Type II error occurs when we do not
reject a false null hypothesis.
P(Type II error) = β
The error probabilities α and β are inversely
related, and any attempt to reduce one will
 increase the other. The following table
summarizes the terminology and the concepts.

DECISION H0 IS TRUE H0 IS FALSE

Defendant is Defendant is guilty
innocent

REJECT H0 TYPE I ERROR CORRECT DECISION

Convict defendant P(TYPE I ERROR)
=α

DO NOT REJECT H0 CORRECT TYPE II ERROR

Acquit defendant DECISION P(TYPE II ERROR) = β
11-1 The Blackstone R
Our justice system is arranged by placing the
burden of proof on the prosecution.
The judges instruct the jury to find the
defendant guilty only if there is “evidence
beyond a reasonable doubt”, and in the
absence of enough evidence, the jury must
acquit.
It follows that the probability of acquitting guilty
people (Type II error) is greater than the
probability of convicting innocent people (Type
I error.)
Sir William Blackstone (1723–1780), an English
legal scholar, famously quipped: “Better for 10
guilty persons escape than that one innocent
suffers.” This is known as the Blackstone Ratio.
.(9:;< > <??@?) 5 !
The Blackstone Ratio: = =
.(9:;< >> <??@?) A !-

Benjamin Franklin (1706–1790) thought that a 1

to 10 ratio was not enough when he stated: “it is
better 100 guilty persons should escape than
that one innocent person should suffer, is a
maxim that has been long and generally
approved.”
11-1 A Statistical Application of Hypothesis
Testing
Recollect Example 10.1 in which the operations
manager at Doll’s Computers wanted to
estimate the mean demand during lead time.
Suppose that the manager wants to know
instead whether the mean demand is different
from 350.
Now the question becomes: “Is there enough
evidence to conclude that μ is not equal to
350?”
The hypotheses are set up to reflect a
manager’s decision problem wherein the null
hypothesis represents the status quo, whereas
the burden of proof is placed on the alternative
hypothesis.
Thus, we write:
H0: μ = 350
H1: μ ≠ 350
And what if the manager wants to know
whether the mean demand has increased from
350?
The new question is now: “Is there enough
evidence to conclude that μ is greater than
350?”
We can write the hypotheses as:
H0: μ = 350
H1: μ > 350
Finally, what if the manager wants to know
whether the mean demand has decreased from
350?
We can phrase the question as: “Is there
enough evidence to conclude that μ is less than
350?”
And write the hypotheses as:
H0: μ = 350
H`1: μ < 350
11-1 The Test Statistic
The test statistic is based on the best estimator
of the population parameter, and it is the
criterion used to make our decision about the
hypotheses.
In Example 10.1, the test statistic is the mean of
a random sample drawn from the population.
We use the test statistic to make one of the
following two decisions:
• If the test statistic’s value is inconsistent with
the hypothesized mean of 350, we reject H0
and conclude that there is sufficient
evidence that H1 is true.
• If the test statistic’s value is close to the
hypothesized mean of 350, we do not reject
H0 and conclude that there is no sufficient
evidence that H1 is true.
In the next section, we show how to use the test
statistic’s sampling distribution to define
“sufficient evidence.”
11-2 Estimating the Population Mean when
the Population Standard Deviation is Known
Example 11.1 – Department Store’s New
Billing System
DATA file: Xm11-01
After a thorough financial analysis, the manager
of a department store has determined that a
new billing system will be cost-effective only if
the mean monthly account is more than $170.
A random sample of 400 monthly accounts has
a sample mean of $178. The manager knows
that the accounts are approximately normally
distributed with a standard deviation of $65.
Question: “can the manager conclude that the
new system will be cost-effective?”
The null hypothesis can be expressed as:
H0: μ ≤ 170
However, as discussed in Section 11-1, we
actually test μ = 170, so we set the hypotheses
as:
H0: μ = 170 (system is not cost-effective: do not
install new system)
H1: μ > 170 (system is cost-effective: install new
system)
11-2a Defining the Rejection Region
Example 11.1 – Department Store’s New
Billing System
To conduct the test, we ask: “is a sample mean
of 178 sufficiently greater than 170 to conclude
that the population mean is greater than 170?”
To make a decision about the sample mean, we
set up the rejection region:
The rejection region is a range of values such
that if the test statistic falls into that range, we
decide to reject H0 in favor of H1.
We define the rejection region such that:

𝑥̅ > 𝑥̅B
Because a Type I error is defined as rejecting a
true null hypothesis, we can write that:

𝑃(Type I error) = 𝑃(𝑥̅ > 𝑥̅B | 𝐻- is true) = 𝛼

11-2a Calculating the Test Statistic from the

Rejection Region
Example 11.1 – Department Store’s New
Billing System
From Section 9.1, we know that 𝑋; is normally
distributed. As a result, we can standardize 𝑥̅ :

𝑥̅ − 𝜇 𝑥̅B − 𝜇 𝑥̅B − 𝜇
𝑃O > P = 𝑃 O𝑍 > P=𝛼
𝜎 ⁄√𝑛 𝜎 ⁄√ 𝑛 𝜎 ⁄√ 𝑛
From Section 8.2, we defined 𝑧5 such that:

𝑃(𝑍 > 𝑧5 ) = 𝛼
Because the two probabilities have the same
distribution, they also have the same limit:
𝑥̅B − 𝜇
= 𝑧5
𝜎⁄√𝑛
Solving for 𝑥̅B yields:

𝑥̅B = 𝜇 + 𝑧5 𝜎 ⁄√𝑛
We can choose a significance level, 𝛼 = 5%,
that is, 𝑧5 = 1.645.
If now we plug in 𝜇 = 170, 𝜎 = 65, 𝑛 = 400,
we obtain:

𝑥̅B = 170 + 1.645R65⁄√400S = 175.34

Therefore, the rejection region is:

𝑥̅ > 175.34
Because the test statistic (sample mean) is 178,
it is in the rejection region.
Thus, we reject the null hypothesis, and
conclude that there is sufficient evidence that
the mean monthly account is greater than $170.
11-2b Standardized Test Statistic
Example 11.1 – Department Store’s New
Billing System
One limitation of the previous method is that
the rejection region is set up in terms of 𝑥̅ . It is
easier to use the standardized test statistic
instead:
𝑥̅ − 𝜇
𝑧>
𝜎 ⁄√ 𝑛
Algebraically, the rejection region can be written
as: 𝑧 > 𝑧5
In our example we have: 𝑧 > 𝑧5 = 𝑧.-2 =
1.645
The standardized test statistic is:
𝑥̅ − 𝜇 178 − 170
𝑧> = = 2.46
𝜎⁄√𝑛 65/√400
Because 2.46 is greater than 1.645, we reject the
null
hypothesis and conclude that there is enough
evidence to
infer that the mean monthly account is greater
than $170.
For simplicity, we will refer to the standardized
test statistic as the test statistic for the rest of
the book.
11-2c p-Value
Example 11.1 – Department Store’s New
Billing System
The main limitation of the rejection region
method is that it produces a yes or no answer
to the question of whether to reject H0 but
offers no further information on the statistical
significance of the test.
One of the implications of using the rejection
region method is the impossibility of managing
considerations on factors such as the cost and
feasibility of restructuring the billing system
should a Type I error occur.
A better approach that provides a measure of
the amount of
statistical evidence supporting H1 uses the p-
value.
The p-value of a test is the probability of
observing a test
statistic at least as extreme as the one
computed given
that the null hypothesis is true.
In Example 11.1, the p-value is the probability
of observing 𝑥̅ ≥ 178 when 𝜇 = 170 is:
(C+D !E4+!E-
p-value = 𝑃(𝑋; > 178) = 𝑃 e'⁄ > f=
√* "2/√3--

𝑃 (𝑍 > 2.46) = 1 − 𝑃(𝑍 < 2.46) = 1 − .9931 =

.0069

11-2d Interpreting the p-Value

Example 11.1 – Department Store’s New
Billing System
In Section 11-2c, we concluded that the
probability of observing 𝑥̅ ≥ 178 when 𝜇 =
170 is .0069, which is a very unlikely event.
Consequently, we have reason to reject H0 and
support
H1. But that we cannot make a probability
statement about a parameter that is not a
random variable.
Incorrect interpretation: the p-value is the
probability that the H0 is true.
Correct interpretation: the p-value measures
the amount of statistical evidence that supports
H1.
The table and the graph to the right show
several values of 𝑥̅ , their z-statistics, and p-
values.
Notice how the farther 𝑥̅ is from the
hypothesized mean of 170, the smaller the p-
value is, and the more the statistical evidence
supports H1.

SAMPLE TEST
p-VALUE
v)
MEAN (𝒙 STATISTIC (z)

170 0 .5000

172 0.62 .2676

174 1.23 .1093

176 1.85 .0322

178 2.46 .0069

11-2c Describing the p-Value

Example 11.1 – Department Store’s New
Billing System
But how small does the p-value have to be to
infer that the alternative hypothesis is true? The
answer depends on several factors, including
the costs of making Type I and Type II errors.
In Example 11.1, a Type I error would occur if
the manager adopts the new billing system
when it is not cost-effective.
If such cost is high, then we want to reject H0
when the p-value is quite low by setting a very
low significance level, such as α = 1%.
In general, the following descriptive terms
apply:
• If p-value ≤ .01, there is overwhelming
evidence that H1 is true. The test is highly
significant.
• If .01 < p-value ≤ .05, there is strong
evidence that H1 is true. The test is
significant.
• If .05 < p-value ≤ .10, there is weak
evidence that H1 is true. The test is not
significant.
• If p-value > .10, there is little to no
evidence that H1 is true. The test is not
 significant.

11-2f The p-Value and Rejection Region

Methods
We can make similar decisions using either
method.
When using the rejection region method, the
following rules apply:
• If the test statistic is within the rejection
region, we reject H0.
• If the test statistic is outside the rejection
region, we do not reject H0.
The rejection region method can be used
manually using the table associated to the
sampling distribution.
When using the p-value method, the following
rules apply:
• If p-value ≤ α, we reject H0.
• If p-value > α, we do not reject H0.
The p-value method can be used manually
using a table only if the sampling distribution is
normal. For any other sampling distribution,
we can use Excel, as shown to the right.

*See notes for instructions on how to use the

Test Statistics Workbook for Example 11.1.
Instructions
1. Type or import the data into one column.
(Open Xm11-01.) In any empty cell,
calculate the sample mean
=AVERAGE(A1:A401)
 2. Open the Test Statistics Workbook and
click the z-Test_Mean tab. In Cell B3, type
or copy the value of the sample mean. In
cells B4–B7, type the value of s (65), the
value of n (400), the value of m under the
null hypothesis (170), and the value of a
(.05), respectively.
The spreadsheet reports the value of the test
statistic, z = 2.46. The p-value* of the test
is .0069. Excel reports this probability as
P(Z <= z) one{tail)
*Excel provides two probabilities in its printout.
The way in which we determine the p-value of
the test from the printout is somewhat more
complicated. Interested students are advised to
read the online appendix Converting Excel’s
Probabilities to p-Values.
11-2i One- and Two-Tail Tests: Identify
Example 11.2 – The Effects of Kiosks on Sales
in Fast-Food Restaurants
Example 11.1 was called a one-tail test because
the rejection region is located only in one tail of
the sampling distribution. We now present an
example of a two-tail test.
A restaurant franchise, concerned about
possible changes in the sales due to the
installation of self-serve kiosks, takes a random
sample of 100 customers who use the new
machines at McDonald’s.
It is known that before the machine’s
installation, the average customer at
McDonald’s spent $6.03 with a standard
deviation of $0.91.
Do the data provide enough evidence at the 5%
significance level of a change in the size of the
transaction for an individual customer?
DATA file: Xm 11-02
From the data, we calculate:
∑ 𝑥8 591
𝑥̅ = = = 5.91
𝑛 100
From the problem statement, we also know
that:
μ = 6.03 and σ = 0.91.
Question: Has the mean expenditure changed
from $6.03?
We set the hypotheses as:
H0: μ = 6.03
H1: μ ≠ 6.03
11-2i One- and Two-Tail Tests: Compute and
Interpret
Example 11.2 – The Effects of Kiosks on Sales
in Fast-Food Restaurants
Because we can reject the null hypothesis when
the test statistic is large or when it is small, and
the total area in the rejection region must be α,
we divide this probability by 2:
𝑧 < 𝑧5/# or 𝑧 > 𝑧5/#
Because α = .05, then α / 2 = .025, and 𝑧..-#2 =
1.96.
Thus, we can write the rejection region as:
𝑧 < −1.96 or 𝑧 > 1.96
The value of the test statistic is:
𝑥̅ − 𝜇 5.91 − 6.03
𝑧= = = −1.32
𝜎 ⁄√ 𝑛 0.91/√100
Because −1.32 is not in the rejection region, we
cannot reject H0.
The p-value corresponds to the area of both
tails:
p-value = 𝑃(𝑧 < −1.32) + 𝑃(𝑧 > 1.32)
= .0934 + .0934 = .1868 > 𝛼
There is not enough evidence to conclude that
the self-serve kiosks result in a change from the
amounts incurred with a cashier.

11-2j When Do We Conduct One- and Two-

Tail Tests?
We conduct a two-tail test whenever we want
to know whether there is enough evidence to
infer that the mean is not equal to the value
stated in the null hypothesis:
H0: μ = μ0
H1: μ ≠ μ0
There are two one-tail tests.
We conduct a right one-tail test whenever we
want to know whether there is enough evidence
to infer that the mean is greater than the
quantity specified by the null hypothesis:
H0: μ = μ0
H1: μ > μ0
We conduct a left one-tail test whenever we
want to know whether there is enough evidence
to infer that the mean is less than the quantity
specified by the null hypothesis:
H0: μ = μ0
H1: μ < μ0
11-2k Testing Hypotheses and Confidence
Interval Estimators
Example 11.2 – The Effects of Kiosks on Sales
in Fast-Food Restaurants
The test statistic and the confidence interval
estimator are both derived from the sampling
distribution. It follows that we can use the
confidence interval estimator to test
hypotheses.
The 95% confidence interval estimate of the
population mean is:

𝜎 0.91
𝑥̅ ± 𝑧5⁄# = 5.91 ± 1.96 = 5.91 ± 0.18
√𝑛 √100
LCL = 5.73 and UCL = 6.09
We estimate that μ lies between $5.73 and
$6.09. Because the interval includes $6.03, we
cannot conclude that there is sufficient evidence
to infer that the mean has changed.
This process is equivalent to the rejection region
approach.
11-3 Calculating the Probability of a Type II
Error
Example 11.1 – Department Store’s New
Billing System
The rejection region (α = 5%) was found to be:

𝑥̅ > 175.34
A Type II error occurs when a false H0 is not
rejected. We defined its probability as:

𝛽 = 𝑃(𝑋; < 175.34 | 𝐻- false)

To compute β we need a value for μ. Suppose
that the manager knows that when μ ≥ 180,
the new system is certain to be cost-effective.
Thus:

𝛽 = 𝑃(𝑋; < 175.34 | 𝜇 = 180)

Because 𝑋; is normally distributed, we can
standardize it:

𝑋; − 𝜇 175.34 − 180
𝛽 = 𝑃O < P
𝜎 ⁄√ 𝑛 65/√400
= 𝑃(𝑍 < −1.43) = .0764
This tells us that when the mean account is
actually $180, the probability of incorrectly not
rejecting the null hypothesis is .0764.
Notice that to calculate the probability of a Type
II error, we had to specify a value for μ other
than the one shown in the null hypothesis.

11-3a Effect on β of Changing α

Example 11.1 – Department Store’s New
Billing System
Suppose we use a significance level of 1%
instead. The rejection region, in terms of z,
becomes:

𝑧 > 𝑧.-! = 2.33

Which can be expressed as:
 𝑥̅ − 170
= 2.33
65⁄√400
Solving for 𝑥̅ we obtain:

𝑥̅ = 170 + 2.33R65⁄√400S = 177.57

The probability of a Type II error when μ = 180
is:

𝑋; − 𝜇177.57 − 180
𝛽 = 𝑃O < P
𝜎 ⁄√ 𝑛 65/√400
= 𝑃(𝑍 < −0.75) = . 2266
By decreasing the significance level from 5% to
1%, we have shifted the critical value of the
rejection region to the right and thus enlarged
the area where the null hypothesis is not
rejected.
As a result, the probability of a Type II error
increased from .0764 to .2266.
11-3a Effect on β of Changing the Sample
Size
Example 11.1 – Department Store’s New
Billing System
The manager found β to be .0764 when
choosing a sample size of 400, an α = 5% on
which to base the decision, with the actual
mean equal to 180.
If the cost of a Type II error is high, there are
two approaches to reduce it:
• Increase the significance level, α. (This will
increase the probability of a Type I error)
• Increase the sample size, n.
As an example, suppose the managers selects a
sample size, n = 1,000.
The rejection region is:
$̅ +!E-
𝑧 > 𝑧.-2 = 1.645 or > 1.645
"2/√!,---

Solving for 𝑥̅ yields:

𝑥̅ > 170 + 1.645R65⁄31,000S = 173.38

Thus, the probability of a Type I error is:

𝑋; − 𝜇 173.38 − 180
𝛽 = 𝑃O < P
𝜎 ⁄√ 𝑛 65/√1,000
= 𝑃(𝑍 < −3.22) ≈ 0

11-3d Power of a Test

The sampling distribution of the mean is
narrower because the standard error of the
mean becomes smaller as n increases. Narrower
distributions convey more information. The
increased information is reflected in a smaller
probability of a Type II error.
We define the power of a test as the
probability of rejecting H0 when false: 1 - β
n = 400

n = 1,000
11-3e Operating Characteristic Curve
The operating characteristic (OC) curve plots
the values of β versus the values of μ. It can be
used to express how well a test performs for
different significance levels and sample sizes.
Because of the time-consuming nature of these
calculations, the use of a computer is necessary.
The figure to the right shows the OC curve for
Example 11.1 with n = 100, 400, 1,000,
and 2,000.
The chart provides insight on how:
• To select the sample size, n.
• The alternative value of μ increases
the value of β decreases.
• When μ = 170, the hypothesized mean, β =
1 – α.
*See notes for Excel instructions.
Instructions:
With σ = 65 and n = 400, the standard error of
the mean is:
' "2
𝜎$̅ = = = 3.25
√* √3--

To calculate the probability of a Type II error in

Example 11.1, we open Excel and in any empty
cell type
=NORM.DIST(175.35,[μ],3.25,True)
For example, to compute β when μ = 180, we
type
 =NORM.DIST(175.35,180,3.25,True)
11-4b Problem Objectives
Guide to Statistical Inference Techniques
The table specifies the chapter or section where
the appropriate statistical technique is
presented for each combination of data type
and problem objective.

DATA TYPE

PROBLEM
NOMINAL ORDINAL INTERVAL
OBJECTIVE

Describe a Sections 12.3, Not covered Sections 12.1,

population 15.1 12.2

Compare Sections 13.5, Sections Sections 11.1,

two 15.2 19.1, 19.2 13.3, 13.4,
populations 19.1, 19.2

Compare Section 15.2 Section 19.3 Chapter 14,

two or Section 19.3
more
populations
Analyze the Section 15.2 Section 19.4 Chapter 16
relationship
between
two
variables

Analyze the Not covered Not covered Chapters 17,

relationship 18
between
two or
more
variables

Appendix 11.A1 – XLSTAT Output and

Instructions
Example 11.1 – Z-Test of a Mean
Instructions:
1. Type or import the data into one column.
(Open Xm11-01.)
2. Click XLSTAT, Parametric tests, and One-
sample t-test and z-test.
3. Check One sample under Data format: In
the Data dialog box, type the input range
(A1:A401). Check Column labels if the first
row of the data contains the name of the
variable. Choose Range:, Sheet, or
Workbook depending on where you wish
the results to appear. Under Tests click z-
test.
4. Click Options and choose Mean >
Theoretical mean in the Alternative
hypothesis box. Type the Theoretical
mean (170). Under Variance for the z-
test:, check User defined and type the
2
4225 for σ . Specify the value of α in
percent (5) in the Significance level (%) box.
5. Click Outputs, check Descriptive statistics,
and Detailed results.
Appendix 11.A3 – XLSTAT Output and
Instructions
Example 11.2 – Z-Test of a Mean
Instructions:
1. Type or import the data into one column.
(Open Xm11-02.)
2. Click XLSTAT, Parametric tests, and One-
sample t-test and z-test.
3. Check One sample under Data format: In
the Data dialog box, type the input range
(A1:A101). Check Column labels if the first
row of the data contains the name of the
 variable. Choose Range:, Sheet, or
Workbook depending on where you wish
the results to appear. Under Tests click z-
test.
4. Click Options and choose Mean ≠
Theoretical mean in the Alternative
hypothesis box. Type the Theoretical
mean (6.03). Under Variance for the z-
test:, check User defined and type
2
the .8281 for σ . Specify the value of α in
percent (5) in the Significance level (%) box.
5. Click Outputs, check Descriptive statistics,
and Detailed results.
Appendix 11.B1 – Stata Output and
Instructions
Example 11.1 – Testing a Population Mean:
Standard Deviation Known
Instructions:
1. Import the data into one column. (Click
File/Import/Excel
spreadsheet(*xls,*xlsx)/Chapter11/Xm11-
01.) Check Import first row as variable
names.
2. Click Statistics, Summaries, tables and
tests, Classical tests of hypotheses, and z-
test (mean comparison test known
variance).
3. Select One-sample, select Accounts in the
Variable name: box and type 170 in the
Hypothesized mean: box. Type 65 for the
Standard deviation.
Appendix 11.B3 – Stata Output and
Instructions
Example 11.2 – Testing a Population Mean:
Standard Deviation Known
Instructions:
1. Import the data into one column. (Click
File/Import/Excel
spreadsheet(*xls,*xlsx)/Chapter11/Xm11-
02.) Check Import first row as variable
names.
2. Click Statistics, Summaries, tables and
tests, Classical tests of hypotheses, and z-
test (mean comparison test known
variance).
3. Select One-sample, select Accounts in the
 Variable name: box and type 6.03 in the
Hypothesized mean: box. Type 0.91 for the
Standard deviation.

Chapter 12: Inference About a Population

12-1 Inference About a Population Mean:
Standard Deviation Unknow
In 1908, mathematician William S. Gosset found
that when the sample standard deviation is used
in place of the unknown population standard
deviation, the resulting sampling distribution is
no longer normal.
Test Statistic for μ when σ Is Unknown
When the population standard deviation is
unknown and the population is normal, the test
statistic for testing hypotheses about μ is
 𝑥̅ − 𝜇
𝑡=
𝑠/√𝑛
which is Student t-distributed with n = ν − 1
degrees of freedom.
Confidence Interval Estimator of μ when σ Is
Unknown
𝑠
𝑥̅ ± 𝑡5# with 𝑛 = 𝜈 − 1
√𝑛
Example 12.1 – Newspaper Recycling Plant:
Identify
A financial analyst for a recycling company has
recently computed that the firm would make a
profit if the mean weekly newspaper collection
per household exceeded 2.0 pounds.
In a study to determine the feasibility of a
recycling plant, a random sample of 148
households was drawn from a large community,
and the weekly weight of newspapers discarded
for recycling for each household recorded.
Do these data provide sufficient evidence to
allow the analyst to conclude that a recycling
plant would be profitable?
DATA file: Xm12-01
Identify
Problem objective: describe the population of
the amounts of newspaper discarded per
household.
The analyst is testing for a population mean in
excess of 2.0 lbs. Thus, this is a right one-tail
test.
H0: μ = 2.0
H1: μ > 2.0
Because σ is unknown, the test statistic is
𝑥̅ − 𝜇
𝑡= with 𝜈 = 𝑛 − 1
𝑠/√𝑛
The analyst believes that the cost of a Type I
error (falsely rejecting H0) is quite high and sets
α = 1%.
Example 12.1 – Newspaper Recycling Plant:
Compute and Interpret
Compute
To calculate the test statistic, we need 𝑥̅ and s
∑ 𝑥8 322.7
𝑥̅ = = = 2.18
𝑛 148
(∑ 𝑥8 )# 322.7#
∑ 𝑥8# − 845.1 −
𝑠# = 𝑛 = 148 = .962
𝑛−1 148 − 1

𝑠 = 3𝑠 # = √0.962 = 0.981
The test statistic is:
𝑥̅ − 𝜇 2.18 − 2.0
𝑡= = = 2.23
𝑠/√𝑛 . 981/√148
The rejection region is

𝑡 > 𝑡5,*+! = 𝑡.-!,!3E ≈ 𝑡.-!,!2- = 2.351

Excel calculations (*see notes) yield a p-value
of .0134.
We conclude that we cannot reject H0 in favor of
H1, as shown using the rejection region and the
p-value, both.
Interpret
Because we did not reject H0, we conclude that
there is not enough evidence to infer that the
mean weight of discarded newspapers is greater
than 2.0.
Note: there is evidence, but because the analyst
insisted on such a low probability of a Type I
error, we cannot conclude that the plant will be
profitable.
Instructions:
1. Type or import the data into one column.
(Open Xm12-01.) In any empty cell, calculate
the sample mean
=AVERAGE(A1:A149)
and the sample standard deviation
=STDEV.S(A1:A149).
 2. Open the Test Statistics Workbook and
click the t-Test_Mean tab.
Type or copy the sample mean (2.18) and the
sample standard deviation (0.981).
Type the sample size (148), the value of μ under
the null hypothesis (2.0) and the value of α (.01).
Example 12.2 – Tax Collection: Identify
In 2019, almost 200 million tax returns were
filed in the United States. The Internal Revenue
Service (IRS) audited 771,095 of them to
determine if they were properly filled.
To determine how well the IRS auditors are
performing, a random sample of 198 returns
was drawn from the population of 771,095
audited files, and the additional tax recorded.
Estimate with 95% confidence the mean
additional income tax collected from the
population of audited files.
DATA file: Xm12-02
Identify
Problem objective: describe the population of
additional income tax.
The question asks us to estimate the population
mean μ. The confidence interval estimator is:
𝑠
𝑥̅ ± 𝑡5⁄#, *+!
√𝑛
Because we want a 95% confidence interval
estimate

1 − 𝛼 = 95% 𝛼 ⁄2 = . 05⁄2 = .025

Thus:
𝑡5/#,*+! = 𝑡.-#2,!HE ≈ 𝑡.-#2,#-- = 1.972
Example 12.2 – Tax Collection: Compute and
Interpret
Compute
From the data, we determine 𝑥̅ and s
∑ 𝑥8 4.438,221
𝑥̅ = = = 22,415
𝑛 198
(∑ 𝑥8 )#
∑ 𝑥8# −
𝑠# = 𝑛 = 133,262,368
𝑛−1
 𝑠 = 3𝑠 # = 3133,262,368 = 11,544
Thus, the 95% confidence interval estimate of μ
is

𝑠 11,544
𝑥̅ ± 𝑡5⁄#, *+! = 22,415 ± 1.972
√𝑛 √198
or
LCL = 20,797
UCL = 24,033
Interpret
We estimate that the mean additional tax
collected lies between $20,797 and $24,033.
We can use this estimate to help decide
whether the IRS is auditing the individuals who
should be audited.
*See notes for Excel instructions.
Instructions:
1. Type or import the data into one column.
(Open Xm12-02.) In any empty cell, calculate
the sample mean
 =AVERAGE(A1:A199)
and the sample standard deviation
=STDEV.S(A1:A199).
2. Open the Test Statistics Workbook and
click the t-Estimate_Mean tab.
Type or copy the sample mean (22,415) and the
sample standard deviation (11,544).
Type the sample size (198), and the confidence
level (95%).
12-1a Checking the required Conditions
Histogram for Example 12.1

Histogram for Example 12.2

Statisticians have shown that the Student t-
distribution is robust, and even if the
population is nonnormal, the results of the t-
test and confidence interval estimate are still
valid provided that the population is not
extremely nonnormal.
To check the normality requirement, we draw a
histogram and determine whether the shape is
far from bell shaped. Both histograms for
examples 12.1 and 12.2 suggest that the
variables are not extremely nonnormal, and in
fact, may even be normal.
12-1cde Developing an understanding of
Statistical Concepts
Degrees of freedom
The reason why we divide the sample variance
by n – 1 is because to compute it, we need to
first know the sample mean 𝑥̅ .
Imagine drawing from a population a multitude
of samples having the same sample mean. Then
n – 1 of the data points in the sample can take
any value in the population, but the n-th data
point is “locked” to the value needed to yield
the given sample mean.
Standard error
When the population standard deviation σ is
unknown, we estimate the standard error with
I
.
√*

Student t-distribution
The Student t-distribution is more widely spread
out than the standard normal because the t-
statistic has two variables: the sample mean 𝑥̅
and the sample standard deviation s, both of
which vary from sample to sample; whereas the
z-statistic has only the sample mean.
Because of the greater uncertainty, the t-
statistic will display greater variability.
12-2 Inference About a Population Variance
To draw inferences about a population’s
variability, we need to investigate the
2 2
population variance σ . The sample variance s is
2
the estimator of σ and it has the characteristics
of being an unbiased and consistent estimator
2
of σ .
2
Test Statistic for σ

( ) #
𝑛 − 1 𝑠
𝛸# =
𝜎#
which is chi-squared distributed with n = ν − 1
degrees of freedom when the population
random variable is normally distributed with
2
variance equal to σ .
2
Confidence Interval Estimator of σ
(𝑛 − 1)𝑠 # (𝑛 − 1)𝑠 #
𝐿𝐶𝐿 = # 𝑈𝐶𝐿 = #
𝛸5/# 𝛸!+5/#

Example 12.3 – Consistency of a Container-

Filling Machine, Part 1: Identify
The president of a company that developed a
new type of container-filling machine boasts
that it can fill 1-liter (1,000 cc) containers so
consistently that the variance of the fills is less
than 1 cc.
To examine the veracity of the claim, a random
sample of 25 1-liter fills is taken, and the results
in cc recorded.
Do the data support the president’s claim at the
5% significance level?
DATA file: Xm12-03
Identify
Problem objective: describe the population of
1-liter fills from this machine.
This is a left one-tail test, because we want to
2
determine whether the population variance σ is
less than 1 cc.
H0: σ2 = 1
H1: σ2 < 1
The test statistic is

( ) #
𝑛 − 1 𝑠
𝛸# = with 𝜈 = 𝑛 − 1
𝜎#
Example 12.3 – Consistency of a Container-
Filling Machine, Part 1: Compute and
Interpret
Compute
From the data, we find

b 𝑥8 = 24,992 and b 𝑥8# = 24,984,017,76

(∑ 𝑥8 )#
∑ 𝑥8# −
𝑠# = 𝑛 = 0.6333
𝑛−1
The test statistic is:

( ) # (25 − 1)0.6333
#
𝑛 − 1 𝑠
𝛸 = #
= = 15.20
𝜎 1
The rejection region is

𝛸 # < 𝛸 #!+5,*+! = 𝛸 # .H2,#3 = 13.85

Excel calculations (*see notes) yield a p-value
of .0852.
We conclude that we cannot reject H0 in favor of
H1, as shown using the rejection region and the
p-value, both.

Interpret
Because we did not reject H0, we conclude that
there is not enough evidence to infer that the
population variance is less than 1 cc.
Instructions:
1. Type or import the data into one column.
(Open Xm12-03.) In any empty cell, calculate
the sample variance
=VAR.S(A1:A26)
2. Open the Test Statistics Workbook and
click the Chi-Squared Test_Variance tab.
Type or copy the sample variance (0.6333).
2
Type the sample size (25), the value of σ under
the null hypothesis (1) and the value of α (.05).
Example 12.4 – Consistency of a Container-
Filling Machine, Part 2
Estimate with 99% confidence the variance of fills
in Example 12.3.
Compute
Recollect from Example 12.3 we found that:
(𝑛 − 1)𝑠 # = (25 − 1)0.6333 = 15.20
From Table 5 in Appendix B, we find
𝛸 # 5/#,*+! = 𝛸 # .--2,#3 = 45.6
𝛸 #!+5/#,*+! = 𝛸 # .HH2,#3 = 9.89
*See notes for Excel instructions.
(𝑛 − 1) 𝑠 # 15.20
𝐿𝐶𝐿 = # = = 0.3333
𝛸5/# 45.6

(𝑛 − 1)𝑠 # 15.20
𝑈𝐶𝐿 = # = = 1.537
𝛸!+5/# 9.89

We are 99% confident that the variance of fills

lies between .3333 and 1.537 cc.
Interpret
Part of the confidence interval is above 1, which
tells us that the variance may be larger than 1.
We may also be able to use the estimate to
predict the percentage of overfilled and
underfilled bottles.
Instructions:
1. Type or import the data into one column.
(Open Xm12-03.) In any empty cell, calculate
the sample variance
=VAR.S(A1:A26)
2. Open the Test Statistics Workbook and
click the Chi-Squared Estimate_Variance
tab.
Type or copy the sample variance (0.6333).
Type the sample size (25) and the confidence
level (.99).
12-2c Checking the required Conditions
 2
The chi-squared test and estimator of σ
theoretically require that the sample population
be normal.
In practice, the technique is valid so long as the
population is not extremely nonnormal. We can
gauge the extent of nonnormality by drawing
the histogram as shown in the figure to the
right.
As you can see, the fills appear to be somewhat
asymmetric. However, the variable does not
appear to be unusually nonnormal.
We conclude that the normality requirement is
not seriously violated.
Histogram for Examples 12.3 and 12.4
12-3 Inference About a Population
Proportion
To draw inferences about the population’s
proportion p of a nominal random variable, we
need to compute the sample proportion 𝑝̂ ,
$
defined as 𝑝̂ =
*

Where x is the number of successes in the

sample and n the sample size.
Test Statistic for p
𝑝̂ − 𝑝
𝑧=
3𝑝(1 − 𝑝)/𝑛
which is approximately normal when np and n(1
– p) are both greater than 5.
Confidence Interval Estimator of p

𝑝̂ ± 𝑧5/# 3𝑝̂ (1 − 𝑝̂ )/𝑛

which is valid provided that 𝑛𝑝̂ and n(1 − 𝑝̂ )
are both greater than 5.
Example 12.5 – Election Day Exit Poll: Identify
In the exit poll from the state of Florida during
the 2000 presidential elections, the pollsters
recorded the votes of the two top candidates,
Democrat Al Gore (code = 1) and Republican
GW Bush (code = 2). The polls close at 8:00 p.m.
Can the networks conclude from these data that
the Republican candidate will win the state?
Should the network announce at 8:01 p.m. that
the Republican candidate will win?
Use a significance level, α = .05.
DATA file: Xm12-05
Identify
Problem objective: describe the population of
votes for the 2000 presidential elections in
Florida.
This is a right one-tail test, because we want to
determine whether the Republican candidate
has enough votes to win Florida.
H0: p = .5
H1: p > .5
The test statistic is
𝑝̂ − 𝑝
𝑧=
3𝑝(1 − 𝑝)/𝑛
Example 12.5 – Election Day Exit Poll:
Compute and Interpret
Compute
From the data file, we count the number of
“successes” as the number of votes for the
Republican candidate and find x = 407.
Hence, the sample proportion is
𝑥 407
𝑝̂ = = = .532
𝑛 765
The test statistic is:
𝑝̂ − 𝑝 . 532 − .5
𝑧= = = 1.77
3𝑝(1 − 𝑝)/𝑛 3. 5(1 − .5)/765
The rejection region is

𝑧 > 𝑧5 = 𝑧.-2 = 1.645

Because the test statistic is (approximately)
normally distributed, we can determine the p-
value as
p-value = 𝑃(𝑍 > 1.77) = 1 − 𝑃(𝑍 < 1.77)
= 1 − .9616 = .0384
We conclude that we can reject H0 in favor of
H1, as shown using both the rejection region
and the p-value.
Interpret
We can declare the Republican candidate the
winner of the Florida elections, at a 5%
significance level.
The issue here is the high cost of a Type I error,
such as to declare a winner without sufficient
evidence, which is exactly what happened when
all the networks recanted their earlier 8:01 p.m.
announcement at about 10 p.m., and declared
the Florida elections too close to call.
Instructions:
1. Type or import the data into one column.
(Open Xm12-03.) In any empty cell, calculate
the sample variance
=VAR.S(A1:A26)
 2. Open the Test Statistics Workbook and
click the Chi-Squared Test_Variance tab.
Type or copy the sample variance (0.6333).
2
Type the sample size (25), the value of σ under
the null hypothesis (1) and the value of α (.05).
Chapter Opener – The Number of
Unemployed: Identify
The United States Bureau of Labor Statistics
(BLS) defines the unemployment rate as the
proportion of unemployed persons currently in
the labor force.
We can use the variable work status (WRKSTAT)
included in the General Social Survey to
estimate with 95% confidence the number of
American adults who are either temporarily not
working or unemployed/laid off.
Identify
Problem objective: describe the population of
work status of American adults.
Because the data are nominal, we use the
confidence interval estimator of the population.

𝑝̂ ± 𝑧5/# 3𝑝̂ (1 − 𝑝̂ )/𝑛

Assume that there are 255,200,373 adults 18
and over in the United States.
The responses of the variable WRKSTAT are:
1. Working fulltime
2. Working part time
3. Temporarily not working
4. Unemployed, laid off
5. Retired
6. School
7. Keeping house
8. Other
We can define as the number of successes x the
count of 3s and 4s in the WRKSTAT column.
The counts of 3s and 4s are 53 and 84,
respectively.

𝑥 = 53 + 84 = 137
Chapter Opener – The Number of
Unemployed: Calculate and Interpret
Calculate
!&E
Thus, we have: 𝑝̂ = = .0584
#,&3"

Because we have at least 5 successes and 5

failures, the confidence interval estimator is
assumed normal.
The confidence level is 95%. Therefore, 𝛼 = .05,
and
𝑧5/# = 𝑧.-2/# = 𝑧.-#2 = 1.96
The 95% confidence interval estimate of p is

𝑝̂ (1 − 𝑝̂ )
𝑝̂ ± 𝑧5/# =
𝑛

. 0584(1 − .0584)
= .0584 ± =
2,346
= .0584 ± .0095
Or, in interval form:
LCL = .0489 and UCL = .0679
*See notes for Excel instructions.
Interpret
We estimate that the proportion of unemployed
American adults lies between 4.89% and 6.79%.
To determine the number of unemployed
people, multiply the lower and upper limits by
the U.S. adult population size 255,200,373.
Thus,

𝐿𝐶𝐿 = 255,200,373(. 0489) = 12,479,298

𝑈𝐶𝐿 = 255,200,373(. 0679) = 17,328,105
Instructions:
1. Type or import the data into one column.
(Open GSS2018.) (We copied column X into
another spreadsheet.) In any empty cell,
calculate the number of “successes”
(=COUNTIF(A1:A2349,3) and
=COUNTIF(A1:A2349,4)) Divide this number
(53 + 84) by the sample size (2,346) to
obtain the sample proportion.
2. Open the estimators Workbook and click
 the z-estimate_proportion tab. Type or
copy the sample proportion. Type the value
of the sample size and the value of α.
12-3f Selecting the Sample Size to estimate
the Proportion
As shown in Section 10-3, the sample size
depends on the confidence level and the bound
on the error of estimation B that we are willing
to tolerate. Solving for n, the required sample
size can be estimated as
𝑧5/# #
𝑛=e f 𝑝̂ (1 − 𝑝̂ )
𝐵
The confidence level and bound error of
estimation are usually provided or known, but if
the sample has yet to be taken, how do we
estimate 𝑝̂ ? There are two methods we can use.
Consider an example in which B = .03, and 1 – α
= .95, so that 𝑧5/# = 𝑧.-#2 = 1.96.
Method 1
It turns out that the product 𝑝̂ (1 − 𝑝̂ ) has its
maximum value at 𝑝̂ = .5. Thus
 𝑧5/# # 1.96 #
𝑛=e f 𝑝̂ (1 − 𝑝̂ ) = H I (. 5)(. 5)
𝐵 . 03
= 1,068
This method would work well in Example 12.5.
Method 2
If we have some idea of what 𝑝̂ could be, we
can use that. Suppose we may already have the
results for a preliminary smaller sample, with
𝑝̂ = .2

𝑧5/# # 1.96 #
𝑛=e f 𝑝̂ (1 − 𝑝̂ ) = H I (. 2)(. 8) = 683
𝐵 . 03
12-3g Wilson Estimator (Optional)
Consider the case in which the confidence
interval estimator of a proportion is a relatively
rare event. If a sample produced 0 successes,
then 𝑝̂ = 0, and the confidence interval would
be

𝑝̂ ± 𝑧5/# 3𝑝̂ (1 − 𝑝̂ )/𝑛 = 0 ± 𝑧5/# 30(1 − 0)/𝑛

=0±0
To remedy such as unacceptable result, in 1927,
Edwin Wilson proposed replacing 𝑝̂ with a
quantity he called 𝑝‚ (pronounced p “tilde”), the
Wilson estimator, which he defined as follows
𝑥+2
𝑝‚ =
𝑛+4
Thus, the confidence interval estimator of p
using the Wilson estimator becomes

𝑝‚(1 − 𝑝‚)
𝑝‚ ± 𝑧5/# =
𝑛+4

Appendix 12.A1 – XLSTAT Output and

Instruction
Example 12.1 – t-Test of a Mean
In the instructions for Appendix 11.A1, replace:
DATA file: Xm12-01
Data Input range: (A1:A149)
Test: Student’s t-test
Alt. hypothesis: Mean 1 > Theoretical mean (2)
Significance level %: (1)
Example 12.2 – t-Estimate of a Mean
In the instructions for Appendix 10.A, replace:
DATA file: Xm12-02
Data Input range: (A1:A199)
Tests: Student’s t-test

Appendix 12.A2 – XLSTAT Output and

Instructions
Example 12.3 – Test of a Population Variance
1. Import the data into one column. (Open
 Xm12-03)
2. Click XLSTAT, Parametric tests, and One-
sample
variance test.
3. In the Data dialog box type the input range
(A1:A26).
Check Column labels if the first row of the data
contains the name of the variable. Choose
Range:,
Sheet, or Workbook depending on where you
wish
the results to appear.
4. Click Options and choose Variance 1 <
Theoretical variance in the Alternative
hypothesis box. Type the Theoretical
variance (1). Specify the value of α in % (1)
in the Significance level (%) box.
5. Click Outputs and check Descriptive
statistics, and Detailed results.
See notes for Excel instructions on Example 12.4
– Estimate of a Population Variance.
Instructions Example 12-4 - Estimate of a
Population Variance
Follow the instructions above.
Specify Variance 1 ≠ Theoretical variance in
the Alternative hypothesis box.
Enter the Significance level (%) (1).
Click Outputs and check Descriptive statistics,
Detailed results, and Confidence interval.
Appendix 12.A3 – XLSTAT Output and
Instructions
Example 12.5 – Test of a Population
Proportion
 1. Import the data into one column. (Open
Xm12-05)
2. Click XLSTAT, Parametric tests, and Test
for One Proportion.
3. Type the Frequency (407) and the Sample
size (765). Type the Test proportion (.5).
Select Frequency under Data format:
Check the z-test.
4. Click the Options tab and select
Proportion – Test proportion > D in the
Alternative hypothesis box. Type the
Hypothesized difference (D) (0).

Appendix 12.B1 – Stata Output and

Instructions
Example 12.1 – t-Test of a Mean

In the instructions for Appendix 11.B1, replace:

DATA file: Xm12-01
Test: t-test (mean-comparison test)
Variable name: Newspaper
Hypothesized mean: (2)
Example 12.2 – t-Estimate of a Mean

In the instructions for Appendix 10.B, replace:

DATA file: Xm12-02
Test: t-test (mean-comparison test)
Variable name: AdditionalIncomeTax
Appendix 12.B2 – Stata Output and
Instruction
Example 12.3 – Test and Estimate of a
Population Variance
1. Import the data into one column. (Click
File/Import/Excel
spreadsheet(*xls,*xlsx)/Chapter12/Xm12-
03.) Check Import first row as variable
names.
2. Click Statistics, Summaries, tables and
tests, Classical tests of hypotheses, and
Variance-Comparison Test.
3. Select One-sample, select Fills in the
Variable name: box and type 1 in the
Hypothesized mean: box.
Appendix 12.B3 – Stata Output and
Instructions
Example 12.5 – Test of a Proportion
1. Import the data into one column. (Click
File/Import/Excel
spreadsheet(*xls,*xlsx)/Chapter12/Xm12-
05.) Check Import first row as variable
names.
2. Click Data, Create or change data, and
Create new variables. In the Variable
name box, write Rvotes. In the Specify a
value or an expression box type Votes==2.
Click OK. This instruction tells Stata to
convert 2s into 1s and 1s into 0s.
3. Click Statistics, Summaries, tables and
 tests, Classical tests of hypotheses, and
Proportion Test.
4. Select One-sample, select Rvotes in the
Variable name: box and type .5 in the
Hypothesized mean: box.