ASSIGNMENT 1

Design and Analysis of Multi-

Factored Experiments
One and Two-Factor Factorial Design

Submitted by :
Kshama Sundar Roy
Student ID : 201088010
  Question no: 1

Consider an experiment that you have conducted in the past. What was the response, the factors
and levels considered, and how was the experiment analyzed? What was the outcome or conclusion
of that experiment? If you have never conducted an experiment in the past, think of an experiment
that you may conduct that might be of interest and list the response (s), factors, possible levels, and
what you hope the experiment will answer.

o Solution:

 Title of the Experiment :

Influence of modulus of subgrade reaction & cantilever length on the deflection of mat
foundation.

 Objective:

To determine the deflection of mat foundation for different modulus of subgrade reactions &
cantilever lengths.

 Description :

Analysis of Mat foundation and study of results reveal that mat foundation is extensively
over-designed in the current approach. Mat foundation of constant thickness is designed by
ACI method, Conventional method or computer analysis by Finite Difference method or
Finite Element method. This over-design is due to uncertainty in the analysis methods and
lack of understanding of mat behaviour. In the conventional present practices soil pressure is
considered as uniform on mat foundation. But it is not always true because soil does not give
uniform pressure. To improve upon this procedure, an extensive investigation has been
carried out here on the behaviour of mat foundation using finite element program named
ETABS. The most significant parameters for mat foundation are mat thicknesses and
modulus of subgrade reaction. Moreover the effect of cantilever length on the mat foundation
is also important. The effect of two parameters modulus of subgrade reaction and cantilever
length for a constant mat thickness on the deflection of mat foundation was investigated here.

 Data :
Elastic foundation (spring)
Z-direction only
Column size = 2ft * 2ft

Assignment 1, St ID: 201088010

2
 Uniform soil pressure = 4 ksf
No. of Columns on footing = 12
Total load on footing = 16800k
Thickness, t = 5ft

Length of Cantilever portion, x1 = 5ft

x2 = 0ft

Spring Constants/ Modulus of subgrade reaction used,

K1 = 150 K/ft3
K2 = 200 K/ft3
K3 = 300 K/ft3

 Result :

Deflection in inches

Modulus of subgrade reaction in K/ft3

K=150 K=200 K=300

L=5 1.3 1 0.6

Cantilever
length L=0 1 0.8 0.4

So from the above consideration we can say that for the above experiment,

Criteria No Description
Factor Modulus of subgrade reaction,
2
cantilever length
Modulus of
subgrade 3 150,200,300
Level reaction
cantilever
2 5,0
length
Response
1 deflection

From the result chart, it is clear that, deflection of mat foundation decreases with the increase of
Modulus of subgrade reaction and also decreases with the decrease of cantilever length. We can also
understand the relationship among them by graphical representation.

Assignment 1, St ID: 201088010

3
 Graphical Representation:

Fig. 1: Variation of deflection with modulus of subgrade reaction at different cantilever length

Fig. 2: Variation of deflection with cantilever length at different modulus of subgrade reaction

Assignment 1, St ID: 201088010

4
  Conclusions:

Figure 1 & 2 shows the influence of modulus of subgrade reaction & cantilever length on the deflection of
mat foundation more clearly. From Fig 1 we can see that deflection of mat foundation decreases with the
increase of Modulus of subgrade reaction at different cantilever length. From fig 2 we can see that
deflection increases with the increase of cantilever length.

So we can conclude that the deflection of mat foundation at constant thickness depends on modulus of
subgrade reaction and on cantilever length.

 Question No : 2

Air pollution concentrations at 11 locations of a building were measured after implementing certain
control measures. Measurements were taken at two different times and the following values in
parts per million of SO2 were obtained:

Time 1: 66.3 65.5 64.9 62.8 63.3 64.4 65.2 63.5 69.4 63.2 67.4
Time 2: 70.3 60.4 65.6 63.9 68.9 71.1 62.8 68.9 65.8 67.2 61.2

It is desired to show that the pollution values obtained at Time 1 are not significantly different from
that at Time 2 at the 5% significance level.

a) Which statistical test should be carried out? Justify your choice. What is the
conclusion from the test?
b) What is the main assumption of the test used? Verify the assumption of the test
used.
c) What alternative tests to the one you used can also be used?

o Solution :

a) Here there are two sets of data collected at eleven (11) different locations and only one factor
is considered. So we can apply t-test.

There are three different types of t-test. They are –

1) Two independent sample t-test

i) Variances are equal
ii) Variances are not equal
2) Paired t-test

Here the given data groups are not independent. They are dependent on another factor called location that
means that SO2 values are obtained at eleven specific locations and in each location at two different times

Assignment 1, St ID: 201088010

5
Time1 and Time 2. In other way, the pollution values (SO2) are measured twice at Time 1 and Time 2 at
eleven different locations. So the data groups are co-related to each other and to the locations.

So paired t-test is the right answer and will be carried out here for this problem.

o Manual calculations :

First assume that the pollution values obtained at Time 1 and time 2 are not different.

So Null Hypothesis, Ho: µ1=µ2

Alternative Hypothesis, Ha: µ1≠µ2

From the Box plot

Box Plot of Time 1, Time 2

72

70

68
Data

66

64

62

60
Time1 Time2

We can see that the means of Time 1 and Time 2 are different. Now we will check whether it is
statistically significant or not.

Assignment 1, St ID: 201088010

6
 Amount of SO2(ppm)
Location
Time 1 (X1) Time 2 (X2) D(X1- X2 ) D2

1 66.3 70.3 -4 16

2 65.5 60.4 5.1 26.01

3 64.9 65.6 -0.7 0.49

4 62.8 63.9 -1.1 1.21

5 63.3 68.9 -5.6 31.36

6 64.4 71.1 -6.7 44.89

7 65.2 62.8 2.4 5.76

8 63.5 68.9 -5.4 29.16

9 69.4 65.8 3.6 12.96

10 63.2 67.2 -4 16

11 67.4 61.2 6.2 38.44

X 1= 65.08 X 2= 66.01 ∑D = -10.2 ∑D2 =222.28

Here, N= 11

D = ∑D/N = -10.2/11 = -0.93= X 1 - X 2 =-0.93

Sd2 = (N∑D2 - (∑D)2)/(N(N-1)) = (11*222.28 – (-10)2)/(11*10) = 21.28

Sd =√21.28 =4.613

to = D /( Sd/√N) = (-0.93)/(4.613/√11) = -0.67

for 5% significance level α = 0.05

From Table for critical values of t (Ref : Table V, Introductory Statistics, 4 th edition by Ronald
J.Wonnacott , Thomas H. Wannacott) we get

t0.025,10 = 2.23

So Ho will be rejected if |to| > t0.025,10 = 2.23

And because |to|= 0.67 < t0.025,10 = 2.23

Assignment 1, St ID: 201088010

7
We cannot reject the Hypothesis Ho: µ1=µ2

That means the pollution values obtained at Time 1 are not significantly different from at Time 2 at 5%
significance level.

t Distribution Plot
df=10

0.4

0.3
Probability Density

0.2

0.1

0.025 0.025
0.0
-2.23 t0= -0.67 0 2.23
t

The fig shows that for to distribution with 10 degrees of freedom, the reference distribution for this test
with the value of to shown relative to the critical region.

o Calculation by Minitab 15:

The following table shows the computer output from the Minitab 15 paired t-test procedure for this
problem.

Assignment 1, St ID: 201088010

8
Again for 5% significance level Ho will be rejected if p value< 0.05

From the result we can see p value = 0.520 > 0.05

So failed to reject null hypothesis of Ho: µ1=µ2

So the pollution values at Time 1 and Time 2 are not significantly different.

So in conclusion we can say that paired t-test is appropriate for this statistical test and the pollution values
obtained at Time 1 are not significantly different from that at Time 2 at the 5% significance level.

b) Main assumption :

Both groups of data are normally distributed around their respective means.

Other assumptions are –

1. Both groups are assumed to have identical distributions which differ in their central location.
2. T-test is a test for differences in central location only.
3. There is an additive difference between the two means if any different exists.

Now we will check the above assumptions of t-test for Time 1 & Time 2. For this purpose, the data are
given input to MINITAB 15 and different plots are obtained which are stated below.

Probability Plot of Time 1

Normal - 95% CI
99
Mean 65.08
StDev 2.015
95 N 11
AD 0.349
90
P-Value 0.406
80
70
Percent

60
50
40
30
20

10

1
60 63 66 69 72
Time 1

Assignment 1, St ID: 201088010

9
 Probability Plot of Time 2
Normal - 95% CI
99
Mean 66.01
StDev 3.633
95 N 11
AD 0.212
90
P-Value 0.807
80
70
Percent

60
50
40
30
20

10

1
50 55 60 65 70 75 80
Time 2

The two plots mentioned above are normality test of Time 1 and Time 2 with 5% confidence level
respectively. These plots show that the distributions are normal.

Probability Plot of Time 1, time 2

Normal - 95% CI
99
Variable
C1
95 C2

90 Mean StDev N AD P
65.08 2.015 11 0.349 0.406
80 66.01 3.633 11 0.212 0.807
70
Percent

60
50
40
30
20

10

1
50 55 60 65 70 75 80
Data

The above normal probability plot indicates that the collected data are around the means and within 95%
confidence interval i.e. the both data group are normally distributed.

Assignment 1, St ID: 201088010

10
 Normal Distribution Plot
Mean of Time 1=65.08
0.20 Mean StDev
65.08 2.02
66.01 3.63

0.15
Density

0.10 Time 1 :
_______
Time 2 :
0.05 ------

0.00
55 60 65 70 75 80
X
Mean of Time 2=66.01

From the above plot we can see that both data groups form Bell shape curve at two different mean
positions and the bell shape curve is the characteristic property of normal distribution.

So we can conclude that the assumption of the t-test is satisfied for this problem.

c) Alternative test :

I think that ANOVA can also be used here.

But as the data are paired so we have to consider location as a factor. Because the both set data are co-
related with location and the response, pollution value can be changed if the location is changed. So two
factor ANOVA will be an alternate approach.

From the given data we can find the following table

Criteria No Description
Factor
2 Time, Location
Time 2 Time 1, Time 2
Level
Location 11 1,2,3,4,5,6,7,8,9,10,11
Response
1 SO2

Assignment 1, St ID: 201088010

11
So two factor ANOVA is used to analyze this problem.

Now let’s consider the Null & Alternative Hypotheses for this problem.

Null Hypothesis, Ho:

-- The pollution values obtained at Time 1are not significantly different from that at Time 2. (α =
0.05)

Alternative hypothesis, Ha:

-- The pollution values obtained at Time 1are significantly different from that at Time 2. (α =
0.05)

Design-Expert 7.1.3 is used to analyze the problem.

The above calculation shows that, MS(treatment) is smaller than MS(error) which results lower F values.

Again we can see that for location P=0.7668 >0.05 and for time P=0.5201>0.05 that means they are not
significant. So we can not reject null hypothesis. So the pollution values obtained at Time 1 are not
significantly different from that at Time 2.

Moreover the P value for time here 0.5201 is exactly equal to the P value 0.520 which was obtained from
the paired t-test.

So we can conclude that two factor ANOVA is an alternative test of paired t-test.

Assignment 1, St ID: 201088010

12
  Question no : 3

A large portion of chromium contaminated water was divided into 20 identical samples. Five
samples were sent to each of 4 laboratories and the following data were produced. Are the
laboratories making consistent measurements?

Lab Chromium concentrations

1 26.1 21.5 22.0 22.6 24.9

2 18.3 19.7 18.0 17.4 22.6
3 19.1 13.9 15.7 18.6 19.1
4 30.7 27.3 20.9 29.0 26.1

a) Are the concentrations measured the same at each lab? Use = 0.05.
b) Analyze the residuals from this experiment. Are the analysis of variance assumptions
satisfied?
c) If there is a statistically significant difference, which labs are different and which are
similar?

o Solution:

a) From the given data we can find the following table

Criteria No Description
Factor
1 Lab
Level
4 Lab 1, Lab 2, Lab 3, Lab 4
Response
1 Chromium concentrations
Replication
5 For each lab

So one way ANOVA is used to analyze this problem.

Now let’s consider the Null & Alternative Hypotheses for this problem.

Null Hypothesis, Ho:

-- the concentrations measured are the same at each lab. (α = 0.05)

Alternative hypothesis, Ha:

-- at least, mean concentration of one lab will be different.

Design-Expert 7.1.3 is used to analyze the problem.

Assignment 1, St ID: 201088010

13
The above calculation shows that, MS(treatment)=91.25 is larger than MS(error)=6.93 which
results a larger F value.

From the F-Table for 5% (Ref: Appendix 4, page no 618, Design and Analysis of Experiments, 7th
edition by Douglas C. Montgomery) we get Fcritical=3.239 (for dfnum= 3 and dfden= 16)

So F=13.16 > Fcritical=3.239 which means that test is significant.

Moreover p-value << 0.05 also means test is highly significant. So reject Ho, which indicates the
means are different.

Again in terms of Treatment effects:

Assignment 1, St ID: 201088010

14
Here, Prob > |t| < 0.05 in the cases of 1vs 2, 1 vs 3, 2 vs 4 and 3 vs 4 indicates the difference in
two treatment means are significant. So reject Ho and accept Ha which is at least mean
concentration of one lab will be different.

We can also compare the equality of means by Box plot (using MINITAB 15).

Box plot of chromium concentrations at 4 laboratories

32.5

30.0

27.5

25.0
Data

22.5

20.0

17.5

15.0

1 2 3 4

The box plot also gives the clear evidence that the means are not equal i.e. the concentrations
measured are different at each lab.

Therefore we can conclude that the concentrations measured are not the same at each lab.

Assignment 1, St ID: 201088010

15
 b) Residual Analysis :

Assumption of ANOVA:

1. All samples are random samples from their respective populations.

2. All samples are independent of one another.
3. Departures from group mean are normally distributed for all groups.
4. All groups have equal variance.

Design Expert 7.1.3 is used to represent the residual analysis from the experiment.

I. Normal plot :

The normal probability plot shows that most of points follow a straight line that means the
distribution of residuals is almost normal.

Assignment 1, St ID: 201088010

16
 II. Residuals vs. Predicted :

The plot of Residuals Vs Predicted looks like well scattered and is not like a outward opening funnel
shape, so no transformation required. Moreover the randomly scattered plot indicates the constant range
of residuals across the graph.

III. Residuals vs Run :

From the above plot we find that most of the data are randomly scattered which indicates no influence of
lurking variables.

Assignment 1, St ID: 201088010

17
 IV. Predicted vs Actual :

The plot shows that the data points are split evenly by the 450 line which is also an indication for no
transformation.

V. Box-Cox Plot for Power Transforms :

Finally from the Box-Cox plot we can see that the current line(blue line) is between the ranges(the red
lines) and near the best value line (the green line). This recommends for no transformation.

So, we can conclude that the assumptions of ANOVA are satisfied.

Assignment 1, St ID: 201088010

18
 c) From the ANOVA we saw that there is a statistically significant difference.

So from the Box plot given below

Box plot of chromium concentrations at 4 laboratories

32.5

30.0

27.5

25.0
Data

22.5

20.0

17.5

15.0

1 2 3 4

We can see that Lab 2 and Lab 3 are similar whereas Lab 1 and Lab 4 are different.

Now the Model graph obtained from the Design Expert 7.1.3 is given below.

We can see that Lab 2 and Lab 3 are similar because there is a slight difference between their means and
Lab 1 and Lab 4 are different because there exists a big difference between their means.

So we can conclude that Lab 2 and Lab 3 are similar whereas Lab 1 and Lab 4 are different.

Assignment 1, St ID: 201088010

19
  Question No : 4

An article in the Journal of Testing and Evaluation (Vol 16 No. 2 pp 508-515) investigated the
effects of cycling loading and environmental conditions on fatigue crack growth at a constant 22
MPa stress for a particular material. The data from the experiment are shown below. The response
is crack growth rate.

Environment

Frequency Air H2O Salt H2O

2.29 2.06 1.90

10 2.47 2.05 1.93
2.48 2.23 1.75
2.12 2.03 2.06

2.65 3.20 3.10

1 2.68 3.18 3.24
2.06 3.96 3.98
2.38 3.64 3.24

2.24 11.0 9.96

0.1 2.71 11.0 10.01
2.81 9.06 9.36
2.08 11.30 10.40
_____________________________________________

Analyze the data and draw the appropriate conclusions, checking all necessary
assumptions. Interpret the results.

Solution :

From the given data we can find the following table

Criteria No Description
Factor
2 Frequency, Environment
Frequency 3 10,1,0.1
Level
Environment 3 Air,H2O,Salt H2O
Response
1 Crack growth rate
Replication
4

Assignment 1, St ID: 201088010

20
 So Two Factor ANOVA is used to analyze this problem.

Now let’s consider the Null & Alternative Hypotheses for this problem.

Null Hypothesis, Ho:

-- Means of all factors and their interactions are equal. (α = 0.05)

Alternative hypothesis, Ha:

-- at least, in one case mean will be different.

Design-Expert 7.1.3 is used to analyze the problem.

o Analysis :

The above calculation shows that, MS(treatments) are larger than MS(error) which results a larger F value
for both Factors A and B and their interaction AB.

From the F-Table for 5% (Ref: Appendix 4, page no 618, Design and Analysis of Experiments, 7th edition
by Douglas C. Montgomery) we get for Factor A, Fcritical=3.354 (for dfnum= 2 and dfden= 27), for Factor B,
Fcritical=3.354 (for dfnum= 2 and dfden= 27), for Interaction AB, Fcritical=2.728 (for dfnum= 4 and dfden= 27)

Assignment 1, St ID: 201088010

21
So for Factor A, F=522.40 > Fcritical=3.354, for Factor B, F=159.92 > Fcritical=3.354, for Interaction AB,
F=126.89 > Fcritical=2.728 which means that test is significant.

Moreover p-value << 0.05 for all cases also mean that the test is highly significant. So reject Ho, which
indicates the means are different.

o Assumptions :

Assumption of ANOVA:

1. All samples are random samples from their respective populations.

2. All samples are independent of one another.
3. Departures from group mean are normally distributed for all groups.
4. All groups have equal variance.

Design Expert 7.1.3 is used to represent the residual analysis from the experiment.

I. Normal plot :

The normal probability plot shows that most of points follow a straight line that means the distribution of
residuals is almost normal.

Assignment 1, St ID: 201088010

22
 II. Residuals vs. Predicted :

The plot of Residuals Vs Predicted does not look like well scattered and is like a outward opening funnel
shape, so possibly transformation is required.

III. Residuals vs Run :

From the above plot we find that most of the data are randomly scattered but one point is well below the
red line.

Assignment 1, St ID: 201088010

23
 IV. Predicted vs Actual :

The plot shows that the data points are not split evenly by the 450 line which is also an indication for
transformation.

V. Box-Cox Plot for Power Transforms :

Assignment 1, St ID: 201088010

24
Finally from the Box-Cox plot we can see that the current line(blue line) is far beyond the ranges(the red
lines) and away from the best value line (the green line). This recommends for transformation and the
Design Expert is suggesting for Log transformation.

So, we will analyze the problem again with Natural Log transformation.

o Analysis with Log Transformation:

Now we can see that the results are different than previous.

Again the above calculation shows that, MS(treatments) are larger than MS(error) which results a larger F
value for both Factors A and B and also their interaction AB.

From the F-Table for 5% (Ref: Appendix 4, page no 618, Design and Analysis of Experiments, 7th edition
by Douglas C. Montgomery) we get for Factor A, Fcritical=3.354 (for dfnum= 2 and dfden= 27), for Factor B,
Fcritical=3.354 (for dfnum= 2 and dfden= 27), for Interaction AB, Fcritical=2.728 (for dfnum= 4 and dfden= 27)

So now for Factor A, F=404.09 > Fcritical=3.354, for Factor B, F=125.85 > Fcritical=3.354, for Interaction
AB, F=94.17 > Fcritical=2.728 which means that test is significant.

Moreover p-value << 0.05 for all cases also mean that the test is highly significant. So reject Ho, which
indicates the means are different.

Assignment 1, St ID: 201088010

25
 o Checking Assumptions :
o Normal plot :

The normal probability plot shows that most of points follow a straight line that means the distribution of
residuals is almost normal.

o Residuals vs. Predicted :

The plot of Residuals Vs Predicted now looks like well scattered and is not like an outward opening
funnel shape as seen before, so now the assumption is satisfied.

Assignment 1, St ID: 201088010

26
 o Residuals vs Run :

From the above plot now we find that most of the data are randomly scattered and the point which was
well below the red line before is now within the red lines. So the scattered plot indicates that the
sequence of runs were random.

o Predicted vs Actual :

Now the plot shows that the data points are split evenly by the 450 line which is also an indication for no
transformation.

Assignment 1, St ID: 201088010

27
 o Box-Cox Plot for Power Transforms :

Finally from the Box-Cox plot we now see that the current line (blue line) is within the ranges (the red
lines) and very near to the best value line (the green line).

So all these recommend that the log transformation we did is right and all the assumptions are satisfied.

Assignment 1, St ID: 201088010

28
 o Conclusions:

The model graph shows that for air environment the crack growth rate does not depend on frequency that
means in air the crack growth rate is almost same for all three different frequencies. In water the crack
growth rate increases as frequency decreases that means the crack growth rate is maximum for lowest
frequency and minimum for highest frequency. In salt water the crack growth rate behaves almost same
as in water.

Again if we change the environment from air to water there exists a good interaction as the lines are not
parallel. But if we change the environment from water to salt water the interaction becomes very small as
the lines become almost parallel.

When we change the environment from air to salt and to salt water gradually for high frequency the crack
growth rate does not change significantly but for low frequency the crack growth rate increases rapidly
from air to water and then remains almost same from water to salt water.

Again for low frequency and for water environment, the crack growth rate is maximum and for high
frequency and in salt water environment, the crack growth rate is minimum

We can also see the same results from 3-D Surface plot.

Assignment 1, St ID: 201088010

29
From the above graph we can see that for low frequency and for water environment, the crack growth rate
is maximum.

From the above graph we can see that for high frequency and in salt water environment, the crack growth
rate is minimum.

So we can conclude that for low frequency and for water environment, the crack growth rate is maximum
and for high frequency and in salt water environment, the crack growth rate is minimum.

-------------------------------------------------------End-------------------------------------------------------------------

Assignment 1, St ID: 201088010

30