lOMoARcPSD|40821539

Ch02

Design Of Experiments (Texas Tech University)

Scan to open on Studocu

Studocu is not sponsored or endorsed by any college or university

Downloaded by Sara ([email protected])
 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

Chapter 2
Simple Comparative Experiments
Solutions

2.1. Computer output for a random sample of data is shown below. Some of the quantities are
missing. Compute the values of the missing quantities.

Variable N Mean SE Mean Std. Dev. Variance Minimum Maximum

Y 9 19.96 ? 3.12 ? 15.94 27.16

SE Mean = 1.04 Variance = 9.73

2.2. Computer output for a random sample of data is shown below. Some of the quantities are
missing. Compute the values of the missing quantities.

Variable N Mean SE Mean Std. Dev. Sum

Y 16 ? 0.159 ? 399.851

Mean = 24.991 Std. Dev. = 0.636

2.3. Suppose that we are testing H0: µ = µ0 versus H1: µ ≠ µ0. Calculate the P-value for the following
observed values of the test statistic:

(a) Z0 = 2.25 P-value = 0.02445

(b) Z0 = 1.55 P-value = 0.12114

(c) Z0 = 2.10 P-value = 0.03573

(d) Z0 = 1.95 P-value = 0.05118

(e) Z0 = -0.10 P-value = 0.92034

2.4. Suppose that we are testing H0: µ = µ0 versus H1: µ > µ0. Calculate the P-value for the
following observed values of the test statistic:

(a) Z0 = 2.45 P-value = 0.00714

(b) Z0 = -1.53 P-value = 0.93699

2-1

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(c) Z0 = 2.15 P-value = 0.01578

(d) Z0 = 1.95 P-value = 0.02559

(e) Z0 = -0.25 P-value = 0.59871

2.5. Consider the computer output shown below.

One-Sample Z
Test of mu = 30 vs not = 30
The assumed standard deviation = 1.2
N Mean SE Mean 95% CI Z P

16 31.2000 0.3000 (30.6120, 31.7880) ? ?

(a) Fill in the missing values in the output. What conclusion would you draw?

Z=4 P = 0.00006; therefore, the mean is not equal to 30.

(b) Is this a one-sided or two-sided test?

Two-sided.

(c) Use the output and the normal table to find a 99 percent CI on the mean.

CI = 30.42725, 31.97275

(d) What is the P-value if the alternative hypothesis is H1: µ > 30

P-value = 0.00003

2.6. Suppose that we are testing H0: µ1 = µ2 versus H1: µ1 = µ2 with a sample size of n1 = n2 = 12.
Both sample variances are unknown but assumed equal. Find bounds on the P-value for the
following observed values of the test statistic:

(a) t0 = 2.30 Table P-value = 0.02, 0.05 Computer P-value = 0.0313

(b) t0 = 3.41 Table P-value = 0.002, 0.005 Computer P-value = 0.0025

(c) t0 = 1.95 Table P-value = 0.1, 0.05 Computer P-value = 0.0640

(d) t0 = -2.45 Table P-value = 0.05, 0.02 Computer P-value = 0.0227

Note that the degrees of freedom is (12 +12) – 2 = 22. This is a two-sided test

2.7. Suppose that we are testing H0: µ1 = µ2 versus H1: µ1 > µ2 with a sample size of n1 = n2 = 10.
Both sample variances are unknown but assumed equal. Find bounds on the P-value for the
following observed values of the test statistic:

(a) t0 = 2.31 Table P-value = 0.01, 0.025 Computer P-value = 0.01648

2-2

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(b) t0 = 3.60 Table P-value = 0.001, 0.0005 Computer P-value = 0.00102

(c) t0 = 1.95 Table P-value = 0.05, 0.025 Computer P-value = 0.03346

(d) t0 = 2.19 Table P-value = 0.01, 0.025 Computer P-value = 0.02097

Note that the degrees of freedom is (10 +10) – 2 = 18. This is a one-sided test.

2.8. Consider the following sample data: 9.37, 13.04, 11.69, 8.21, 11.18, 10.41, 13.15, 11.51, 13.21,
and 7.75. Is it reasonable to assume that this data is from a normal distribution? Is there evidence to
support a claim that the mean of the population is 10?

Minitab Output

According to the output, the Anderson-Darling Normality Test has a P-Value of 0.435. The data can
be considered normal. The 95% confidence interval on the mean is (9.526,12.378). This confidence
interval contains 10, therefore there is evidence that the population mean is 10.

2.9. A computer program has produced the following output for the hypothesis testing problem:

Difference in sample means: 2.35

Degrees of freedom: 18
Standard error of the difference in the sample means: ?
Test statistic: to = 2.01
P-Value = 0.0298

2-3

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(a) What is the missing value for the standard error?

y1 − y2 2.35
t0 = = = 2.01
1 1 StdError
Sp +
n1 n2
StdError = 2.35 / 2.01 = 1.169
(b) Is this a two-sided or one-sided test? One-sided test for a t0 = 2.01 is a P-value of 0.0298.

(c) If α=0.05, what are your conclusions? Reject the null hypothesis and conclude that there is a
difference in the two samples.

(d) Find a 90% two-sided CI on the difference in the means.

1 1 1 1
y1 − y2 − tα 2, n1 + n2 − 2 Sp + ≤ µ1 − µ1 ≤ y1 − y2 + tα 2, n1 + n2 − 2 Sp +
n1 n2 n1 n2
1 1 1 1
y1 − y2 − t0.05,18S p + ≤ µ1 − µ1 ≤ y1 − y2 + t0.05,18S p +
n1 n2 n1 n2
2.35 − 1.734 (1.169 ) ≤ µ1 − µ1 ≤ 2.35 + 1.734 (1.169 )
0.323 ≤ µ1 − µ1 ≤ 4.377

2.10. A computer program has produced the following output for the hypothesis testing problem:

Difference in sample means: 11.5

Degrees of freedom: 24
Standard error of the difference in the sample means: ?
Test statistic: to = -1.88
P-Value = 0.0723

(a) What is the missing value for the standard error?

y1 − y2 −11.5
t0 = = = −1.88
1 1 StdError
Sp +
n1 n2
StdError = −11.5 / −1.88 = 6.12
(b) Is this a two-sided or one-sided test? Two-sided test for a t0 = -1.88 is a P-value of 0.0723.

(c) If α=0.05, what are your conclusions? Accept the null hypothesis, there is no difference in the
means.

2-4

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(d) Find a 90% two-sided CI on the difference in the means.

1 1 1 1
y1 − y2 − tα 2, n1 + n2 − 2 Sp + ≤ µ1 − µ1 ≤ y1 − y2 + tα 2, n1 + n2 − 2 Sp +
n1 n2 n1 n2
1 1 1 1
y1 − y2 − t0.05,24 S p + ≤ µ1 − µ1 ≤ y1 − y2 + t0.05,24 S p +
n1 n2 n1 n2
−11.5 − 1.711 ( 6.12 ) ≤ µ1 − µ1 ≤ −11.5 + 1.711( 6.12 )
−21.97 ≤ µ1 − µ1 ≤ −1.03

2.11. A two-sample t-test has been conducted and the sample sizes are n1 = n2 = 10. The computed
value of the test statistic is t0 = 2.15. If the null hypothesis is two-sided, an upper bound on the P-
value is

(a) 0.10

(b) 0.05

(c) 0.025

(d) 0.01

(e) None of the above.

2.12. A two-sample t-test has been conducted and the sample sizes are n1 = n2 = 12. The computed
value of the test statistic is t0 = 2.27. If the null hypothesis is two-sided, an upper bound on the P-
value is

(a) 0.10

(b) 0.05

(c) 0.025

(d) 0.01

(e) None of the above.

2.13. Suppose that we are testing H0: µ = µ0 versus H1: µ > µ0 with a sample size of n = 15. Calculate
bounds on the P-value for the following observed values of the test statistic:

(a) t0 = 2.35 Table P-value = 0.01, 0.025 Computer P-value = 0.01698

(b) t0 = 3.55 Table P-value = 0.001, 0.0025 Computer P-value = 0.00160

(c) t0 = 2.00 Table P-value = 0.025, 0.005 Computer P-value = 0.03264

2-5

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(d) t0 = 1.55 Table P-value = 0.05, 0.10 Computer P-value = 0.07172

The degrees of freedom are 15 – 1 = 14. This is a one-sided test.

2.14. Suppose that we are testing H0: µ = µ0 versus H1: µ ≠ µ0 with a sample size of n = 10. Calculate
bounds on the P-value for the following observed values of the test statistic:

(a) t0 = 2.48 Table P-value = 0.02, 0.05 Computer P-value = 0.03499

(b) t0 = -3.95 Table P-value = 0.002, 0.005 Computer P-value = 0.00335

(c) t0 = 2.69 Table P-value = 0.02, 0.05 Computer P-value = 0.02480

(d) t0 = 1.88 Table P-value = 0.05, 0.10 Computer P-value = 0.09281

(e) t0 = -1.25 Table P-value = 0.20, 0.50 Computer P-value = 0.24282

2.15. Consider the computer output shown below.

One-Sample T: Y
Test of mu = 91 vs. not = 91
Variable N Mean Std. Dev. SE Mean 95% CI T P

Y 25 92.5805 ? 0.4675 (91.6160, ? ) 3.38 0.002

(a) Fill in the missing values in the output. Can the null hypothesis be rejected at the 0.05 level?
Why?

Std. Dev. = 2.3365 UCI = 93.5450

Yes, the null hypothesis can be rejected at the 0.05 level because the P-value is much lower at
0.002.

(b) Is this a one-sided or two-sided test?

Two-sided.

(c) If the hypothesis had been H0: µ = 90 versus H1: µ ≠ 90 would you reject the null hypothesis at
the 0.05 level?

Yes.

(d) Use the output and the t table to find a 99 percent two-sided CI on the mean.

CI = 91.2735, 93.8875

(e) What is the P-value if the alternative hypothesis is H1: µ > 91?

P-value = 0.001.

2-6

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

2.16. Consider the computer output shown below.

One-Sample T: Y
Test of mu = 25 vs > 25
Variable N Mean Std. Dev. SE Mean 95% Lower Bound T P

Y 12 25.6818 ? 0.3360 ? ? 0.034

(a) How many degrees of freedom are there on the t-test statistic?

(N-1) = (12 – 1) = 11

(b) Fill in the missing information.

Std. Dev. = 1.1639 95% Lower Bound = 2.0292

2.17. Consider the computer output shown below.

Two-Sample T-Test and CI: Y1, Y2

Two-sample T for Y1 vs Y2
N Mean Std. Dev. SE Mean

Y1 20 50.19 1.71 0.38

Y2 20 52.52 2.48 0.55

Difference = mu (X1) – mu (X2)

Estimate for difference: -2.33341

95% CI for difference: (-3.69547, -0.97135)

T-Test of difference = 0 (vs not = ) : T-Value = -3.47

P-Value = 0.01 DF = 38

Both use Pooled Std. Dev. = 2.1277

(a) Can the null hypothesis be rejected at the 0.05 level? Why?

Yes, the P-Value of 0.001 is much less than 0.05.

(b) Is this a one-sided or two-sided test?

Two-sided.

(c) If the hypothesis had been H0: µ1 - µ2 = 2 versus H1: µ1 - µ2 ≠ 2 would you reject the null
hypothesis at the 0.05 level?

Yes.

2-7

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(d) If the hypothesis had been H0: µ1 - µ2 = 2 versus H1: µ1 - µ2 < 2 would you reject the null
hypothesis at the 0.05 level? Can you answer this question without doing any additional
calculations? Why?

Yes, no additional calculations are required because the test is naturally becoming more
significant with the change from -2.33341 to -4.33341.

(e) Use the output and the t table to find a 95 percent upper confidence bound on the difference in
means?

95% upper confidence bound = -1.21.

(f) What is the P-value if the alternative hypotheses are H0: µ1 - µ2 = 2 versus H1: µ1 - µ2 ≠ 2?

P-value = 1.4E-07.

2.18. The breaking strength of a fiber is required to be at least 150 psi. Past experience has indicated
that the standard deviation of breaking strength is σ = 3 psi. A random sample of four specimens is
tested. The results are y1=145, y2=153, y3=150 and y4=147.

(a) State the hypotheses that you think should be tested in this experiment.

H0: µ = 150 H1: µ > 150

(b) Test these hypotheses using α = 0.05. What are your conclusions?

n = 4, σ = 3, y = 1/4 (145 + 153 + 150 + 147) = 148.75

y − µo 148.75 − 150 −1.25

zo = = = = −0.8333
σ 3 3
n 4 2

Since z0.05 = 1.645, do not reject.

(c) Find the P-value for the test in part (b).

From the z-table: P ≅ 1− ⎣⎡0.7967 + ( 2 3)( 0.7995 − 0.7967) ⎦⎤ = 0.2014

(d) Construct a 95 percent confidence interval on the mean breaking strength.

The 95% confidence interval is

σ σ
y − zα 2 ≤ µ ≤ y + zα 2
n n
148.75 − (1.96 )( 3 2 ) ≤ µ ≤ 148.75 + (1.96 )( 3 2 )

145.81 ≤ µ ≤ 151.69

2-8

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

2.19. The viscosity of a liquid detergent is supposed to average 800 centistokes at 25°C. A random
sample of 16 batches of detergent is collected, and the average viscosity is 812. Suppose we know that
the standard deviation of viscosity is σ = 25 centistokes.

(a) State the hypotheses that should be tested.

H0: µ = 800 H1: µ ≠ 800

(b) Test these hypotheses using α = 0.05. What are your conclusions?

y − µo 812 − 800 12
zo = = = = 1.92 Since zα/2 = z0.025 = 1.96, do not reject.
σ 25 25
n 16 4

(c) What is the P-value for the test?

(d) Find a 95 percent confidence interval on the mean.

The 95% confidence interval is

σ σ
y − zα 2 ≤ µ ≤ y + zα 2
n n
812 − (1.96)( 25 4) ≤ µ ≤ 812 + (1.96)( 25 4)
812 −12.25 ≤ µ ≤ 812 + 12.25
799.75 ≤ µ ≤ 824.25

2.20. The diameters of steel shafts produced by a certain manufacturing process should have a mean
diameter of 0.255 inches. The diameter is known to have a standard deviation of σ = 0.0001 inch. A
random sample of 10 shafts has an average diameter of 0.2545 inches.

(a) Set up the appropriate hypotheses on the mean µ.

H0: µ = 0.255 H1: µ ≠ 0.255

(b) Test these hypotheses using α = 0.05. What are your conclusions?

n = 10, σ = 0.0001, y = 0.2545

y − µo 0.2545 − 0.255
zo = = = −15.81
σ 0.0001
n 10
Since z0.025 = 1.96, reject H0.

(c) Find the P-value for this test. P = 2.6547x10-56

2-9

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(d) Construct a 95 percent confidence interval on the mean shaft diameter.

The 95% confidence interval is

σ σ
y − zα 2 ≤ µ ≤ y + zα 2
n n
⎛ 0.0001 ⎞ ⎛ 0.0001 ⎞
0.2545 − (1.96 ) ⎜ ⎟ ≤ µ ≤ 0.2545 + (1.96 ) ⎜ ⎟
⎝ 10 ⎠ ⎝ 10 ⎠

0.254438 ≤ µ ≤ 0.254562

2.21. A normally distributed random variable has an unknown mean µ and a known variance σ2 = 9.
Find the sample size required to construct a 95 percent confidence interval on the mean that has total
length of 1.0.

Since y ∼ N(µ,9), a 95% two-sided confidence interval on µ is

If the total interval is to have width 1.0, then the half-interval is 0.5. Since zα/2 = z0.025 = 1.96,
⎛ 3 ⎞
(1.96 ) ⎜ ⎟ = 0.5 (1.96 )
⎝ n⎠
⎛ 3 ⎞
n = (1.96 ) ⎜ ⎟ = 11.76
⎝ 0.5 ⎠
2
n = (11.76 ) = 138.30 ≅ 139

2.22. The shelf life of a carbonated beverage is of interest. Ten bottles are randomly selected and
tested, and the following results are obtained:

Days
108 138
124 163
124 159
106 134
115 139

(a) We would like to demonstrate that the mean shelf life exceeds 120 days. Set up appropriate
hypotheses for investigating this claim.

H0: µ = 120 H1: µ > 120

(b) Test these hypotheses using α = 0.01. What are your conclusions?

y = 131
S2 = 3438 / 9 = 382
S = 382 = 19.54

y − µ0 131 − 120
t0 = = = 1.78
S n 19.54 10

since t0.01,9 = 2.821; do not reject H0

2-10

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

Minitab Output
T-Test of the Mean

Test of mu = 120.00 vs mu > 120.00

Variable N Mean StDev SE Mean T P

Shelf Life 10 131.00 19.54 6.18 1.78 0.054

T Confidence Intervals

Variable N Mean StDev SE Mean 99.0 % CI

Shelf Life 10 131.00 19.54 6.18 ( 110.91, 151.09)

(c) Find the P-value for the test in part (b). P=0.054

(d) Construct a 99 percent confidence interval on the mean shelf life.

S S
The 99% confidence interval is y − tα 2,n−1 ≤ µ ≤ y + tα 2,n−1 with α = 0.01.
n n

⎛ 19.54 ⎞ ⎛ 19.54 ⎞
131 − ( 3.250 ) ⎜ ⎟ ≤ µ ≤ 131 + ( 3.250 ) ⎜ ⎟
⎝ 10 ⎠ ⎝ 10 ⎠

110.91 ≤ µ ≤ 151.08

2.23. Consider the shelf life data in Problem 2.22. Can shelf life be described or modeled adequately
by a normal distribution? What effect would violation of this assumption have on the test procedure
you used in solving Problem 2.22?

A normal probability plot, obtained from Minitab, is shown. There is no reason to doubt the
adequacy of the normality assumption. If shelf life is not normally distributed, then the impact of
this on the t-test in problem 2.22 is not too serious unless the departure from normality is severe.

2-11

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

2.24. The time to repair an electronic instrument is a normally distributed random variable
measured in hours. The repair time for 16 such instruments chosen at random are as follows:

Hours
159 280 101 212
224 379 179 264
222 362 168 250
149 260 485 170

(a) You wish to know if the mean repair time exceeds 225 hours. Set up appropriate hypotheses
for investigating this issue.

H0: µ = 225 H1: µ > 225

(b) Test the hypotheses you formulated in part (a). What are your conclusions? Use α = 0.05.

y = 241.50
S2 =146202 / (16 - 1) = 9746.80

S = 9746.8 = 98.73

y − µo 241.50 − 225
to = = = 0.67
S 98.73
n 16

since t0.05,15 = 1.753; do not reject H0

T-Test of the Mean

Test of mu = 225.0 vs mu > 225.0

Variable N Mean StDev SE Mean T P

Hours 16 241.5 98.7 24.7 0.67 0.26

T Confidence Intervals

Variable N Mean StDev SE Mean 95.0 % CI

Hours 16 241.5 98.7 24.7 ( 188.9, 294.1)

(c) Find the P-value for this test. P=0.26

(d) Construct a 95 percent confidence interval on mean repair time.

S S
The 95% confidence interval is y − tα 2,n−1 ≤ µ ≤ y + tα 2,n−1
n n

⎛ 98.73 ⎞ ⎛ 98.73 ⎞
241.50 − ( 2.131) ⎜ ⎟ ≤ µ ≤ 241.50 + ( 2.131) ⎜ ⎟
⎝ 16 ⎠ ⎝ 16 ⎠

188.9 ≤ µ ≤ 294.1

2-12

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

2.25. Reconsider the repair time data in Problem 2.24. Can repair time, in your opinion, be
adequately modeled by a normal distribution?

The normal probability plot below does not reveal any serious problem with the normality
assumption.

2.26. Two machines are used for filling plastic bottles with a net volume of 16.0 ounces. The filling
processes can be assumed to be normal, with standard deviation of σ1 = 0.015 and σ2 = 0.018. The
quality engineering department suspects that both machines fill to the same net volume, whether or
not this volume is 16.0 ounces. An experiment is performed by taking a random sample from the
output of each machine.

Machine 1 Machine 2
16.03 16.01 16.02 16.03
16.04 15.96 15.97 16.04
16.05 15.98 15.96 16.02
16.05 16.02 16.01 16.01
16.02 15.99 15.99 16.00

(a) State the hypotheses that should be tested in this experiment.

H0: µ1 = µ2 H1: µ1 ≠ µ2

(b) Test these hypotheses using α=0.05. What are your conclusions?

y1 = 16.015 y2 = 16.005
σ 1 = 0.015 σ 2 = 0.018
n1 = 10 n2 = 10

2-13

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

y1 − y2 16.015 − 16.018
zo = = = 1.35
2 2
σ 1 σ 2 0.0152 0.0182
+ +
n1 n2 10 10

z0.025 = 1.96; do not reject

(c) What is the P-value for the test? P = 0.1770

(d) Find a 95 percent confidence interval on the difference in the mean fill volume for the two
machines.

The 95% confidence interval is

σ 12 σ 22 σ 12 σ 22
y1 − y2 − zα 2 + ≤ µ1 − µ2 ≤ y1 − y2 + zα 2 +
n1 n2 n1 n2
0.0152 0.0182 0.0152 0.0182
(16.015 − 16.005) − (1.96) + ≤ µ1 − µ2 ≤ (16.015 − 16.005) + (1.96) +
10 10 10 10
−0.0045 ≤ µ1 − µ2 ≤ 0.0245

2.27. Two types of plastic are suitable for use by an electronic calculator manufacturer. The
breaking strength of this plastic is important. It is known that σ1 = σ2 = 1.0 psi. From random
samples of n1 = 10 and n2 = 12 we obtain y 1 = 162.5 and y 2 = 155.0. The company will not adopt
plastic 1 unless its breaking strength exceeds that of plastic 2 by at least 10 psi. Based on the sample
information, should they use plastic 1? In answering this questions, set up and test appropriate
hypotheses using α = 0.01. Construct a 99 percent confidence interval on the true mean difference in
breaking strength.

H0: µ1 - µ2 =10 H1: µ1 - µ2 >10

y1 = 162.5 y2 = 155.0
σ1 = 1 σ2 =1
n1 = 10 n2 = 10

y1 − y2 − 10 162.5 − 155.0 − 10
zo = = = −5.84
2 2
σ 1 σ 2 12 12
+ +
n1 n2 10 12

z0.01 = 2.325; do not reject

The 99 percent confidence interval is

σ 12 σ 22 σ 12 σ 22
y1 − y2 − zα 2 + ≤ µ1 − µ2 ≤ y1 − y2 + zα 2 +
n1 n2 n1 n2
2 2
1 1 12 12
(162.5 − 155.0 ) − (2.575) + ≤ µ1 − µ 2 ≤ (162.5 − 155.0 ) + (2.575) +
10 12 10 12

6.40 ≤ µ1 − µ2 ≤ 8.60

2-14

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

2.28. The following are the burning times (in minutes) of chemical flares of two different
formulations. The design engineers are interested in both the means and variance of the burning
times.

Type 1 Type 2
65 82 64 56
81 67 71 69
57 59 83 74
66 75 59 82
82 70 65 79

(a) Test the hypotheses that the two variances are equal. Use α = 0.05.

H0 : σ12 = σ 22
H1 : σ12 ≠ σ 22
Do not reject.

(b) Using the results of (a), test the hypotheses that the mean burning times are equal. Use α =
0.05. What is the P-value for this test?

Do not reject.

From the computer output, t=0.05; do not reject. Also from the computer output P=0.96

Minitab Output
Two Sample T-Test and Confidence Interval

Two sample T for Type 1 vs Type 2

N Mean StDev SE Mean

Type 1 10 70.40 9.26 2.9
Type 2 10 70.20 9.37 3.0

95% CI for mu Type 1 - mu Type 2: ( -8.6, 9.0)

T-Test mu Type 1 = mu Type 2 (vs not =): T = 0.05 P = 0.96 DF = 18
Both use Pooled StDev = 9.32

(c) Discuss the role of the normality assumption in this problem. Check the assumption of
normality for both types of flares.

The assumption of normality is required in the theoretical development of the t-test. However,
moderate departure from normality has little impact on the performance of the t-test. The normality
assumption is more important for the test on the equality of the two variances. An indication of
nonnormality would be of concern here. The normal probability plots shown below indicate that
burning time for both formulations follow the normal distribution.

2-15

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

2.29. An article in Solid State Technology, "Orthogonal Design of Process Optimization and Its
Application to Plasma Etching" by G.Z. Yin and D.W. Jillie (May, 1987) describes an experiment to
determine the effect of C2F6 flow rate on the uniformity of the etch on a silicon wafer used in
integrated circuit manufacturing. Data for two flow rates are as follows:

C2F6 Uniformity Observation

(SCCM) 1 2 3 4 5 6
125 2.7 4.6 2.6 3.0 3.2 3.8
200 4.6 3.4 2.9 3.5 4.1 5.1

2-16

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(a) Does the C2F6 flow rate affect average etch uniformity? Use α = 0.05.

No, C2F6 flow rate does not affect average etch uniformity.

Minitab Output
Two Sample T-Test and Confidence Interval

Two sample T for Uniformity

Flow Rat N Mean StDev SE Mean

125 6 3.317 0.760 0.31
200 6 3.933 0.821 0.34

95% CI for mu (125) - mu (200): ( -1.63, 0.40)

T-Test mu (125) = mu (200) (vs not =): T = -1.35 P = 0.21 DF = 10
Both use Pooled StDev = 0.791

(b) What is the P-value for the test in part (a)? From the Minitab output, P=0.21

(c) Does the C2F6 flow rate affect the wafer-to-wafer variability in etch uniformity? Use α = 0.05.

H 0 : σ 12 = σ 22
H1 : σ 12 ≠ σ 22
F0.025,5,5 = 7.15
F0.975,5,5 = 0.14
0.5776
F0 = = 0.86
0.6724
Do not reject; C2F6 flow rate does not affect wafer-to-wafer variability.

(d) Draw box plots to assist in the interpretation of the data from this experiment.

The box plots shown below indicate that there is little difference in uniformity at the two gas flow
rates. Any observed difference is not statistically significant. See the t-test in part (a).

2-17

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

2.30. A new filtering device is installed in a chemical unit. Before its installation, a random sample
yielded the following information about the percentage of impurity: y 1 = 12.5, S 12 =101.17, and
n1 = 8. After installation, a random sample yielded y 2 = 10.2, S 22 = 94.73, n2 = 9.

(a) Can you conclude that the two variances are equal? Use α = 0.05.

H0 : σ12 = σ 22
H1 : σ12 ≠ σ 22
F0.025,7,8 = 4.53
S12 101.17
F0 = = = 1.07
S22 94.73
Do not reject. Assume that the variances are equal.

(b) Has the filtering device reduced the percentage of impurity significantly? Use α = 0.05.

H0 : µ1 = µ2
H1 : µ1 > µ2
( n1 − 1)S12 + ( n2 − 1)S22 (8 − 1)(101.17) + (9 − 1)(94.73)
S p2 = = = 97.74
n1 + n2 − 2 8+9−2
S p = 9.89
y1 − y2 12.5 − 10.2
t0 = = = 0.479
1 1 1 1
Sp + 9.89 +
n1 n2 8 9
t0.05,15 = 1.753

Do not reject. There is no evidence to indicate that the new filtering device has affected the mean.

2.31. Photoresist is a light-sensitive material applied to semiconductor wafers so that the circuit
pattern can be imaged on to the wafer. After application, the coated wafers are baked to remove the
solvent in the photoresist mixture and to harden the resist. Here are measurements of photoresist
thickness (in kÅ) for eight wafers baked at two different temperatures. Assume that all of the runs
were made in random order.

95 ºC 100 ºC
11.176 5.623
7.089 6.748
8.097 7.461
11.739 7.015
11.291 8.133
10.759 7.418
6.467 3.772
8.315 8.963

2-18

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(a) Is there evidence to support the claim that the higher baking temperature results in wafers with
a lower mean photoresist thickness? Use α = 0.05.

H0 : µ1 = µ2
H1 : µ1 > µ2
( n1 − 1)S12 + ( n2 − 1)S22 (8 − 1)(4.41) + (8 − 1)(2.54)
S p2 = = = 3.48
n1 + n2 − 2 8+8−2
S p = 1.86
y1 − y2 9.37 − 6.89
t0 = = = 2.65
1 1 1 1
Sp + 1.86 +
n1 n2 8 8
t0.05,14 = 1.761

Since t0.05,14 = 1.761, reject H0. There appears to be a lower mean thickness at the higher temperature.
This is also seen in the computer output.

Minitab Output
Two-Sample T-Test and CI: Thickness, Temp

Two-sample T for Thick@95 vs Thick@100

N Mean StDev SE Mean

Thick@95 8 9.37 2.10 0.74
Thick@10 8 6.89 1.60 0.56

Difference = mu Thick@95 - mu Thick@100

Estimate for difference: 2.475
95% lower bound for difference: 0.833
T-Test of difference = 0 (vs >): T-Value = 2.65 P-Value = 0.009 DF = 14
Both use Pooled StDev = 1.86

(b) What is the P-value for the test conducted in part (a)? P = 0.009

(c) Find a 95% confidence interval on the difference in means. Provide a practical interpretation
of this interval.

From the computer output the 95% lower confidence bound is 0.833 ≤ µ1 − µ2 . This lower
confidence bound is greater than 0; therefore, there is a difference in the two temperatures on the
thickness of the photoresist.

2-19

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(d) Draw dot diagrams to assist in interpreting the results from this experiment.

(e) Check the assumption of normality of the photoresist thickness.

2-20

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

There are no significant deviations from the normality assumptions.

(f) Find the power of this test for detecting an actual difference in means of 2.5 kÅ.

Minitab Output
Power and Sample Size

2-Sample t Test

Testing mean 1 = mean 2 (versus not =)

Calculating power for mean 1 = mean 2 + difference
Alpha = 0.05 Sigma = 1.86

Sample
Difference Size Power
2.5 8 0.7056

(g) What sample size would be necessary to detect an actual difference in means of 1.5 kÅ with a
power of at least 0.9?.

Minitab Output
Power and Sample Size

2-Sample t Test

Testing mean 1 = mean 2 (versus not =)

Calculating power for mean 1 = mean 2 + difference
Alpha = 0.05 Sigma = 1.86

Sample Target Actual

Difference Size Power Power
1.5 34 0.9000 0.9060

This result makes intuitive sense. More samples are needed to detect a smaller difference.

2-21

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

2.32. Front housings for cell phones are manufactured in an injection molding process. The time the
part is allowed to cool in the mold before removal is thought to influence the occurrence of a
particularly troublesome cosmetic defect, flow lines, in the finished housing. After manufacturing,
the housings are inspected visually and assigned a score between 1 and 10 based on their appearance,
with 10 corresponding to a perfect part and 1 corresponding to a completely defective part. An
experiment was conducted using two cool-down times, 10 seconds and 20 seconds, and 20 housings
were evaluated at each level of cool-down time. All 40 observations in this experiment were run in
random order. The data are shown below.

10 Seconds 20 Seconds
1 3 7 6
2 6 8 9
1 5 5 5
3 3 9 7
5 2 5 4
1 1 8 6
5 6 6 8
2 8 4 5
3 2 6 8
5 3 7 7

(a) Is there evidence to support the claim that the longer cool-down time results in fewer
appearance defects? Use α = 0.05.

From the analysis shown below, there is evidence that the longer cool-down time results in fewer
appearance defects.

Minitab Output
Two-Sample T-Test and CI: 10 seconds, 20 seconds

Two-sample T for 10 seconds vs 20 seconds

N Mean StDev SE Mean

10 secon 20 3.35 2.01 0.45
20 secon 20 6.50 1.54 0.34

Difference = mu 10 seconds - mu 20 seconds

Estimate for difference: -3.150
95% upper bound for difference: -2.196
T-Test of difference = 0 (vs <): T-Value = -5.57 P-Value = 0.000 DF = 38
Both use Pooled StDev = 1.79

(b) What is the P-value for the test conducted in part (a)? From the Minitab output, P = 0.000

(c) Find a 95% confidence interval on the difference in means. Provide a practical interpretation
of this interval.

From the Minitab output, µ1 − µ 2 ≤ −2.196 . This lower confidence bound is less than 0. The two
samples are different. The 20 second cooling time gives a cosmetically better housing.

2-22

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(d) Draw dot diagrams to assist in interpreting the results from this experiment.

(e) Check the assumption of normality for the data from this experiment.

2-23

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

There are no significant departures from normality.

2.33. Twenty observations on etch uniformity on silicon wafers are taken during a qualification
experiment for a plasma etcher. The data are as follows:

Etch Uniformity
5.34 6.65 4.76 5.98 7.25
6.00 7.55 5.54 5.62 6.21
5.97 7.35 5.44 4.39 4.98
5.25 6.35 4.61 6.00 5.32

(a) Construct a 95 percent confidence interval estimate of σ 2.

( n − 1) S 2 ≤ σ 2 ≤ ( n − 1) S 2
χα2 ,n −1
χ(12 −α ),n −1
2 2
2 2
( 20 − 1)( 0.88907 ) ≤σ ≤
( 20 − 1)( 0.88907 )
2

32.852 8.907
0.457 ≤ σ 2 ≤ 1.686
(b) Test the hypothesis that σ 2 = 1.0. Use α = 0.05. What are your conclusions?

Do not reject. There is no evidence to indicate that σ2 ≠1

(c) Discuss the normality assumption and its role in this problem.

The normality assumption is much more important when analyzing variances then when analyzing
means. A moderate departure from normality could cause problems with both statistical tests and
confidence intervals. Specifically, it will cause the reported significance levels to be incorrect.

2-24

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(d) Check normality by constructing a normal probability plot. What are your conclusions?

The normal probability plot indicates that there is not a serious problem with the normality
assumption.

2.34. The diameter of a ball bearing was measured by 12 inspectors, each using two different kinds
of calipers. The results were:

Inspector Caliper 1 Caliper 2 Difference Difference^2

1 0.265 0.264 0.001 0.000001
2 0.265 0.265 0 0
3 0.266 0.264 0.002 0.000004
4 0.267 0.266 0.001 0.000001
5 0.267 0.267 0 0
6 0.265 0.268 -0.003 0.000009
7 0.267 0.264 0.003 0.000009
8 0.267 0.265 0.002 0.000004
9 0.265 0.265 0 0
10 0.268 0.267 0.001 0.000001
11 0.268 0.268 0 0
12 0.265 0.269 -0.004 0.000016
∑ = 0.003 ∑ = 0.000045

2-25

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(a) Is there a significant difference between the means of the population of measurements
represented by the two samples? Use α = 0.05.

H 0 : µ1 = µ 2 H 0 : µd = 0
or equivalently
H 1 : µ1 ≠ µ 2 H1 : µ d ≠ 0

Minitab Output
Paired T-Test and Confidence Interval

Paired T for Caliper 1 - Caliper 2

N Mean StDev SE Mean

Caliper 12 0.266250 0.001215 0.000351
Caliper 12 0.266000 0.001758 0.000508
Difference 12 0.000250 0.002006 0.000579

95% CI for mean difference: (-0.001024, 0.001524)

T-Test of mean difference = 0 (vs not = 0): T-Value = 0.43 P-Value = 0.674

(b) Find the P-value for the test in part (a). P=0.674

(c) Construct a 95 percent confidence interval on the difference in the mean diameter
measurements for the two types of calipers.

Sd S
d − tα ≤ µ D ( = µ1 − µ 2 ) ≤ d + tα ,n −1 d
, n −1
2 n 2 n
0.002 0.002
0.00025 − 2.201 ≤ µ d ≤ 0.00025 + 2.201
12 12
−0.00102 ≤ µ d ≤ 0.00152

2.35. An article in the journal of Neurology (1998, Vol. 50, pp.1246-1252) observed that the
monozygotic twins share numerous physical, psychological and pathological traits. The investigators
measured an intelligence score of 10 pairs of twins. The data are obtained as follows:

Pair Birth Order: 1 Birth Order: 2

1 6.08 5.73
2 6.22 5.80
3 7.99 8.42
4 7.44 6.84
5 6.48 6.43
6 7.99 8.76
7 6.32 6.32
8 7.60 7.62
9 6.03 6.59
10 7.52 7.67

2-26

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(a) Is the assumption that the difference in score is normally distributed reasonable?

Minitab Output

By plotting the differences, the output shows that the Anderson-Darling Normality Test shows a P-
Value of 0.860. The data is assumed to be normal.

(b) Find a 95% confidence interval on the difference in the mean score. Is there any evidence that
mean score depends on birth order?

The 95% confidence interval on the difference in mean score is (-0.366415, 0.264415) contains the
value of zero. There is no difference in birth order.

(c) Test an appropriate set of hypothesis indicating that the mean score does not depend on birth
order.

H 0 : µ1 = µ 2 H 0 : µd = 0
or equivalently
H 1 : µ1 ≠ µ 2 H1 : µ d ≠ 0

Minitab Output
Paired T for Birth Order: 1 - Birth Order: 2

N Mean StDev SE Mean

Birth Order: 1 10 6.967 0.810 0.256
Birth Order: 2 10 7.018 1.053 0.333
Difference 10 -0.051 0.441 0.139

95% CI for mean difference: (-0.366, 0.264)

T-Test of mean difference = 0 (vs not = 0): T-Value = -0.37 P-Value = 0.723

Do not reject. The P-value is 0.723.

2-27

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

2.36. An article in the Journal of Strain Analysis (vol.18, no. 2, 1983) compares several procedures
for predicting the shear strength for steel plate girders. Data for nine girders in the form of the ratio
of predicted to observed load for two of these procedures, the Karlsruhe and Lehigh methods, are as
follows:

Girder Karlsruhe Lehigh Method Difference Difference^2

Method
S1/1 1.186 1.061 0.125 0.015625
S2/1 1.151 0.992 0.159 0.025281
S3/1 1.322 1.063 0.259 0.067081
S4/1 1.339 1.062 0.277 0.076729
S5/1 1.200 1.065 0.135 0.018225
S2/1 1.402 1.178 0.224 0.050176
S2/2 1.365 1.037 0.328 0.107584
S2/3 1.537 1.086 0.451 0.203401
S2/4 1.559 1.052 0.507 0.257049
Sum = 2.465 0.821151
Average = 0.274

(a) Is there any evidence to support a claim that there is a difference in mean performance between
the two methods? Use α = 0.05.

µ =µ µ =
or equivalently
µ ≠µ µ ≠

1 n 1
d = ∑
n i =1
d i = ( 2.465 ) = 0.274
9
1
⎡ n 2 1 ⎛ n ⎞2 ⎤
2
1

⎢ ∑ di − ⎜ ∑ di ⎟ ⎥ ⎡ 1 2 ⎤ 2

⎢ 0.821151 − (2.465) ⎥
n ⎝ i =1 ⎠ ⎥
sd = ⎢ i =1 =⎢ 9
⎥ = 0.135
⎢ n −1 ⎥ 9 − 1
⎢ ⎥ ⎢ ⎥
⎣ ⎦
⎣⎢ ⎦⎥

d 0.274
t0 = = = 6.08
Sd 0.135
n 9
tα 2 , n −1 = t0.025,8 = 2.306 , reject the null hypothesis.
Minitab Output
Paired T-Test and Confidence Interval

Paired T for Karlsruhe - Lehigh

N Mean StDev SE Mean

Karlsruh 9 1.3401 0.1460 0.0487
Lehigh 9 1.0662 0.0494 0.0165
Difference 9 0.2739 0.1351 0.0450

95% CI for mean difference: (0.1700, 0.3777)

T-Test of mean difference = 0 (vs not = 0): T-Value = 6.08 P-Value = 0.000

2-28

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(b) What is the P-value for the test in part (a)?

P=0.0002

(c) Construct a 95 percent confidence interval for the difference in mean predicted to observed
load.

Sd S
d − tα ≤ µ d ≤ d + tα ,n −1 d
, n −1
2 n 2 n
0.135 0.135
0.274 − 2.306 ≤ µ d ≤ 0.274 + 2.306
9 9
0.17023 ≤ µ d ≤ 0.37777

(d) Investigate the normality assumption for both samples.

The normal probability plots of the observations for each method follow. There are no serious
concerns with the normality assumption, but there is an indication of a possible outlier (1.178) in the
Lehigh method data.

2-29

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(a) Investigate the normality assumption for the difference in ratios for the two methods.

There is no issue with normality in the difference of ratios of the two methods.

(b) Discuss the role of the normality assumption in the paired t-test.

As in any t-test, the assumption of normality is of only moderate importance. In the paired t-test, the
assumption of normality applies to the distribution of the differences. That is, the individual sample
measurements do not have to be normally distributed, only their difference.

2.37. The deflection temperature under load for two different formulations of ABS plastic pipe is
being studied. Two samples of 12 observations each are prepared using each formulation, and the
deflection temperatures (in °F) are reported below:

Formulation Formulation
1 2
206 193 192 177 176 198
188 207 210 197 185 188
205 185 194 206 200 189
187 189 178 201 197 203

2-30

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(a) Construct normal probability plots for both samples. Do these plots support assumptions of
normality and equal variance for both samples?

(b) Do the data support the claim that the mean deflection temperature under load for formulation
1 exceeds that of formulation 2? Use α = 0.05.

No, formulation 1 does not exceed formulation 2 per the Minitab output below.

Minitab Output
Two Sample T-Test and Confidence Interval

N Mean StDev SE Mean

Form 1 12 194.5 10.2 2.9
Form 2 12 193.08 9.95 2.9

Difference = mu Form 1 - mu Form 2

Estimate for difference: 1.42
95% lower bound for difference: -5.64
T-Test of difference = 0 (vs >): T-Value = 0.34 P-Value = 0.367 DF = 22
Both use Pooled StDev = 10.1

2-31

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(c) What is the P-value for the test in part (a)?

P = 0.367

2.38. Refer to the data in problem 2.37. Do the data support a claim that the mean deflection
temperature under load for formulation 1 exceeds that of formulation 2 by at least 3 °F?

No, formulation 1 does not exceed formulation 2 by at least 3 °F.

Minitab Output
Two-Sample T-Test and CI: Form1, Form2

Two-sample T for Form 1 vs Form 2

N Mean StDev SE Mean

Form 1 12 194.5 10.2 2.9
Form 2 12 193.08 9.95 2.9

Difference = mu Form 1 - mu Form 2

Estimate for difference: 1.42
95% lower bound for difference: -5.64
T-Test of difference = 3 (vs >): T-Value = -0.39 P-Value = 0.648 DF = 22
Both use Pooled StDev = 10.1

2.39. In semiconductor manufacturing, wet chemical etching is often used to remove silicon from the
backs of wafers prior to metalization. The etch rate is an important characteristic of this process.
Two different etching solutions are being evaluated. Eight randomly selected wafers have been
etched in each solution and the observed etch rates (in mils/min) are shown below:

Solution 1 Solution 2
9.9 10.6 10.2 10.6
9.4 10.3 10.0 10.2
10.0 9.3 10.7 10.4
10.3 9.8 10.5 10.3

(a) Do the data indicate that the claim that both solutions have the same mean etch rate is valid?
Use α = 0.05 and assume equal variances.

No, the solutions do not have the same mean etch rate. See the Minitab output below.

Minitab Output
Two Sample T-Test and Confidence Interval

Two-sample T for Solution 1 vs Solution 2

N Mean StDev SE Mean

Solution 8 9.950 0.450 0.16
Solution 8 10.363 0.233 0.082

Difference = mu Solution 1 - mu Solution 2

Estimate for difference: -0.413
95% CI for difference: (-0.797, -0.028)
T-Test of difference = 0 (vs not =): T-Value = -2.30 P-Value = 0.037 DF = 14
Both use Pooled StDev = 0.358

2-32

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

(b) Find a 95% confidence interval on the difference in mean etch rate.

From the Minitab output, -0.797 to –0.028.

(c) Use normal probability plots to investigate the adequacy of the assumptions of normality and
equal variances.

Both the normality and equality of variance assumptions are valid.

2.40. Two popular pain medications are being compared on the basis of the speed of absorption by
2
the body. Specifically, tablet 1 is claimed to be absorbed twice as fast as tablet 2. Assume that σ 1
2
and σ 2 are known. Develop a test statistic for

H0: 2µ1 = µ2
H1: 2µ1 ≠ µ2

2-33

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

⎛ 4σ 2 σ 2 ⎞
2 y1 − y2 ~ N ⎜ 2µ1 − µ2 , 1 + 2 ⎟, assuming that the data is normally
⎝ n1 n2 ⎠
distributed.
2 y1 − y2
The test statistic is: zo = , reject if z o > zα
4σ12 σ 22 2

+
n1 n2

2.41. Continuation of Problem 2.40. An article in Nature (1972, pp.225-226) reported on the levels of
monoamine oxidase in blood platelets for a sample of 43 schizophrenic patients resulting in = 2.69
and s1 = 2.30 while for a sample of 45 normal patients the results were = 6.35 and s2 = 4.03. The
units are nm/mg protein/h. Use the results of the previous problem to test the claim that the mean
monoamine oxidase level for normal patients is at least twice the mean level for schizophrenic
patients. Assume that the sample sizes are large enough to use the sample standard deviations as the
true parameter values.

2 y1 − y2 2(2.69) − 6.35 −0.97

zo = = = = −1.05
4σ 2
1 σ 2
2 4(2.30) 4.03 2 2 .92357
+ +
n1 n2 43 45
z0 = -1.05; using α =0.05, , do not reject.

2.42. Suppose we are testing

H0: µ1 = µ2
H1: µ1 ≠ µ2

2 2
where σ 1 > σ 2 are known. Our sampling resources are constrained such that n1 + n2 = N. Show
that an allocation of the observation n1 and n2 to the two samples leads to the most powerful test in
the ratio n1 / n2 = σ1 / σ2.

The most powerful test is attained by the n1 and n2 that maximize zo for given y1 − y2.
y1 − y2
Thus, we chose n1 and n2 to max zo = , subject to n1 + n2 = N.
σ12 σ 22
+
n1 n2
2 2 2
σ σ σ σ 22
This is equivalent to min L = 1
+ 2
= 1
+ , subject to n1 + n2 = N.
n1 n2 n1 N − n1
dL −σ12 σ 22
Now = + = 0 , implies that n1 / n2 = σ1 / σ2.
dn1 n12 ( N − n1 )2
Thus n1 and n2 are assigned proportionally to the ratio of the standard deviations. This has
intuitive appeal, as it allocates more observations to the population with the greatest variability.

2-34

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

2.43. Continuation of Problem 2.42. Suppose that we want to construct a 95% two-sided confidence
interval on the difference in two means where the two sample standard deviations are known to be σ1
= 4 and σ2 = 8. The total sample size is restricted to N = 30. What is the length of the 95% CI if the
sample sizes used by the experimenter are n1 = n2 = 15? How much shorter would the 95% CI have
been if the experiment had used the optimal sample size calculation?

The 95% confidence interval for n1 = n2 = 15 is

σ 12 σ 22 σ 12 σ 22
( y1 − y2 ) − zα 2 + ≤ µ1 − µ2 ≤ ( y1 − y2 ) + zα 2 +
n1 n2 n1 n2
4 2 82 4 2 82
( y1 − y2 ) − zα 2 + ≤ µ1 − µ2 ≤ ( y1 − y2 ) + zα 2 +
15 15 15 15
( y1 − y2 ) − zα 2 (2.31) ≤ µ1 − µ2 ≤ ( y1 − y2 ) + zα 2 (2.31)

The 95% confidence interval for the proportions is,

Therefore n2 = 20 and n1 = 10

σ 12 σ 22 σ 12 σ 22
( y1 − y2 ) − zα 2 + ≤ µ1 − µ2 ≤ ( y1 − y2 ) + zα 2 +
n1 n2 n1 n2
4 2 82 4 2 82
( y1 − y2 ) − zα 2 + ≤ µ1 − µ2 ≤ ( y1 − y2 ) + zα 2 +
10 20 10 20
( y1 − y2 ) − zα 2 (2.19) ≤ µ1 − µ2 ≤ ( y1 − y2 ) + zα 2 (2.19)

The confidence interval decreases from a multiple of 2.31 to a multiple of 2.19.

2.44. Develop Equation 2.46 for a 100(1 - α) percent confidence interval for the variance of a normal
distribution.

SS ⎧ SS ⎫
2
~ χ n2−1 . Thus, P ⎨ χ 2 α ≤ 2 ≤ χ α2 ,n−1 ⎬ = 1 − α . Therefore,
σ ⎩ 2 1− ,n −1
σ 2 ⎭
⎧ ⎫
⎪ SS SS ⎪
P ⎨ 2 ≤ σ 2 ≤ 2 ⎬ = 1−α ,
⎪⎩ χ α2 ,n−1 χ 1−α ,n−1 ⎪
2 ⎭
⎡ ⎤
SS SS ⎥
so ⎢ , is the 100(1 - α)% confidence interval on σ 2.
⎢ χ α2 ,n−1 χ12−α ,n−1 ⎥
⎣ 2 2 ⎦

2-35

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

2 2
2.45. Develop Equation 2.50 for a 100(1 - α) percent confidence interval for the ratio σ 1 / σ 2 ,
2 2
where σ 1 and σ 2 are the variances of two normal distributions.

S 22 σ 22
~ Fn2 −1, n1 −1
S12 σ 12
⎧ S22 σ 22 ⎫
P ⎨F1−α 2,n2 −1,n1 −1 ≤ 2 2 ≤ Fα ⎬ = 1 − α or
,n −1,n1−1
⎩ S1 σ1 2 2
⎭
⎧ S12 σ12 S12 ⎫
P ⎨ 2 F1−α 2,n2 −1,n1 −1 ≤ 2 ≤ 2 Fα ⎬ = 1−α
⎩ S2 σ 2 S2 2,n2−1,n1−1 ⎭

2.46. Develop an equation for finding a 100(1 - α) percent confidence interval on the difference in
2 2
the means of two normal distributions where σ 1 ≠ σ 2 . Apply your equation to the portland cement
experiment data, and find a 95% confidence interval.

( y1 − y2 ) − ( µ1 − µ2 ) ~ t
2 ,υ
α
S12 S22
+
n1 n2
S12 S22 S2 S2
tα 2,υ + ≤ ( y1 − y2 ) − ( µ1 − µ2 ) ≤ tα 2,υ 1 + 2
n1 n2 n1 n2
S12 S22 S2 S2
( y1 − y2 ) − t α
2 ,υ
+ ≤ ( µ1 − µ2 ) ≤ ( y1 − y2 ) + tα 2,υ 1 + 2
n1 n2 n1 n2
2
⎛ S12 S22 ⎞
⎜ + ⎟
⎝ n1 n2 ⎠
where υ = 2 2
⎛ S12 ⎞ ⎛ S22 ⎞
⎜ ⎟ ⎜ ⎟
⎝ n1 ⎠ + ⎝ n2 ⎠
n1 − 1 n2 − 1
Using the data from Table 2.1

0.100138 0.0614622
(16.764 − 17.343 ) − 2.110 + ≤ ( µ1 − µ 2 ) ≤
10 10
0.100138 0.0614622
(16.764 − 17.343 ) + 2.110 +
10 10
2
⎛ 0.100138 0.0614622 ⎞
⎜ + ⎟
10 10
where υ = ⎝ 2
⎠
2
= 17.024 ≅ 17
⎛ 0.100138 ⎞ ⎛ 0.0614622 ⎞
⎜ ⎟ ⎜ ⎟
⎝ 10 ⎠ +⎝ 10 ⎠
10 − 1 10 − 1
− 1.426 ≤ ( µ 1 − µ 2 ) ≤ − 0.889

2-36

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

2.47. Construct a data set for which the paired t-test statistic is very large, but for which the usual
two-sample or pooled t-test statistic is small. In general, describe how you created the data. Does
this give you any insight regarding how the paired t-test works?

A B delta
7.1662 8.2416 -1.0754
2.3590 2.4555 -0.0965
19.9977 21.1018 -1.1041
0.9077 2.3401 -1.4324
-15.9034 -15.0013 -0.9021
-6.0722 -5.5941 -0.4781
9.9501 10.6910 -0.7409
-1.0944 -0.1358 -0.9586
-4.6907 -3.3446 -1.3461
-6.6929 -5.9303 -0.7626

Minitab Output
Paired T-Test and Confidence Interval

Paired T for A - B
N Mean StDev SE Mean
A 10 0.59 10.06 3.18
B 10 1.48 10.11 3.20
Difference 10 -0.890 0.398 0.126

95% CI for mean difference: (-1.174, -0.605)

T-Test of mean difference = 0 (vs not = 0): T-Value = -7.07 P-Value = 0.000

Two Sample T-Test and Confidence Interval

Two-sample T for A vs B

N Mean StDev SE Mean

A 10 0.6 10.1 3.2
B 10 1.5 10.1 3.2

Difference = mu A - mu B
Estimate for difference: -0.89
95% CI for difference: (-10.37, 8.59)
T-Test of difference = 0 (vs not =): T-Value = -0.20 P-Value = 0.846 DF = 18
Both use Pooled StDev = 10.1

These two sets of data were created by making the observation for A and B moderately different
within each pair (or block), but making the observations between pairs very different. The fact that
the difference between pairs is large makes the pooled estimate of the standard deviation large and
the two-sample t-test statistic small. Therefore the fairly small difference between the means of the
two treatments that is present when they are applied to the same experimental unit cannot be
detected. Generally, if the blocks are very different, then this will occur. Blocking eliminates the
variability associated with the nuisance variable that they represent.

2.48. Consider the experiment described in problem 2.28. If the mean burning times of the two
flames differ by as much as 2 minutes, find the power of the test. What sample size would be required
to detect an actual difference in mean burning time of 1 minute with a power of at least 0.90?

2-37

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

From the Minitab output below, the power is 0.0740. This answer was obtained by using the pooled
estimate of σ from Problem 2-11, Sp = 9.32. Because the difference in means is very small relative to
the standard deviation, the power is very low.

Minitab Output
Power and Sample Size

2-Sample t Test

Testing mean 1 = mean 2 (versus not =)

Calculating power for mean 1 = mean 2 + difference
Alpha = 0.05 Sigma = 9.32

Sample
Difference Size Power
2 10 0.0740

From the Minitab output below, the required sample size is 1827. The sample size is huge because the
difference in means is very small relative to the standard deviation.

Minitab Output
Power and Sample Size

2-Sample t Test

Testing mean 1 = mean 2 (versus not =)

Calculating power for mean 1 = mean 2 + difference
Alpha = 0.05 Sigma = 9.32

Sample Target Actual

Difference Size Power Power
1 1827 0.9000 0.9001

2.49. Reconsider the bottle filling experiment described in Problem 2.26. Rework this problem
assuming that the two population variances are unknown but equal.

Minitab Output
Two-Sample T-Test and CI: Machine 1, Machine 2

Two-sample T for Machine 1 vs Machine 2

N Mean StDev SE Mean

Machine 10 16.0150 0.0303 0.0096
Machine 10 16.0050 0.0255 0.0081

Difference = mu Machine 1 - mu Machine 2

Estimate for difference: 0.0100
95% CI for difference: (-0.0163, 0.0363)
T-Test of difference = 0 (vs not =): T-Value = 0.80 P-Value = 0.435 DF = 18
Both use Pooled StDev = 0.0280

The hypothesis test is the same: H0: µ1 = µ2 H1: µ1 ≠ µ2

The conclusions are the same as Problem 2.19, do not reject H0. There is no difference in the
machines. The P-value for this analysis is 0.435.
The confidence interval is (-0.0163, 0.0363). This interval contains 0. There is no difference in
machines.

2-38

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

2.50. Consider the data from problem 2.26. If the mean fill volume of the two machines differ by as
much as 0.25 ounces, what is the power of the test used in problem 2.19? What sample size could
result in a power of at least 0.9 if the actual difference in mean fill volume is 0.25 ounces?

The power is 1.0000 as shown in the analysis below.

Minitab Output
Power and Sample Size

2-Sample t Test

Testing mean 1 = mean 2 (versus not =)

Calculating power for mean 1 = mean 2 + difference
Alpha = 0.05 Sigma = 0.028

Sample
Difference Size Power
0.25 10 1.0000

The required sample size is 2 as shown below.

Minitab Output
Power and Sample Size

2-Sample t Test

Testing mean 1 = mean 2 (versus not =)

Calculating power for mean 1 = mean 2 + difference
Alpha = 0.05 Sigma = 0.028

Sample Target Actual

Difference Size Power Power
0.25 2 0.9000 0.9805

2.51. An experiment has been performed with a factor that has only two levels. Samples of size n1 =
n2 = 12 have been taken and the resulting sample data is as follows:

Can you conclude that there is no difference in means using α = 0.05? What are bounds on the P-
value for this test? Find a 95% confidence interval on the difference in the two means. Does the
confidence interval provide any information that is useful in interpreting the test of the hypothesis on
the difference in the two means?

Minitab Output
Two-Sample T-Test and CI

Sample N Mean StDev SE Mean

1 12 12.50 1.80 0.52
2 12 13.10 2.10 0.61

Difference = μ (1) - μ (2)

Estimate for difference: -0.600
95% CI for difference: (-2.256, 1.056)
T-Test of difference = 0 (vs ≠): T-Value = -0.75 P-Value = 0.460 DF = 22
Both use Pooled StDev = 1.9558

2-39

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

There is no difference in the means. The P-value is 0.460. If using the t-table in the book, the
bounds are 0.50 < P < 0.80. The 95% confidence interval is (-2.256, 1.056). This interval contains
zero, meaning that there is no difference in the means.

2.52. Reconsider the situation in Problem 2.51. Suppose that the two sample sizes were n1 = n2 = 5.
What difference in conclusions (if any) would you have obtained from the hypothesis test? From the
CI?

Minitab Output
Two-Sample T-Test and CI

Sample N Mean StDev SE Mean

1 5 12.50 1.80 0.80
2 5 13.10 2.10 0.94

Difference = μ (1) - μ (2)

Estimate for difference: -0.60
95% CI for difference: (-3.45, 2.25)
T-Test of difference = 0 (vs ≠): T-Value = -0.49 P-Value = 0.641 DF = 8
Both use Pooled StDev = 1.9558

There is not difference in the conclusion, there is no difference in the means. The confidence interval
for n=5 is much wider than the confidence interval for n=12.

2.53. Suppose that you are testing the hypothesis H0: µ = 50 against the usual two-sided alternative.

The data are normally distributed with known standard deviation σ = 1. The sample average
obtained in the experiment is 50.5, and it is known that if the true population mean is actually 50.5
then this has no practical significance on the problem that motivated the experiment. Find the P-
value for the t-test for the following sample sizes:

(a) n=5

(b) n = 10

(c) n = 25

(d) n = 50

(e) n = 100

(f) n = 1000

Discuss your findings. What does this tell you about relying on P-values in hypothesis testing
situations when sample sizes are large?

Minitab Output
One-Sample T

Test of μ = 50 vs ≠ 50

N Mean StDev SE Mean 95% CI T P

5 50.500 1.000 0.447 (49.258, 51.742) 1.12 0.326

N Mean StDev SE Mean 95% CI T P

10 50.500 1.000 0.316 (49.785, 51.215) 1.58 0.148

2-40

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

N Mean StDev SE Mean 95% CI T P

25 50.500 1.000 0.200 (50.087, 50.913) 2.50 0.020

N Mean StDev SE Mean 95% CI T P

50 50.500 1.000 0.141 (50.216, 50.784) 3.54 0.001

N Mean StDev SE Mean 95% CI T P

100 50.500 1.000 0.100 (50.302, 50.698) 5.00 0.000

N Mean StDev SE Mean 95% CI T P

1000 50.5000 1.0000 0.0316 (50.4379, 50.5621) 15.81 0.000

As the sample size increases, the SE Mean decreases and the 95% confidence interval gets smaller,
driving the P-value smaller. For this problem, the test becomes statistically significant between
sample sizes of 10 and 25. Having more data in this case just means that you are proving that there is
a statistical difference, not a practical difference.

2.54. Consider the situation in Problem 2.53. Calculate the 95% confidence interval on the mean for
each of the sample sizes given. How does the length of the confidence interval change with sample
size?

The 95% confidence interval decreases as sample sizes increases.

2.55. Is the assumption of sampling from a normal distribution critical in the application of the t-
test? Justify your Answer.

The normal distribution is an assumption to develop the formal test procedure. Moderate departures
from normality do not seriously affect the results of a t-test. The t-test is a good approximation of a
randomized design and can be without much concern for normality. A quick procedure for checking
normality is the normal probability plot.

2.56. Why is the random sampling assumption important in statistical inference? Suppose that you
had to select a random sample of 100 items from a production line. How would you propose to do
this?

When choosing a sampling strategy, you want the samples to be independent of each other, and
representative of the population. In high speed production, samples that are consecutive are not
typically independent. Sample n+1 is very much like sample n.

Should you take into account factors such as the production rate, or whether the line operates
continuously or only intermittently?

A sampling strategy should include the speed of production, the number of parts made per unit time,
time of day, and batch size.

2.57. An experiment has been performed with a factor that has only two levels. Sample of size n1 = n2
= 10 have been taken and the resulting sample data is as follows:

It seems likely that the two population variances are not the same. Can you conclude that there is no
difference in means using α = 0.05? What are bounds on the P-value for this test? Find a 95%

2-41

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

confidence interval on the difference in the two means. Does the confidence interval provide any
information that is useful in interpreting the test of the hypothesis on the difference in the two means?

Minitab Output
Test and CI for Two Variances

Method

Null hypothesis σ(First) / σ(Second) = 1

Alternative hypothesis σ(First) / σ(Second) ≠ 1
Significance level α = 0.05

F method was used. This method is accurate for normal data only.

Statistics
95% CI for
Sample N StDev Variance StDevs
First 10 1.500 2.250 (1.032, 2.738)
Second 10 4.100 16.810 (2.820, 7.485)

Ratio of standard deviations = 0.366

Ratio of variances = 0.134

95% Confidence Intervals

CI for
CI for StDev Variance
Method Ratio Ratio
F (0.182, 0.734) (0.033, 0.539)

Tests
Test
Method DF1 DF2 Statistic P-Value
F 9 9 0.13 0.006

Two-Sample T-Test and CI

Sample N Mean StDev SE Mean
1 10 10.70 1.50 0.47
2 10 15.10 4.10 1.3

Difference = μ (1) - μ (2)

Estimate for difference: -4.40
95% CI for difference: (-7.44, -1.36)
T-Test of difference = 0 (vs ≠): T-Value = -3.19 P-Value = 0.009 DF = 11

The variances of the two samples is different. The confidence interval on the variance is (0033,
0.539). This interval does not include 1, therefore they are different. The P-value is 0.006. Using the
two-sample t-test, with unequal variances, the means are different as well. The confidence interval of
the difference of the means at 95% is (-7.44, -1.36). This interval does not include zero, therefore we
can conclude that the means are different.

2.58. Do you think that using a significance level of α = 0.05 is appropriate for all experiments? In
the early stages of research and development work is there a lot of harm in identifying a factor as
important when it really isn’t? Would that seem to justify higher levels of significance such as α = 0.10
or perhaps even α = 0.15 in some situations?

Not necessarily. With the use of P-values, one should not immediately reject the null hypothesis if the
P-value were say, 0.055. The variable would probably become statistically significant with only a
small increase in sample size. For many research and development projects, we may want to consider
an α = 0.10 or higher. In the R&D environment, typically the sample sizes are very small and we are
interested to understand if there might be an effect of the variable that we are sampling.

2-42

Downloaded by Sara ([email protected])

 lOMoARcPSD|40821539

Solutions from Montgomery, D. C. (2017) Design and Analysis of Experiments, Wiley, NY

2.59. Power calculation for hypothesis testing are relatively easy to do with modern statistical
software. What do you think “adequate power” should be for an experiment? What issues need to be
considered in answering the question?

80%, or 0.80 is a reasonable level for power. The three levers that determine power are sample size,
the standard deviation or variance of the data, and the difference or delta that you are looking for.
The easiest way to increase power is to increase sample size.

2.60. In the early stages of research and development experimentation which type of error do you
think is most important, type I or type II? Justify your answer.

Type II. If a Type II error is made, an important factor is not identified, and it may never be
considered again. A Type I error leads to a factor that is not important but thought to be active.
This will eventually be identified.

2-43

Downloaded by Sara ([email protected])