Engineering Statistics 31 March, 2022

Lecture No. 10
Resourse Person: Dr. Absar Ul Haq Department: Mechanical Engineering (Narowal Campus).

10.1 Sampling Distributions

The field of statistical inference is basically concerned with generalizations and predictions. For example, we
might claim, based on the opinions of several people interviewed on the street, that in a forthcoming election
60% of the eligible voters in the city of Detroit favor a certain candidate. In this case, we are dealing with a
random sample of opinions from a very large finite population. As a second illustration we might state that
the average cost to build a residence in Charleston, South Carolina, is between $330,000 and $335,000, based
on the estimates of 3 contractors selected at random from the 30 now building in this city. The population
being sampled here is again finite but very small. Finally, let us consider a soft-drink machine designed
to dispense, on average, 240 milliliters per drink. A company official who computes the mean of 40 drinks
obtains x̄ = 236 milliliters and, on the basis of this value, decides that the machine is still dispensing drinks
with an average content of µ = 240 milliliters. The 40 drinks represent a sample from the infinite population
of possible drinks that will be dispensed by this machine.

Inference about the Population from Sample Information

In each of the examples above, we computed a statistic from a sample selected from the population, and
from this statistic we made various statements concerning the values of population parameters that may or
may not be true. The company official made the decision that the soft-drink machine dispenses drinks with
an average content of 240 milliliters, even though the sample mean was 236 milliliters, because he knows
from sampling theory that, if µ = 240 milliliters, such a sample value could easily occur. In fact, if he ran
similar tests, say every hour, he would expect the values of the statistic x̄ to fluctuate above and below
µ = 240 milliliters. Only when the value of x̄ is substantially different from 240 milliliters will the company
official initiate action to adjust the machine. Since a statistic is a random variable that depends only on the
observed sample, it must have a probability distribution.

Definition 10.1 The probability distribution of a statistic is called a sampling distribution.

The sampling distribution of a statistic depends on the distribution of the population, the size of the samples,
and the method of choosing the samples. In the remainder of this chapter we study several of the important
sampling distributions of frequently used statistics. Applications of these sampling distributions to problems
of statistical inference are considered throughout most of the remaining chapters. The probability distribution
of X̄ is called the sampling distribution of the mean.

What Is the Sampling Distribution of X̄

We should view the sampling distributions of X̄ and S 2 as the mechanisms from which we will be able to
make inferences on the parameters µ and s2 . The sampling distribution of X̄ with sample size n is the
distribution that results when an experiment is conducted over and over (always with sample size n) and
the many values of X̄ result. This sampling distribution, then, describes the variability of sample averages

10-1
Lecture No. 10 10-2

around the population mean µ. In the case of the soft-drink machine, knowledge of the sampling distribution
of X̄ arms the analyst with the knowledge of a ”typical” discrepancy between an observed x̄ value and true
µ. The same principle applies in the case of the distribution of S 2 . The sampling distribution produces
information about the variability of s2 values around s2 in repeated experiments.

Sampling Distribution of Means and the Central Limit Theorem

The first important sampling distribution to be considered is that of the mean X̄. Suppose that a random
sample of n observations is taken from a normal population with mean µ and variance s2 . Each observation
Xi , i = 1, 2, · · ·, n, of the random sample will then have the same normal distribution as the population being
sampled. Hence, by the reproductive property of the normal distribution established in Theorem below:

Theorem 10.2 If X1 , X2 , · · ·, Xn are independent random variables having normal distributions with means
µ1 , µ2 , · · ·, µn and variances σ12 , σ22 , · · ·, σn2 , respectively, then the random variable

Y = a1 X1 + a2 X2 + · · · + an Xn

has a normal distribution with mean

µY = a1 µ1 + a2 µ2 + · · · + an µn

and variance
σY2 = a21 σ12 + a22 σ22 + · · · + a2n σn2 .

we conclude that
1
X̄ = (x1 + x2 + · · · + xn )
n
has a normal distribution with mean
1
µX̄ = (µ + µ + · · · + µ) = µ
n | {z }
n−terms

and variance
2 1 2 σ2
σX̄ = (σ + σ 2 + · · · + σ 2 ) =
n | {z } n
n−terms

If we are sampling from a population with unknown distribution, either finite or infinite, the sampling
2
distribution of X̄ will still be approximately normal with mean µ and variance σn , provided that the sample
size is large. This amazing result is an immediate consequence of the following theorem, called the Central
Limit Theorem.

Theorem 10.3 (Central Limit Theorem) If X̄ is the mean of a random sample of size n taken from a
population with mean µ and finite variance σ 2 , then the limiting form of the distribution of
X −µ
Z=
√σ
n

as n → ∞, is the standard normal distribution n(z; 0, 1).

Lecture No. 10 10-3

Remark 10.4 The normal approximation for X̄ will generally be good if n = 30, provided the population
distribution is not terribly skewed. If n < 30, the approximation is good only if the population is not too
different from a normal distribution and, as stated above, if the population is known to be normal, the sampling
distribution of X̄ will follow a normal distribution exactly, no matter how small the size of the samples. The
sample size n = 30 is a guideline to use for the Central Limit Theorem. However, as the statement of the
theorem implies, the presumption of normality on the distribution of X̄ becomes more accurate as n grows
larger. In fact, Figure 10.1 illustrates how the theorem works. It shows how the distribution of X̄ becomes
closer to normal as n grows larger, beginning with the clearly nonsymmetric distribution of an individual

Figure 10.1: Illustration of the Central Limit Theorem (distribution of X̄ for n = 1, moderate n, and large
n).

observation (n = 1). It also illustrates that the mean of X̄ remains µ for any sample size and the variance
of X̄ gets smaller as n increases.

ˆ An electrical firm manufactures light bulbs that have a length of life that is approximately normally
distributed, with mean equal to 800 hours and a standard deviation of 40 hours. Find the probability
that a random sample of 16 bulbs will have an average life of less than 775 hours.
Solution:
√ The sampling distribution of X̄ will be approximately normal, with µX̄ = 800 and σX̄ =
40/ 16 = 10. The desired probability is given by the area of the shaded region in Figure 10.2
Corresponding to x̄ = 775, we find that

Figure 10.2: Area for Example.

 Lecture No. 10 10-4

775 − 800
z= = −2.5,
10
and therefore
P (X̄ < 775) = P (Z < −2.5) = 0.0062.

Inferences on the Population Mean

One very important application of the Central Limit Theorem is the determination of reasonable values of
the population mean µ. Topics such as hypothesis testing, estimation, quality control, and many others make
use of the Central Limit Theorem. The following example illustrates the use of the Central Limit Theorem
with regard to its relationship with µ, the mean of the population, although the formal application to the
foregoing topics is relegated to future chapters. In the following case study, an illustration is given which
draws an inference that makes use of the sampling distribution of X̄. In this simple illustration, µ and s are
both known. The Central Limit Theorem and the general notion of sampling distributions are often used
to produce evidence about some important aspect of a distribution such as a parameter of the distribution.
In the case of the Central Limit Theorem, the parameter of interest is the mean µ. The inference made
concerning µ may take one of many forms. Often there is a desire on the part of the analyst that the data
(in the form of x̄) support (or not) some predetermined conjecture concerning the value of µ. The use of
what we know about the sampling distribution can contribute to answering this type of question. In the
following case study, the concept of hypothesis testing leads to a formal objective that we will highlight in
future chapters.

Case Study 10.1 Automobile Parts: An important manufacturing process produces cylindrical component parts for
the automotive industry. It is important that the process produce parts having a mean diameter of
5.0 millimeters. The engineer involved conjectures that the population mean is 5.0 millimeters. An
experiment is conducted in which 100 parts produced by the process are selected randomly and the
diameter measured on each. It is known that the population standard deviation is s = 0.1 millimeter.
The experiment indicates a sample average diameter of x̄ = 5.027 millimeters. Does this sample
information appear to support or refute the engineer’s conjecture?
Solution: This example reflects the kind of problem often posed and solved with hypothesis testing
machinery introduced in future chapters. We will not use the formality associated with hypothesis
testing here, but we will illustrate the principles and logic used. Whether the data support or refute
the conjecture depends on the probability that data similar to those obtained in this experiment
(x̄ = 5.027) can readily occur when in fact µ = 5.0 (Figure 10.3). In other words, how likely is it that
one can obtain x̄ = 5.027 with n = 100 if the population mean is µ = 5.0? If this probability suggests
that x̄ = 5.027 is not unreasonable, the conjecture is not refuted. If the probability is quite low, one
can certainly argue that the data do not support the conjecture that µ = 5.0. The probability that
we choose to compute is given by P (|X̄ − 5| = 0.027) In other words, if the mean µ is 5, what is the
chance that X̄ will deviate by as much as 0.027 millimeter?

P (|X̄ − 5| = 0.027) = P (X̄ − 5 = 0.027) + P (X̄ − 5 = −0.027) (10.1)

 
X̄ − 5
= 2P √ ≥ 2.7 . (10.2)
0.1/ 100

Here we are simply standardizing X̄ according to the Central Limit Theorem. If the conjecture µ = 5.0
X̄−5
is true, 0.1/ √
100
should follow N (0, 1). Thus,
 
X̄ − 5
2P √ ≥ 2.7 = 2P (Z ≥ 2.7) = 2(0.0035) = 0.007.
0.1/ 100
Lecture No. 10 10-5

Figure 10.3: Area for Case Study

Therefore, one would experience by chance that an x̄ would be 0.027 millimeter from the mean in only
7 in 1000 experiments. As a result, this experiment with x̄ = 5.027 certainly does not give supporting
evidence to the conjecture that µ = 5.0. In fact, it strongly refutes the conjecture!
ˆ Traveling between two campuses of a university in a city via shuttle bus takes, on average, 28 minutes
with a standard deviation of 5 minutes. In a given week, a bus transported passengers 40 times. What
is the probability that the average transport time was more than 30 minutes? Assume the mean time
is measured to the nearest minute.
Solution: In this case, µ = 28 and σ = 3. We need to calculate the probability P (X̄ > 30) with
n = 40. Since the time is measured on a continuous scale to the nearest minute, an x̄ greater than 30
is equivalent to x̄ = 30.5. Hence,
 
X̄ − 28 30.5 − 28
P (X̄ > 30) = P √ ≥ √ = P (Z ≥ 3.16) = 0.0008.
5/ 40 5/ 40

There is only a slight chance that the average time of one bus trip will exceed 30 minutes. An

Figure 10.4: Area for Example.

illustrative graph is shown in Figure 10.4.

 Lecture No. 10 10-6

Sampling Distribution of the Difference between Two Means

The illustration in above Case Study deals with notions of statistical inference on a single mean µ. The
engineer was interested in supporting a conjecture regarding a single population mean. A far more important
application involves two populations. A scientist or engineer may be interested in a comparative experiment
in which two manufacturing methods, 1 and 2, are to be compared. The basis for that comparison is µ1 -
µ2 , the difference in the population means. Suppose that we have two populations, the first with mean µ1
and variance σ12 , and the second with mean µ2 and variance σ22 . Let the statistic X̄1 represent the mean of
a random sample of size n1 selected from the first population, and the statistic X̄2 represent the mean of
a random sample of size n2 selected from the second population, independent of the sample from the first
population. What can we say about the sampling distribution of the difference X̄1 − X̄2 for repeated samples
of size n1 and n2 ? According to Central Limit Theorem, the variables X̄1 and X̄2 are both approximately
σ2 σ2
normally distributed with means µ1 and µ2 and variances n11 and n22 , respectively. This approximation
improves as n1 and n2 increase. By choosing independent samples from the two populations we ensure that
the variables X̄1 and X̄2 will be independent, and then using Theorem 10.2, with a1 = 1 and a2 = −1, we
can conclude that X̄1 − X̄2 is approximately normally distributed with mean

µX̄1 −X̄2 = µX̄1 − µX̄2 = µ1 − µ2

and variance
σ12 σ2
2
σX̄ 1 −X̄2
2
= σX̄ 2
− σX̄ = − 2
1 2
n1 n2
The Central Limit Theorem can be easily extended to the two-sample, two-population case.

Theorem 10.5 If independent samples of size n1 and n2 are drawn at random from two populations, discrete
or continuous, with means µ1 and µ2 and variances σ12 and σ22 , respectively, then the sampling distribution
of the differences of means, X̄1 − X̄2 , is approximately normally distributed with mean and variance given
by
µX̄1 −X̄2 = µ1 − µ2
and
σ12 σ2
2
σX̄ 1 −X̄2
= − 2.
n1 n2
Hence,
(X̄1 − X̄2 ) − (µ1 − µ2 )
Z= r   
σ12 σ22
n1 + n2

is approximately a standard normal variable.

If both n1 and n2 are greater than or equal to 30, the normal approximation for the distribution of X̄1 − X̄2
is very good when the underlying distributions are not too far away from normal. However, even when
n1 and n2 are less than 30, the normal approximation is reasonably good except when the populations are
decidedly nonnormal. Of course, if both populations are normal, then X̄1 − X̄2 has a normal distribution no
matter what the sizes of n1 and n2 are. The utility of the sampling distribution of the difference between
two sample averages is very similar to that described in Case Study 10.1 for the case of a single mean. Case
Study 10.2 that follows focuses on the use of the difference between two sample means to support (or not)
the conjecture that two population means are the same.

Case Study 10.2 Paint Drying Time: Two independent experiments are run in which two different types of paint are
compared. Eighteen specimens are painted using type A, and the drying time, in hours, is recorded
Lecture No. 10 10-7

for each. The same is done with type B. The population standard deviations are both known to be
1.0. Assuming that the mean drying time is equal for the two types of paint, find P (X̄A − X̄B > 1.0),
where X̄A and X̄B are average drying times for samples of size
nA = nB = 18.
Solution: From the sampling distribution of X̄A − X̄B , we know that the distribution is approximately
normal with mean
µX̄A −X̄B = µA − µB = 0
and
2
2 σA σ2 1 1 1
σX̄ A −X̄B
= + B = + =
nA nB 18 18 9
The desired probability is given by the shaded region in Figure 10.5. Corresponding to the value

Figure 10.5: Area for Example.

X̄A − X̄B = 1.0, we have

1 − (µX̄A −X̄B ) 1−0
z= p =p = 3.0
1/9 1/9
or
P (Z > 3.0) = 1 − P (Z < 3.0) = 1 − 0.9987 = 0.0013.
ˆ The television picture tubes of manufacturer A have a mean lifetime of 6.5 years and a standard
deviation of 0.9 year, while those of manufacturer B have a mean lifetime of 6.0 years and a standard
deviation of 0.8 year. What is the probability that a random sample of 36 tubes from manufacturer A
will have a mean lifetime that is at least 1 year more than the mean lifetime of a sample of 49 tubes
from manufacturer B?
Solution: We are given the following information: If we use Theorem 8.3, the sampling distribution

of X̄1 − X̄2 will be approximately normal and will have a mean and standard deviation
µX̄1 −X̄2 = 6.5 − 6.0 = 0.5
Lecture No. 10 10-8

and r
0.36 0.64
σX̄1 −X̄2 = + = 0.189.
81 49
The probability that the mean lifetime for 36 tubes from manufacturer A will be at least 1 year longer
than the mean lifetime for 49 tubes from manufacturer B is given by the area of the shaded region in
Figure 10.6. Corresponding to the value x̄1 − x̄2 = 1.0, we find that
1.0 − 0.5
z= = 2.65,
0.189
P (X̄1 − X̄2 ≥ 1.0) = P (Z > 2.65) = 1 − P (Z < 2.65) = 1 − 0.9960 = 0.0040.

Figure 10.6: Area for Example.

References
[TT] T.T. Soong, “Fundamentals of probability and statistics for engineers,” John Wiley & Sons
Ltd, The Atrium, Southern Gate, Chichester, West Sussex PO19 8SQ, England, 2004.