CHAPTER 8
Design of experiments
8.1 Introduction
8.2 Concepts from Experimental design
8.2.1 Some terminology
8.2.2 Fundamental principles: replication, randomization and blocking
8.2.3 Some specific designs
8.3 Types of Experiments
8.3.1 one factor at a time Experiments
8.3.2 Full factorial Experiments
8.3.2 Fractional factorial Experiments
8.4 Optimal design
8.5 The Taguchi approach
8.6 Chapter Summary
8.7 Computer examples
Projects for chapter 8
Chapter 8
Exercises 8.2
8.2.1
Response is amount of fat was absorbed by has-brown potatoes
Factors are frying durations and different type of fats
Factor types are frying durations which quantitative and continuous and different type of fats
which is qualitative and discrete.
Treatments
There are 16 treatments
2 min with animal fat I, 2 min with animal fat II, 2 min with vegetable fat I and 2 min with
vegetable fat II
3 min with animal fat I, 3 min with animal fat II, 3 min with vegetable fat I and 3 min with
vegetable fat II
4 min with animal fat I, 4 min with animal fat II, 4 min with vegetable fat I and 4 min with
vegetable fat II
5 min with animal fat I, 5 min with animal fat II, 5 min with vegetable fat I and 5 min with
vegetable fat II
8.2.2
(a) Response is the number of offspring born for a specific species of mice
(b) Factor is different type of vitamins and they are qualitative and discrete type.
8.2.3
Procedure for random assignment
1. Number the experimental units from 1 to 30.
2. Use a random number table or a statistical software to get a list of numbers that is a random
permutation of the numbers 1 to 30.
3. Give treatment 1 to the experimental units having the first 10 numbers in the list. Treatment
2 will be given to the next 10 numbers in the list, and so on, give treatment 3 to the last 10
units in the list.
Here response is rose bushes and facor is different fertilizers.
24, 12, 30, 21, 8, 3, 20, 1, 11,18, 13, 15, 28, 5, 25, 29, 4, 10, 14, 19, 26, 9, 2, 6, 22, 16, 23, 7,
27,17
Brand Subject
A 24 12 30 21 8 3 20 1 11 18
B 13 15 28 5 25 29 4 10 14 19
C 26 9 2 6 22 16 23 7 27 17
8.2.4
Procedure for randomization in a randomized complete block design
1. Group the experimental units into 3 groups (blocks) containing 5 homogeneous
experimental units.
2. In group 1, number the experimental units from 1 to 5 and obtain a random permutation of
numbers 1 to k using a random number generator.
3. In group 1, the experimental unit corresponding to the first number in the permutation
receives treatment 1, the experimental unit corresponding to the second number in
permutation receives treatment 2, and so on.
4. Repeat steps 2 and 3 for each of the remaining blocks.
G R J
3(A) 5(A) 1(A)
1(B) 4(B) 2(B)
2(C) 3(C) 4(C)
5(A) 1(B) 3(A)
4(B) 2(C) 5(C)
8.2.5
Procedure for a randomized complete block design with 3 replications
1. Group the experimental units into 3 groups (called blocks), each containing 3*3
homogeneous experimental units.
2. In group 1, number the experimental units from 1 to 9 and generate a list of numbers
which are random permutation of the numbers 1 to 9.
3. In group 1, assign treatment 1 to the experimental units having numbers given by the first
3 numbers in the list. Assign treatment 2 to the experiments having next 3numbers in the
list, and so on until treatment 3 receives 3 experimental units.
4. Repeat steps 2 and 3 for the remaining blocks of experimental units.
G R J
3(A) 5(A) 1(A)
1(A) 4(A) 2(A)
2(A) 3(A) 4(A)
6(B) 9(B) 7(B)
8(B) 1(B) 8(B)
7(B) 6(B) 5(B)
9(C) 8(C) 6(C)
5(C) 7(C) 3(C)
4(C) 2(C) 9(C)
8.2.6.
The main objective of randomization it to negate the effects of all uncontrolled extraneous
variables.
8.2.7.
19, 37, 52, 42, 13, 34, 56, 48, 44, 43, 24, 12, 5, 32, 40, 23, 41, 11, 10, 6, 30, 26, 18, 8, 2, 29,
21, 36, 1, 54, 20, 39, 33, 27, 49, 16, 51, 15, 28, 47, 53, 35, 31, 3, 38, 25, 17, 55, 4, 50, 14, 22,
9, 46, 45, 7
Brand Subject
A 19 37 52 42 13 34 56 48 44 43 24 12 5 32
B 40 23 41 11 10 6 30 26 18 8 2 29 21 36
C 1 54 20 39 33 27 49 16 51 15 28 47 53 35
D 31 3 38 25 17 55 4 50 14 22 9 46 45 7
8.2.8.
AI stands for animal fat I and so on.
four pans are A, B, C and D.
AI AII VI VII
7(A) 14(A) 12(A) 5(A)
20(B) 22(B) 19(B) 2(A)
4(C) 3(C) 10(C) 23(A)
21(D) 13(B) 8(A) 6(A)
11(A) 9(B) 17(C) 18(A)
1(B) 16(C) 24(D) 15(D)
By latin square
pans
fat 1 2 3 4
AI D B A C
AII C A D B
VI B D C A
VII A C B D
8.2.9.
Start
New material
Days 1 3 2 4 5
2 A B C D E
3 B C D E A
5 C D E A B
1 D E A B C
4 E A B C D
Then
New material
Days 1 3 2 4 5
1 D E A B C
2 A B C D E
3 B C D E A
4 E A B C D
5 C D E A B
The final is
New material
Days 1 2 3 4 5
1 D A E B C
2 A C B D E
3 B D C E A
4 E B A C D
5 C E D A B
8.2.10.
A stands for Anna and C stands for Carol and so on.
Man
Woman E J M R
A Su F T Sa
C Sa T Su F
J F Su Sa T
N T Sa F Su
T stands for Thursday, F stands for Friday, Sa stands for Saturday and Su stands for Sunday.
8.2.11.
Grid
Grid 1 2 3 4
1 D A B C
2 C B D A
3 A D C B
4 B C A D
8.2.12.
Z is 1.64
E=5
S=(60-20)/4=10
Sample size is 10.758
Therefore we need 11.
Exercises 8.3
8.3.1
one factor at a time experiment to predict average amount of profit
Assume we fix proportion at 40% then if we increase the quality from ordinary to fine we get
average profit decrease from 25,000 to 10,000. If we fix proportion at 60% then if we increase
the quality from ordinary to fine we get average profit decrease from 9500 to 3000.
Then if we change the setting from 40% and ordinary to 60% and fine then we get from
25,000 to 3000.
8.3.2
8.3.3
In fractional factorial experiment, only a fraction of the possible treatments are actually
used in the experiment. A full factorial design is most ideal design through which we could
obtain information on all main effects and interactions. Due to prohibitive size of the
experiments, such designs are not practical to run. The total number of distinct treatments will
be .
Fractional factorial experiments are used in which trials are conducted at only a well-
balanced subset of the possible combinations of levels of factors. This allows the
experimenter to obtain information about all main effects and interactions while maintaining
the size of the experiment manageable.
The experiment is carried out in a single systematic effort.
8.3.4
(a) factor-level combinations are possible in this experiment
(b) by randomize the order of factor levels for both row and columns and then give treatment
for each level.
Exercises 8.4
8.4.1．
and
8.4.2
Then
Thus n has to be infinite for these equations to hold.
Exercise 8.5
8.5.1
8.5.2
(a)
For run 1 the test statistic is 7.358804
For run 2 the test statistic is 7.586248
For run 3 the test statistic is 6.484102
For run 4 the test statistic is 7.018697
Therefore run 3 is the best and run 4 is the 2nd best, then run 1 and last run 2.
(b)
For run 1 the test statistic is 2.033047
For run 2 the test statistic is 2.288109
For run 3 the test statistic is 1.170693
For run 4 the test statistic is 1.718720
Therefore run 3 is the best and run 4 is the 2nd best, then run 2 and last run 1.
The resluts are very similar except run 1 and run 2 switch the order.
