Solution – Selected Problems Chapter 3

3.1. An experimenter has conducted a single-factor experiment with four levels of the factor, and each factor level
has been replicated six times. The computed value of the F-statistic is F0 = 3.26. Find bounds on the P-value.

# levels of factor = 4
# of replications = 6
Df numerator = 4 -1 = 3
Df denominator = 24-4 = 20

Table P-value = 0.025, 0.050 Computer P-value = 0.043

3.3. A computer ANOVA output is shown below. Fill in the blanks. You may give bounds on the P-value.

One-way ANOVA
Source DF SS MS F P
Factor 3 36.15 ? ? ?
Error ? ? ?
Total 19 196.04

Completed table is:

One-way ANOVA
Source DF SS MS F P
Factor 3 36.15 12.05 1.21 0.3395
Error 16 159.89 9.99
Total 19 196.04

3.7. The tensile strength of Portland cement is being studied.Four different mixing techniques can be used
economically. A completely randomized experiment was conducted and the following data were collected.

Mixing
Technique Tensile Strength (lb/in2)
1 3129 3000 2865 2890
2 3200 3300 2975 3150
3 2800 2900 2985 3050
4 2600 2700 2600 2765

(a) Test the hypothesis that mixing techniques affect the strength of the cement. Use  = 0.05.
Response: Tensile Strengthin lb/in^2
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of Mean F
Source Squares DF Square Value Prob > F
Model 4.897E+005 3 1.632E+005 12.73 0.0005 significant
A 4.897E+005 3 1.632E+005 12.73 0.0005
Residual 1.539E+005 12 12825.69
Cor Total 6.436E+005 15

The Model F-value of 12.73 implies the model is significant. There is only
a 0.05% chance that a "Model F-Value" this large could occur due to noise.

Treatment Means (Adjusted, If Necessary)

Estimated Standard
Mean Error
1-1 2971.00 56.63
2-2 3156.25 56.63
3-3 2933.75 56.63
4-4 2666.25 56.63

Mean Standard t for H0

Treatment Difference DF Error Coeff=0 Prob > |t|
1 vs 2 -185.25 1 80.08 -2.31 0.0392
1 vs 3 37.25 1 80.08 0.47 0.6501
1 vs 4 304.75 1 80.08 3.81 0.0025
2 vs 3 222.50 1 80.08 2.78 0.0167
2 vs 4 490.00 1 80.08 6.12 < 0.0001
3 vs 4 267.50 1 80.08 3.34 0.0059

The F-value is 12.73 with a corresponding P-value of .0005. Mixing technique has an effect.

(b) Construct a graphical display as described in Section 3.5.3 to compare the mean tensile strengths for the four
mixing techniques. What are your conclusions?

MS E 12825.7
S yi .    56.625
n 4
Scaled t Distribution

(4) (3) (1) (2)

2700 2800 2900 3000 3100

Tensile Strength

Based on examination of the plot, we would conclude that 1 and 3 are the same; that  4 differs from 1 and 3 ,
that  2 differs from 1 and 3 , and that  2 and  4 are different.
(c) Use the Fisher LSD method with =0.05 to make comparisons between pairs of means.
2MS E
LSD  t 
2
,N  a n
2( 12825.7 )
LSD  t 0.025,16 4
4
LSD  2.179 6412.85  174.495

Treatment 2 vs. Treatment 4 = 3156.250 - 2666.250 = 490.000 > 174.495

Treatment 2 vs. Treatment 3 = 3156.250 - 2933.750 = 222.500 > 174.495
Treatment 2 vs. Treatment 1 = 3156.250 - 2971.000 = 185.250 > 174.495
Treatment 1 vs. Treatment 4 = 2971.000 - 2666.250 = 304.750 > 174.495
Treatment 1 vs. Treatment 3 = 2971.000 - 2933.750 = 37.250 < 174.495
Treatment 3 vs. Treatment 4 = 2933.750 - 2666.250 = 267.500 > 174.495

The Fisher LSD method is also presented in the computer output above. The results agree with the graphical
method for this experiment.

(d) Construct a normal probability plot of the residuals. What conclusion would you draw about the validity of the
normality assumption?

There is nothing unusual about the normal probability plot of residuals.

N o rm a l p lo t o f re s id ua ls

99

95
N o r m a l % p r o b a b i l i ty

90

80
70

50

30
20

10
5

-1 8 1 .2 5 -9 6 .4 3 7 5 -1 1 .6 2 5 7 3 .1 8 7 5 158

R e s id u a l

(e) Plot the residuals versus the predicted tensile strength. Comment on the plot.

There is nothing unusual about this plot.

 Residuals vs. Predicted
158

73.1875

Res iduals
-11.625

-96.4375

-181.25

2666.25 2788.75 2911.25 3033.75 3156.25

Predicted

(f) Prepare a scatter plot of the results to aid the interpretation of the results of this experiment.

Design-Expert automatically generates the scatter plot. The plot below also shows the sample average for each
treatment and the 95 percent confidence interval on the treatment mean.

One Factor Plot

3300

3119.75
Tens ile Strength

2939.51

2759.26

2
2579.01

1 2 3 4

Technique
3.10. A product developer is investigating the tensile strength of a new synthetic fiber that will be used to make
cloth for men’s shirts. Strength is usually affected by the percentage of cotton used in the blend of materials for the
fiber. The engineer conducts a completely randomized experiment with five levels of cotton content and replicated
the experiment five times. The data are shown in the following table.

Cotton
Weight
Percentage Observations
15 7 7 15 11 9
20 12 17 12 18 18
25 14 19 19 18 18
30 19 25 22 19 23
35 7 10 11 15 11

(a) Is there evidence to support the claim that cotton content affects the mean tensile strength? Use  = 0.05.

One-way ANOVA: Tensile Strength versus Cotton Percentage

Analysis of Variance for Tensile

Source DF SS MS F P
Cotton P 4 475.76 118.94 14.76 0.000
Error 20 161.20 8.06
Total 24 636.96

Yes, the F-value is 14.76 with a corresponding P-value of 0.000. The percentage of cotton in the fiber appears to
have an affect on the tensile strength.

(b) Use the Fisher LSD method to make comparisons between the pairs of means. What conclusions can you draw?

Fisher's pairwise comparisons

Family error rate = 0.264

Individual error rate = 0.0500

Critical value = 2.086

Intervals for (column level mean) - (row level mean)

15 20 25 30

20 -9.346
-1.854

25 -11.546 -5.946
-4.054 1.546

30 -15.546 -9.946 -7.746

-8.054 -2.454 -0.254

35 -4.746 0.854 3.054 7.054

2.746 8.346 10.546 14.546

In the output the pairs of treatments that do not contain zero in the pair of numbers indicates that there is a difference
in the pairs of the treatments. 15% cotton is different than 20%, 25% and 30%. 20% cotton is different than 30%
and 35% cotton. 25% cotton is different than 30% and 35% cotton. 30% cotton is different than 35%.

(c) Analyze the residuals from this experiment and comment on model adequacy.

The residual plots below show nothing unusual.

 Normal Probability Plot of the Residuals
(response is Tensile Strength)
99

95

90

80
70
Percent

60
50
40
30
20

10

1
-5.0 -2.5 0.0 2.5 5.0
Residual

Residuals Versus the Fitted Values

(response is Tensile Strength)

5.0

2.5
Residual

0.0

-2.5

-5.0
10 12 14 16 18 20 22
Fitted Value
3.22. The response time in milliseconds was determined for three different types of circuits that could be used in an
automatic valve shutoff mechanism. The results are shown in the following table.

Circuit Type Response Time

1 9 12 10 8 15
2 20 21 23 17 30
3 6 5 8 16 7

(a) Test the hypothesis that the three circuit types have the same response time. Use  = 0.01.

From the computer printout, F=16.08, so there is at least one circuit type that is different.

Response: Response Time in ms

ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of Mean F
Source Squares DF Square Value Prob > F
Model 543.60 2 271.80 16.08 0.0004 significant
A 543.60 2 271.80 16.08 0.0004
Residual 202.80 12 16.90
Cor Total 746.40 14

The Model F-value of 16.08 implies the model is significant. There is only
a 0.04% chance that a "Model F-Value" this large could occur due to noise.

Treatment Means (Adjusted, If Necessary)

Estimated Standard
Mean Error
1-1 10.80 1.84
2-2 22.20 1.84
3-3 8.40 1.84

Mean Standard t for H0

Treatment Difference DF Error Coeff=0 Prob > |t|
1 vs 2 -11.40 1 2.60 -4.38 0.0009
1 vs 3 2.40 1 2.60 0.92 0.3742
2 vs 3 13.80 1 2.60 5.31 0.0002

(b) Use Tukey’s test to compare pairs of treatment means. Use  = 0.01.

MS E 16.90
S yi .    1.8385
n 5
q0.01,3,12  5.04
t0  1.83855.04 9.266
1 vs. 2: 10.8-22.2=11.4 > 9.266
1 vs. 3: 10.8-8.4=2.4 < 9.266
2 vs. 3: 22.2-8.4=13.8 > 9.266
1 and 2 are different. 2 and 3 are different.

Notice that the results indicate that the mean of treatment 2 differs from the means of both treatments 1 and 3, and
that the means for treatments 1 and 3 are the same. Notice also that the Fisher LSD procedure (see the computer
output) gives the same results.

(c) Use the graphical procedure in Section 3.5.3 to compare the treatment means. What conclusions can you draw?
How do they compare with the conclusions from part (a).

The scaled-t plot agrees with part (b). In this case, the large difference between the mean of treatment 2 and the
other two treatments is very obvious.
 Scaled t Distribution

(3) (1) (2)

5 10 15 20 25

Tensile Strength

(d) Construct a set of orthogonal contrasts, assuming that at the outset of the experiment you suspected the response
time of circuit type 2 to be different from the other two.

H 0  1  2 2  3  0
H1  1  2 2  3  0
C1  y1.  2 y2.  y3.
C1  54  2 111  42  126
126 
2

SSC1   529.2
5 6 
529.2
FC1   31.31
16.9

Type 2 differs from the average of type 1 and type 3.

(e) If you were a design engineer and you wished to minimize the response time, which circuit type would you
select?

Either type 1 or type 3 as they are not different from each other and have the lowest response time.

(f) Analyze the residuals from this experiment. Are the basic analysis of variance assumptions satisfied?

The normal probability plot has some points that do not lie along the line in the upper region. This may indicate
potential outliers in the data.
 Normal plot of residuals Residuals vs. Predicted
7.8

99

95
4.55
90
Norm al % probability

80
70

Res iduals
50 1.3

30
20
10
-1.95
5

-5.2

-5.2 -1.95 1.3 4.55 7.8 8.40 11.85 15.30 18.75 22.20

Res idual Predicted

Residuals vs. Circuit Type

7.8

4.55
Res iduals

1.3

-1.95

-5.2

1 2 3

Circuit Type
3.49. Consider the data shown in Problem 3.22.

(a) Write out the least squares normal equations for this problem, and solve them for  and  i , using the usual

constraint  3
ˆ  0
i 1 i
. Estimate   .
1 2

15̂  5ˆ1  5ˆ2  5ˆ3 =207

5̂  5ˆ1 =54

5̂  5ˆ2 =111
5̂  5ˆ3 =42

3
Imposing ˆ
i 1
i  0 , therefore ˆ  13.80 , ˆ1  3.00 , ˆ2  8.40 , ˆ3  5.40

ˆ1  ˆ2  3.00  8.40  11.40

(b) Solve the equations in (a) using the constraint ˆ3  0 . Are the estimators ˆi and ̂ the same as you found in
(a)? Why? Now estimate  1   2 and compare your answer with that for (a). What statement can you make
about estimating contrasts in the  i ?

Imposing the constraint, ˆ3  0 we get the following solution to the normal equations:
ˆ  8.40 ,
ˆ1  2.40 ,
ˆ2  13.8 , and
ˆ3  0 .
These estimators are not the same as in part (a). However, ˆ1  ˆ2  2.40  13.80  11.40 , is the same as in
part (a). The contrasts are estimable.

(c) Estimate    1 , 2 1   2   3 and    1   2 using the two solutions to the normal equations. Compare the
results obtained in each case.

Contrast Estimated from Part (a) Estimated from Part (b)

1   1 10.80 10.80
2 2 1   2   3 -9.00 -9.00
3  1  2 19.20 24.60

Contrasts 1 and 2 are estimable, 3 is not estimable.

3.34 An article in the Journal of Quality Technology (Vol. 13, No. 2, 1981, pp. 111-114) describes and experiment
that investigates the effects of four bleaching chemicals on pulp brightness. These four chemicals were selected at
random from a large population of potential bleaching agents. The data are as follows:

Chemicals Brightness
1 77.199 74.466 92.746 76.208 82.876
2 80.522 79.306 81.914 80.346 73.385
3 79.417 78.017 91.596 80.802 80.626
4 78.001 78.358 77.544 77.364 77.386

(a) Is there a difference in the chemical types? Use Use α=0.05.

From the analysis below, there does not appear to be a difference

Source DF Sum of Squares Mean Square F Ratio Prob > F

Model 3 53.98207 17.9940 0.7498 0.5383
Error 16 383.99085 23.9994
C. Total 19 437.97292

(c) Estimate the variability due to random error.

From the output shown above, the variance component due to random error is 23.999428.

3.51 Use the Kruskal-Wallis test for the experiment in Problem 3.23. Are the results comparable to those found by
the usual analysis of variance?

Fluid Type Life (in h) at 35 kV Load

1 17.6 18.9 16.3 17.4 20.1 21.6
2 16.9 15.3 18.6 17.1 19.5 20.3
3 21.4 23.6 19.4 18.5 20.5 22.3
4 19.3 21.1 16.9 17.5 18.3 19.8

Wilcoxon Scores (Rank Sums) for Variable Life

Classified by Variable FluidType
FluidType N Sum of Expected Std Dev Mean
Scores Under H0 Under H0 Score
1 6 67.00 75.0 14.996739 11.166667
2 6 53.50 75.0 14.996739 8.916667
3 6 111.00 75.0 14.996739 18.500000
4 6 68.50 75.0 14.996739 11.416667
Average scores were used for ties.

Kruskal-Wallis Test
Chi-Square 6.2177
DF 3
Pr > Chi-Square0.1015
Graduate Students Exercises

3.36. Consider testing the equality of the means of two normal populations, where the variances are unknown but
are assumed to be equal. The appropriate test procedure is the pooled t test. Show that the pooled t test is equivalent
to the single factor analysis of variance.

y1.  y 2.
t0  ~ t 2 n  2 assuming n1 = n2 = n
2
Sp
n
n n 2 n

 y1 j  y1. 2   y2 j  y2. 2  yij  y1. 2

j 1 j 1 i 1 j 1
Sp    MS E for a=2
2n  2 2n  2

2
2 n  yi2. y..2
Furthermore, y1.  y2.    
2

i 1
n 2n
 , which is exactly the same as SSTreatments in a one-way classification with

a=2.

SS Treatments
Thus we have shown that t 02  .
MS E
In general, we know that t u2  F1,u so that t 02 ~ F1,2n 2 .

Thus the square of the test statistic from the pooled t-test is the same test statistic that results from a single-factor
analysis of variance with a=2.

a a
3.37. Show that the variance of the linear combination 
i 1
c i y i . is  2 n c
i 1
2
i i .

 a  a a  ni  a ni


V  ci yi .  
 i 1 
 V ci yi .    ci2V   yij  
 j 1 
 V y , V yij   2
ci2 ij .
i 1 i 1   i 1 j 1
a
 c
i 1
2
i ni 
2
3.38. In a fixed effects experiment, suppose that there are n observations for each of four treatments. Let
Q12 , Q22 , Q32 be single-degree-of-freedom components for the orthogonal contrasts. Prove that
SS Treatments  Q12  Q22  Q32 .

Since there are four treatments, there are 3 independent treatment contrasts. These can be selected to be orthogonal.
Suppose the coefficients of the contrast are ci  ci1 ci 2 ci 3 ci 4  , with c ij  0 for i= 1, 2, 3.
j

 c   c 
2 2
We could assume without loss of generality that ij  1 (if ij  1 one could normalize the constants
cij
and select cij*  )
c 2
ij

We could write Qi  n  cij yi.  ci' z where ci  ci1 ci 2 ci 3 ci 4  and z.  n  y1. y2. y3. y4. 
. Because the contrasts are orthogonal (and the number of observations are the same in all treatments) then
ci  c 'j   cij crj  0 .
j

Define the matrix  

C   ' c1' c2' c`'3 where   1 / 4  1 1 1 1 / 4 . Note that C ' C  I 4 x 4 and
Q  Q  Q  zCRC z with CR  c1' c2' c3'
2
1
2
2
2
3
' '
R  
Define the matrix W  I 4 x 4   ' . Note that SST  zWz.'
Now note that Wci'  ( I 4 x 4   ' )ci'  ci' . Therefore the ci' s are eigenvectors of W associated with eigenvalue of
one. Also observe that W '   '   '  0 so  ' is also an eigenvector associated with the eigenvalue of zero.
Because the set of eigenvector  , c1 , c2 , c3  is orthogonal it forms a basis for the column vectors of W .
' ' ' '

0 0  0 0  '
C 'WC  
Therefore,  and W  C  C  CRCR . This implies that
'

0 I 3 x3  0 I 3 x3 

SST  zWz.  zCRCR z  Q1  Q2  Q3 .

' ' ' 2 2 2