CHAPTER 6

SAMPLING AND
SAMPLING DISTRIBUTIONS

6-1 The major drawback of judgmental sampling is the lack of a measure of the validity of the sample
as representative of the population.

6-2 They are not necessarily mutually exclusive. For example, a situation could arise in which the
history ( or lack of history ) of the population of interest, indicates no clear-cut trends. Thus, the
personal opinion of the individual conducting the study could be that probability sampling is best.

6-3 Sampling is less costly, less time-consuming, more convenient, and more efficient than complete
enumeration in determining something about judgment samples.

6-4 Probability Sampling involves more rational analysis and planning at the beginning of a study and
usually takes more time and money than judgment samples.

6-5 It is probably both. It is a parameter for the population of shareholders attending the meeting.
However, unless all shareholders were at the meeting, it is only a sample statistic for the
population of the shareholders.

6-6 From what we have been told in the problem, Jean’s position is apparently quite defensible.
Perhaps what makes statistical sampling unique is that it permits statistical inference to be made
about a population and its parameters. This is apparently what jean has done. There are no hard
and fast rules as to the size of the sample that must be drawn before inferences can be made.
Specifically there is nothing magic about the 50% mark. Common sense would seem to point out
that gathering data from 50% of some population might tend to be almost as difficult as gathering
data from the entire population-for instance, the population of the United States, or the world. The
defense for jean’s position lies in empirical evidence and some explanation and reasoning with the
project leaders, educating them about the abilities of statistical inference.

6-7 If the deceased victims have relative or friends who were told of the heater’s behavior by the
deceased, and if they could be located, the census could be completed. Otherwise, only a census of
surviving victims of some other reduced population is possible.

6-8 In (b), the distributions have greater between-group variance and less within-group variance than
in (a).

6-9 To make life easy, we’ll ignore pages 1000-1026. Starting at the 4th line of the 2nd column, we’ll
read groups of 3 digits to identify the selected pages. When we reach the end of the 4th line of the
5th column, we’ll go to the 5th line in the 1st column. The selected pages and numbers of words in
italics are given below. (we did not count symbols or number as words.)

119 

 
 Page : 543 581 078 896 708 529 131 291 265 730 489 003 130 500 995

Words: 8 1 6 0 4 0 0 0 0 12 0 32 0 2 0

In all like likelihood, your results were different, because you used a different way to pick your
pages from Table 6-3.

6-10 Assuming a non-leap year:

1/6 1/24 2/11 3/1 3/19 4/6 4/24 5/12 5/30 6/17

7/5 7/23 8/10 8/28 9/15 10/3 1021 11/8 11/26 12/14

6-11 (c)

6-12 The probability that a 4, 7, or 2 will appear is .10, since each digit is equally likely to be selected
by a random number generator. Since there are 115 ten digit numbers, we would expect to see each
number appear 11.5 times. Actually, the numbers appear as follows:
Number 4 7 2
Appearances 16 13 10
Assuming that our table is random, we would expect, in a larger version of this table, the number
of appearances of any digit in any given position to be close to one tenth of the possible
appearances. The large deviations from 12.5 in this case are 'due to insufficient sample size.

6-13 Since there is so much variation within each group, stratified sampling would not be
recommended. Since the groups are very different from each other, cluster sampling is also
inappropriate. Random sampling is reasonable, because a list of all the subscribers is readily
available. Depending on the order in which the subscribers are listed, systematic sampling may or
may not be advisable.

6-14 No. Between noon and 5 pm on a weekday, no one will be at home if both parents work, and their
children are of pre-school age. Thus, the heaviest users of daycare centers are excluded from a poll
on daycare centers.

6-15 Very possibly not. The population that Peterson is interested in is all of Piedmont's customers. The
intersection of the set of customers and the set of persons with telephones may be large enough for
Peterson to make inferences, but quite probably it is not. Peterson's poll might (1) include homes
without electricity and (2) definitely does not include customers who do not have phones. If a large
percentage of the population of the service area does not have telephone service, it would be very
dangerous for Peterson to infer too much from his sample. In short, for his purposes, the sample is
probably not random.

120 

 
6-16 Every seventh is better. If every fifth lot were inspected, then the output of only two machines
would be inspected. If every seventh lot is inspected, then each machine's output will be inspected.

6-17 Yes. Systematic sampling is particularly useful when the population being sampled is randomly
distributed. In this case, accidents are filed by date. For the purpose that the department has in
mind, such a filing is random. The department has no control over when the accidents happen 80
there seems little chance that there is any underlying systematic distribution; hence the results
obtained from the sample will be representative of the population.

6-18 We need to remember that when we use stratified sampling, we are only acknowledging that the
population is already divided into groups of different sizes. However, these groups are required to
be relatively homogeneous before we can use this sampling method. In the case under question,
Mary proposes dividing the population into groups such as urban, suburban, and rural; examination
of the differences among these groups should convince you that they are reasonably homogeneous,
and thus stratified sampling would work.

6-19 a) The only way to guarantee that there will be no sample error (guarantee sample mean is
equal to population mean) is to sample the entire population.
 
b) No sample size will guarantee a zero standard error of the mean, unless each member of the
population has the same value i. sample entire population.

6-20 Sampling error, the chance error which is due to the particular elements selected as a sample of
the population.

6-21 Your box fell short of the company's claim by 0.1 cups, which is only 0.5 standard deviations.
Since this is so close to the claimed average, you have no reason to doubt the validity of the claim.

6-22 A sample mean overestimating the true mean is no better than one under- estimating the true
mean. In this case.. 30 cents,is closer than 35 cents to the true mean, in absolute differences:
 30  31.4  = 31.4  30 = 1.4 <  35 – 31.411 = 35  31.4 = 3.6

6-23 No. In a sampling distribution of the mean, all samples are of the same size. It is very unlikely that
she interviewed the same number of people each day, unless she did so intentionally. In any case,
the sampling distribution in question is the distribution of the means of all possible samples of the
given size, and she certainly has not looked at all such samples.

6-24 Average weekly sales have decreased from 3538 cartons to 3462 cartons.

6-25 No. The standard deviation of the mean of all possible samples is referred to as the standard error
of he mean. It is a measure of the dispersion of the theoretical sampling distribution rather than, as
the secretary assumes, an allowable deviation from the population mean.

121 

 
6-26 The information gathered concerns mean customer satisfaction for groups of 30 customers, not for
single, customers, so it is a sample from the sampling distribution of the mean of samples of me 30
drawn from the customer population. It is not a sample from the customer population.

6-27 n=16 =150 ² =256 = / = 16/ =4

a) P( < 160) = P = P(z < 2.5) = .5 + .4938 = .9938

b) P( >142) = P = P(z >  2) = .5 + .4772 = .9772

n == 9 = / = 16/ = 5.33

c) P( < 160) = P = P(z < 1.88) = .5 + .4699 = .9699

d) P( > 142) = P = P(z >  1.5) = .5 + .4332 = .9332

6-28 n=19 =18  =4.8 = / = 4.8/ =1.101

P(16 < < 20) = P

= P(1.82 < z < 1.82) = 2(.4656) =.9312

b) The same as (a), since the distribution is continuous

c) n=48, =18  =4.8 = / = 4.8/ = 0.693

P(16 < < 20) = P

= P(2.89 < z < 2.89) =2(.4981) =.9962

6-29  = 56  = 21 P(z > 1.28) = .90

Want P(x > 52) = .90 => 1.28 => 6.72 => n  45.16
A sample size of 46 will be sufficient.

6-30  = 315  = 48 P(-1.96 < z < 1.96) = .95

Want P(370 < < 380)  .95 =>  .475 =>

 1.96 =>  18.816 => n  354.04

122 

 
 A sample size of 355 will be sufficient.

6-31  = 50 ² = 9  = 3

a) n=4 = / = 3/ =1.5

P(  48) = p = P(z  1.33) = .5 + .4082 = .9082

b) The probability will rise, because additional sampling decreases the

standard error of the, sampling distribution. Thus, P(48   50) increases,
meaning that
P( 48) = .5 + P(48   50) increases.

6-32  = $62,000  = $4,200

a) n=1 = / = $4,200

P( 65,000) =p = P(z  .71)

= .5  .2611 = .2389

b) The probability is less. More sampling decreases the standard

error of the sampling distribution of the mean.

n=2 = $4200/ = $2970

P( 65000) =p = P(z  1.01)

= .5  .3438 = .1562

The decrease is .0827, or 35%.

6-33 The sample size of 75 is large enough to use the central limit theorem.

 = 86  = 16 n = 75 = / = 16/ = 1.848

P( < 84) =p = P(z <  1.08) = .5  .3599 = .1401

P( > 90) = = P(z > 2.16) = .5 - .4846 = .0154

Thus, P( <84 or > 90) = .1401 + .0154 = .1555

123 

 
6-34 a)  = 4300  = 730 n = 3 = / = 730/ = 421.5
For a set of monitors to last 13000 hours, they must each last
4333.33 hours on average.

P(  4333.33) =
= P(z > .08) =.5  .0319 =.4681

b) For the set to last at most 12630 hours, the average life cannot exceed
4210 hours.

P(  4210) =
= P(z  .21) =.5 - .0832 =.4168
6-35  =64  = = 4.195 n =35 =/ = 4.195/ = 0.709

a) P( > 72) = P = P(z > 11.28)  0

b) P(64 < < 72) = P(O < z < 11.28)  0.5

c) P( = 64) = P(z=0) = 0

d) P( > 94) = p = P(z > 42.31)  0

e) No. A sample average of 100 ppm is so far above what is

expected, and hence so extremely unlikely, that it casts very strong doubt
on the reliability of the study.

6-36 The sample size of 48 is large enough to use the central limit theorem.
 =110  = 64 n =35 = / = 64/ = 9.238

P(  120) =
= P(z  .1.08) =.5 - .3599 =.8599  .80
The overhaul will not be ordered.

6-37  =7500  = 3300 n =1800 = / = 3300/ = 77.782

a) P(  7700) =
124 

 
 = P(z  2.57) =.5  .4949 =.0051

b) P(  7400) =
= P(z  1.29) =.5  .4015 =.0985

c) P(7275   7650) =
= P(2.89  z  1.93) = .4981 + .4732 =.9713

6-38  = 120  =12 n = 60

a) =  = 120
b) = / = 12/ = 1.549

c) P(  123.8) =
= P(z  2.45 ) =.5  .4929 =.0071

d) ) P(117   122) =
= P(1.94  z  1.29) = .4738 + .4015 =.8753

6-39  =168  = =19 n =1800 = / = 19/ = 3.8

a)P(total weight > 4250) = P(  170) =

= P(z  0.53 ) =.5 - .2019 =.2981

b) Note : .05 = p(x > 1.64)

We want to find c such that P(total weight > c) = .05

p = .05 p = .05

p = .05 = 1.64 c – 4200 = 155.8

c = 4355.8 is the 95th percentile of the distribution of total weight of
ferry passengers.

The ferry is not complying with safety regulations.

125 

 
6-40 N = 75 n = 32 µ = 364 = = 4.243

a)

b) P(366 = p
= p (1.75 3.50) = .4599 + .5 = .9599

c) With replacement, =/ = 4.243 / = 0.750

6.41 N = 80 µ = 22  = 3.2 n = 25

a)

b) P(21 = p
= p (-1.87 2.81) = .4693 + .4975= .9668

6-42 N = 80 n = 2.1 µ = 8.2

a) n = 16 = .4725

b) n=25 = .3504

c) n =49 = .1879

6-43 µ unknown  = 216.4

a) n = 800 = 7.651

P(µ = p
= p (0 39.21) 0.5

b) 0.95 = P(196 = p

Hence 1.96 = = 0.4621 so n = = 17.99

126 

 
 They should sample at least 18 tires.

6-44 N = 45 n = 9 µ = 225,000  = 39,000

11,759

P(9  2,100,000 = p

= p
= p (z  .71) = .5 – 2611 = .2389

6-45  = 150

We want to find n such that =>  6 => n  36

She should take at least 36 readings.

6-46 With 250(.36) = 90 contributors, the average donation must be between $1222.22 and
$1333.33 for the total to be between $112,950 and $116,100. Each 4% gift has
µ = .04(32,000) = $1,280 and  = .04(9,600) = $384.

32.4467

P(1222.22   1333.33) =
= P(1.78  z  1.64) = .4625 + .4495 =.9120

6-47  =96  = 7 n =6 = / = 7/ = 2.858

P(  98) =
= P(z  .70) =.5 - .2580 =.2420

127 

 
 (The population is the large set of planes on which the de-icing system can be installed, not
just the 30 planes on which it has been installed so far.' Thus, it is incorrect to use the finite
population multiplier for this problem.)

6-48 N = 145 n = 36  =1200 174.01

P(µ200  µ + 200) =
= P(1.15 z  1.15) = .3749 + .3749=.7498

6-49 µ= 310  = 150

We want to find an n large enough so that

=  (.015)(310) = 4.65 150  4.65

=  32.285 n  1040.58
Thus, a sample size of at least 1041 is needed.

6-50 (Ex 6.5/Pg. No. 318-19)

 = 90 minutes=1.5 hrs, degree of confidence (1-α) 95% so Z= 1.96, margin of error

allowed (e) =0.5hrs
n = (* Z/ e)2 = (1.5* 1.96/0.5)2 = 34.57 = 35

6-51 (Ex 6.5/Pg. No. 318-19)

e = .03, p =0.4, q= 0.6, 1-α= 95% so Z= 1.96.

n = Z2 * pq/ e2
= 1.962 * 0.4* 0.6/ (.03)2 = 102.44 =103

6-52 (Ex 6.5/Pg. No. 318-19)

e = .04, pq =0.25 (p, q are unknown so taking maximum possible value of pq i.e 0.25), 1-
α= 95% so Z= 1.96.
Z = e / (pq)1/2 * {(N-n)/N}1/2
Solving for n, n= 545.66 = 546

Review & Application Exercise (Pg. No. 323-325)

6-50 Judgmental, because the sample (those skates inspected) is determined by whether the skate
is Crash's size.

128 

 
6-51 Proponents of random sampling often argue that it is better than judgmental sampling
because more statistical inferences can be made and the information obtained is more
reliable. In many situations, these arguments are true. In this case, the opposite may be true.
Judgmental sampling seems o have worked well. Given the cost of the proposed
alternative, it may be better to leave well enough alone.

6-52 Stratified sampling is used when we have reason to believe that we can divide the
population into groups which are relatively homogeneous. In this instance, the manager
feels that the residents fall into various age and income levels. Since he is trying to measure
residents' attitudes, stratified sampling may be his best method, especially since he feels
residents in different age and income groups may have different attitudes.

6-53 No. It must be noted that the employees of different departments probably have varying
backgrounds and concerns. For these reasons, cluster sampling will not give representative
results to the management. The problems in a research and development department may
be entirely different from those in an assembly department. Similarly, cultural differences
may be . prevalent in the different departments. The company would obtain better and more
representative results if it incorporated input from every department into its plan.

6-54  = 26  = 5.65 = 5.65/

We want to find n such that: p(25   27)  .9544 = p(-2  z  2)

p  p(-2  z  2)

p(.177 )  p(-2  z  2)

.177  2  11.3 n  127.7

Thus, a sample of at least 128 customers is needed.

6-55 a) Rate Rate

State x x2 State x x2

Alabama 7.5 56.25 Montana 7.3 53.29

Alaska 10.1 102.01 Nebraska 2.8 7.84
Arizona 8.4 70.56 Nevada 6.8 46.24
Arkansas 7.0 49.00 New Hampshire 7.5 56.25
California 8.7 75.69 New Jersey 7.5 56.25
Colorada 6.3 39.69 New Mexico 7.6 57.76
Connecticut 7.4 54.76 New York 8.5 72.25
Delaware 6.4 40.96 North Carolina 6.4 40.96
District of Columbia 8.2 67.24 North Dakota 5.3 28.09
Florida 8.1 65.61 Ohio 7.8 60.84
129 

 
 Georgia 6.3 39.69 Oklahoma 6.8 46.24
Hawaii 3.5 12.25 Oregon 8.6 73.96
Idaho 7.8 60.84 Pennsylvania 7.6 57.76
Illinois 8.2 67.24 Rhode lsland 8.9 79.21
Indiana 6.3 39.69 South Carolina 7.1 50.41
Iowa 5.3 28.09 South Dakota 4.0 16.00
Kansas 3.6 12.96 Tennessee 7.0 49.00
Kentucky 7.0 49.00 Texas 7.4 54.76
Louisiana 6.9 47.61 Utah 5.0 25.00
Maine 8.4 70.56 Vermont 7.1 50.41
Maryland 1A 54.76 Virginia 6.8 46.24
Massachusetts 10.0 100.00 Washington 8.3 68.89
Michigan 10.0 100.00 West Virginia 12.9 168.49
Minnesota 6.3 39.69 Wisconsin 5.7 32.49
Mississippi 8.1 65.61 Wyoming 7.5 56.25
Missouri 5.6 31.36 367.0 2793.92

 = ∑x/N = 367.0/51 = 7.196

² = ∑x²/N - ² = 2793.92/51 – 7.196² = 3.000

= = = 1.732

b) = ∑x/n = (7.5+3.6+10.0+2.8+6.4+)/5=6.06

c) =  = 7.196, = = 0.743

d) Probably not. The sample size is small (n=5), and the population itself, with
N=51, is too small to be considered even approximately normal.
e) p(5.9   6.5) = p(( 5.9 – 7.196)/0.743  z  (6.5 – 7.196)/0.743
= p(-1.74  z  -0.94) = 0.4591 – 0.3264 = 0.1327

6-56 In this situation, Fargo Lanna is not constrained by(1) cost, (2) time, (3) destruction of
population members, or (4) accessibility of the population, Nevertheless, sampling would be entuely
appropriate for this company's purpose. It will be able to obtain the same information wit.houtexpending
nearly the effort required to poll all employees. The company may get the e job done by using a sample,
and will be able to reassign the clerical staff earlier.

6-57 There is no truth to what Simmons said. We would expect that the sample mean would
overstate the population mean as often as it would understate it. Sample variation works in 'both ways, and
to assume that the sample mean always understates is to be guilty of an error in thinking.

130 

 
6-58 Again Simmons is incorrect. A sampling distribution of means is a frequency distribution
of the .means of all possible samples. It is not in any sense a graph of the individual
observations in

6-59  = 275 n=25 => =275/ = 55

If = = 55/2 = 27.5, there must be a sample size n* such that
= 275/ 27.5 => 275/27.5 => 10 => n*= 100
Sampling another 75 women costs $1600 but increases the benefits by $1720. She should
increase the sample size.

6-60  = 42  = 11 n= 5 = /n = 11/ = 4.919

P(  50) = = P(z  1.63 ) =.5 - .4484 =.0516

6-61  = 41  = 8 n= 6 = /n = 8/ = 3.266

P(  50) = = P(z  2.76 ) =.5 + .4971 =.9971

6-62  = 5.8  = .8 n= 20 = /n = .8/ = .1789

P(5.5  6.2) =
= P(-1.68 z 2.24) = .4535 + .4875 =.9410

6-63 Cost = Benefit

4n = = 19.81 = = 4.
95 n = 24.5

Thus, she should sample at least 25 detectors.

6-64 N=70 n=15  = 18 = 4

.9221

a) P(  15.5) = = P(z  -2.71) =.5 - .4966 =.0034

131 

 
 b) P(  20) = = P(z  2.17) =.5 - .4850 =.0150

132