Chapter Three

Design and Analysis of Experiments

A. Introduction
Design of experiment means how to design an experiment in the sense that how the observations or
measurements should be obtained to answer a query in a valid, efficient and economical way.

Designing an experiment means deciding how the observations or measurement should be taken to
answer a particular question in a valid, efficient and economical way.

If the experiment is designed properly keeping in mind the question, then the data generated is valid and
proper analysis of data provides the valid statistical inferences. If the experiment is not well designed, the
validity of the statistical inferences is questionable and may be invalid.

3.1. Approaches to Experiments

1. Conventional approach to Experiment (One Factor At a Time- OFAT)
 It is used when we want evaluate the effect of one factor at a time.

Limitation of OFAT:

 It will not give us the interaction various factors each other. Interaction means the effect
of a factor depends on setting of the other factor.
 We can miss the optimum setting of the factors to get best results
2. Statistically Designed Experiments
 The interaction effect of factors can be manipulated.
 We can test less and learn more.
 We can find optimum settings to get the best from the process.

3.2. Characteristics of Good Experiment

A good experiment is:

 Unbiased ( Randomized and Blinding)

 High precision ( Uniform materials, replication, Stratification)
 Simple

1|Page
  Wide range of application (Deliberate variation, factorial design).
 Able to estimate uncertainty (Replication, Randomization)

3.3. Terminologies used in DOE

 Experimental unit:

For conducting an experiment, the experimental material is divided into smaller parts and each part is
referred to as an experimental unit. The experimental unit is randomly assigned to treatment is the
experimental unit. The phrase “randomly assigned” is very important in this definition.

 Experiment:
 It is a way of getting an answer to a question which the experimenter wants to know.
 A device or a means of getting an answer to the problem under consideration.
 It broadly classified into two categories:
1. Absolute: Designed to calculate certain measures of relationships. Eg. Correlation between
age and height of a group of people, average marks of a class etc.
2. Comparative: Designed to compare the effect of two more characteristics. Eg. Comparison
of effect of two or medicines, fertilizers etc.
 Treatment:

Different objects or procedures which are to be compared in an experiment are called treatments.

 Yield:

The measurements of the variable under study on different experimental units are termed as yield.

 Sampling unit:

The object that is measured in an experiment is called the sampling unit. This may be different from the
experimental unit.

 Factor
 Control Factors (Xs):

We want to evaluate effects of these on the result of response. They can be quantitative (Examples:
temperature, pressure, time) or qualitative factors (Example: Steel, Aluminum, method A and method B
etc)

2|Page
A factor is a variable defining a categorization. A factor can be fixed or random in nature.

 A factor is termed as a fixed factor if all the levels of interest are included in the experiment.
 A factor is termed as a random factor if all the levels of interest are not included in the
experiment and those that are can be considered to be randomly chosen from all the levels of
interest.
 Responses (Ys):

This is what we want optimize. There can be more than one Ys.

Example: Fuel consumption, strength, size, yield, rejection etc…

 Level:
 The settings of factors during experiment.
 Can be values or fixed points. Examples : Temperature 50 OC and 75 OC. Plastic and wood
 Replication:

It is the repetition of the experimental situation by replicating the experimental unit.

 Experimental error:

The unexplained random part of the variation in any experiment is termed as experimental error. An
estimate of experimental error can be obtained by replication.

 Treatment design:

A treatment design is the manner in which the levels of treatments are arranged in an experiment.

Example:

Suppose some varieties of fish food are to be investigated on some species of fishes. The food is placed in
the water tanks containing the fishes. The response is the increase in the weight of fish. The experimental
unit is the tank, as the treatment is applied to the tank, not to the fish. Note that if the experimenter had
taken the fish in hand and placed the food in the mouth of fish, then the fish would have been the
experimental unit as long as each of the fish got an independent scoop of food.

 Precision:

3|Page
The degree of uncertainty with which the inferences are drown from the results of experiments is called
precision of the experiment. It is measured as the reciprocal of the variance of mean

Where r is treatment replication and 𝜎 2 is the variance.

B. Statistically Designed experiment- Design of Experiments (DOE)

3.4. Introduction
 One of the main objectives of designing an experiment is how to verify the hypothesis in an
efficient and economical way. In the contest of the null hypothesis of equality of several means of
normal populations having the same variances, the analysis of variance technique can be used. Note
that such techniques are based on certain statistical assumptions. If these assumptions are violated, the
outcome of the test of a hypothesis then may also be faulty and the analysis of data may be
meaningless. So the main question is how to obtain the data such that the assumptions are met and the
data is readily available for the application of tools like analysis of variance. The designing of such a
mechanism to obtain such data is achieved by the design of the experiment. After obtaining the
sufficient experimental unit, the treatments are allocated to the experimental units in a random
fashion.
 Design of experiment provides a method by which the treatments are placed at random on the
experimental units in such a way that the responses are estimated with the utmost precision possible.

A DOE is a planned of tests on the response(s) with one or more input(s) each at two or more settings
(levels) which will:

 Determine if any factor or combination of factors is significant

 Define prediction equation
 Allow efficient optimization

4|Page
3.4.1. Objectives of DOE
The objective of DOE includes:

 Learning the most from as few runs as possible--- Efficiency

 Identify which factors affect mean, variation both or none
 Screening a large number of factors down to the vital few.
 Modeling the process with a prediction equation
 Optimizing the factor level for desired response
 Validating the results through confirmation.

3.4.2. Steps in DOE

 Planning the Experiment
1. Define project, set up a team and state the objectives.
2. Establish control factors and levels
3. Select response variables and measurement methods.
4. Choose the experimental design and apply replication, randomization and local control
principles
 Performing the Experiment
5. Conduct the experiment.
 Analyzing and verifying the experimental results
6. Analyze the data, determine the optimum levels for control factors and predict performance under
levels
7. Conduct the verification (also called confirmation) experiment and plan future actions.

3.4.3. Principles of experimental design:

There are three basic principles of design which were developed by Sir Ronald A. Fisher.

1. Randomization
2. Replication
3. Local control

5|Page
 3.1.1.1. Randomization
 The principle of randomization involves the allocation of treatment to experimental units at
random to avoid any bias in the experiment resulting from the influence of some extraneous
unknown factor that may affect the experiment. In the development of analysis of variance, we
assume that the errors are random and independent. In turn, the observations also become
random. The principle of randomization ensures this.
 The random assignment of experimental units to treatments results in the following outcomes.
 It eliminates systematic bias.
 It is needed to obtain a representative sample from the population.
 It helps in distributing the unknown variation due to confounded variables throughout
the experiment and breaks the confounding influence.
 Randomization forms a basis of a valid experiment but replication is also needed for the validity
of the experiment.
 If the randomization process is such that every experimental unit has an equal chance of receiving
each treatment, it is called complete randomization.

3.1.1.2. Replication:
In the replication principle, any treatment is repeated a number of times to obtain a valid and more
reliable estimate than which is possible with one observation only. Replication provides an efficient way
of increasing the precision of an experiment. The precision increases with the increase in the number of
observations. Replication provides more observations when the same treatment is used, so it increases
precision. For example, if the variance of x is 2 than variance of the sample mean ̅ based on n
observation is 2 /n. As n increases, variance ( ̅ ) decreases.

6|Page
 3.1.1.3. Blocking or Local control (error control)
This is the process of grouping together experimental units that are similar and assigning all treatments
into each group or block separately and independently. This allows the measurement of variation among
blocks which can be removed from the experimental error. Blocking is therefore one of the measure for
reducing or minimizing experimental error.

 The replication is used with local control to reduce the experimental error. For example, if the
experimental units are divided into different groups such that they are homogeneous within the
blocks, then the variation among the blocks is eliminated and ideally, the error component will
contain the variation due to the treatments only. This will, in turn, increase the efficiency.

3.5. Completely Randomized Design

 The Completely Randomized Design (CRD) is the simplest design based on principle of
replication and randomization.
 All treatments are randomly allocated to the experimental units.
 It allows every experimental unit to have an equal probability of receive a treatment

3.5.1. Layout
Let us suppose that we have p treatments and 𝑖𝑡ℎ treatment is replicated 𝑛𝑖 times (𝑖 = 1,2, … , 𝑝).

7|Page
 C. Types of Design of Experimental Designs
3.6. Simple design (One factor at a time)
 Start with a configuration and vary one factor at a time
 Given k factors and the i-th factor having ni levels
 The required number of experiments

𝑛 ∑ 𝑛

Example: For an experiment that has 3 factors to be considered, and the levels of each factor are k=3,
{n1=3, n2=4, n3=2}

Hence the number of experimental runs to be conducted is calculated using the following formula.

𝑛 ∑ 𝑛

Therefore:

k=3, {n1=3, n2=4, n3=2}

n = 1+ (2 + 3 + 1) = 7

3.7. Factorial Design Experiment

 It facilitates a more rational optimization
 It is always balanced (Equal number of -1 and +1)
 Factorial design is always orthogonal, ∑ =0. The sum of all dot product is always zero.
 It is the most efficient experimental design for two level factors.
 It will give average effects with and the need for replication
 It will account for interaction effects

Response may also depend on certain combination of factors. An interaction is present when the
effect of a factor on the response changes because of different settings of the levels of other factors.

Example: y = f (a.b)

Y= 10+ 2a+3b-4ab

8|Page
 Hence we can say that there is interaction between factor A and factor B.

3.6.1. Full Factorial Design (FFD)

When we run trail with each possible treatment, it is a full factorial experiment. The number of runs in
full factorial designs increases by geometric progression with ratio of numbers of levels of each factor.

Given k factors and the i-th factor having ni levels, The required number of experiment

𝑛 ∏𝑛

Example:

k=3, {n1=3, n2=4, n3=2}

n = 3×4×2 = 24

 It is efficient for two level experiment

 For two factor and two level, the number of experiment is calculated as 2k
 It is two level factorial designs.
 Number of Experiments= 2K, where K number of factors.

Example 1-: For one factor. The numbers of experiments are:-

9|Page
21= 2

Number of Runs tc Factor A

1 1 -1
2 2 +1

Example 2-: For two factor, the numbers of experiments are:-

22= 4

Number of tc Factor (s) Interaction

Runs A B AB
1 1 -1 -1 +1
2 (a) +1 -1 -1
3 (b) -1 +1 -1
4 (ab) +1 +1 +1

Full Factorial Design (FFD) and Level of Interaction

Number Main Order of interaction

of Factors 2 3 4 5 6 7 8 9 10
Factors
2 2 1
3 3 3 1
4 4 6 4 1
5 5 10 10 5 1
6 6 15 20 15 6 1
7 7 21 35 35 21 7 1
8 8 28 56 70 56 28 8 1
9 9 36 84 126 84 36 9 1
10 10 44

10 | P a g e
Example 1: For three factors (ABC); the levels of interaction

2nd level interaction: AB, BC, AC

3rd level interaction: ABC

Example 2: For four factors (ABCD); the levels of interaction

2nd level interaction: AB, AC, AD, BC, BD, CD

3rd level interaction: ABC, ACD, BCD, ABD

4th level interaction: ABCD

Limitation of FFD

 Full factorial design (FFD) results in a huge number of experiments, it may be not possible to run all.

Example: For an experiment having ten factors at two levels, the number of runs will be 210 = 1024

3.6.2. Fractional factorial design

 When full factorial design results in a huge number of experiments, it may be not possible to run all
 It is developed to reduce the number of runs or trails.
 Only selected treatment combinations are tried instead of all combinations.
 It Use subsets of levels of factors and the possible combinations of these
 Given k factors and the i-th factor having ni levels, and selected subsets of levels mi ≤ ni .
 The required number of experiments

𝑛 ∏

Example:

k=3, {n1=3, n2=4, n3=2}, but use {m1=2, m2=2, m3=1} • n = 2×2×1 = 4

Note: For factors having two level, the number of experiments on Fractional Factorial Design

2k-q.

11 | P a g e
Where k is number of factors and q indicates fractions of experiments.

Example: when q=1 it indicates ½ fraction and when q=2 it indicates ¼ fraction

Example 1-: We need to construct design matrix for 6 factors to run 16 experiments

So, Full factorial design: 2k = 26 =64. Hence 16 experiments are 1/4th of the total experiment. That is 2k-q
= 26-2 = 24 =16

Example 1-: We need to construct design matrix for 4 factors to run 8 experiments.

The total number of runs on full factorial design is: 24 =16. But if we need 8 trials it becomes 24-1 =23 =8.

Hence to formulate the matrix of design of experiment, WE have to follow the following procedures.

1st: we have to construct the matrix of design of experiment: 23 =8, for factors A,B,C

Run Factors Interactions

A B C AB AC BC ABC
1 - - - + + + _
2 - - + + _ _ +
3 - + - _ + _ +
4 - + + _ _ + _
5 + _ _ _ _ + +
6 + _ + _ + _ _
7 + + _ + _ _ _
8 + + + + + + +

2nd: Let we decide where to put the 4th factor, D. So. We have to place on the highest order of interaction
(D= ABC).

12 | P a g e
Hence, the experimental design becomes

Run Factors Interactions

BCD ACD ABD CD BD AD D
A B C AB AC BC ABC
1 - - - + + + _
2 - - + + _ _ +
3 - + - _ + _ +
4 - + + _ _ + _
5 + _ _ _ _ + +
6 + _ + _ + _ _
7 + + _ + _ _ _
8 + + + + + + +

Hence: in this design matrix:

I. The Main factors A, B,C, and D are confounded ( aliased) with three factors ( 3rd level of
interactions);
 A confounded with BCD
 B confounded with ACD
 C confounded with ABD
 D confounded with ABC
II. The second order interactions are confounded (aliased)
 AB is confounded with CD
 AC is confounded with BD
 BC is confounded with AD
This means that we are not able to identify whether the change on the response is because of the
interaction between A and B or the interaction between C and D and so on.

13 | P a g e
Note:

 The interaction level beyond second order is not considered and it is too rare.
 In the design of an experiment the main parameters (factors) can be aliased (confounded) with 3 rd
level and above interactions. But, the main factors should not be confounded with 2nd order
interactions. This is because we cannot determine the change on the response is either by the main
factor or the interaction between the two factors.
 Two factor interaction can be confounded with two factor interaction

3.6.2.1. Resolution of an Experiment

 Based on the extent of confounding, experiments are denoted by their resolution codes.
 Resolution design indicates its power and ability to separately estimate effects of factors and
interactions
 The resolution of a design is one more than the smallest order interaction that some main effects are
confounded with.

Example:

1. If some main effects are aliased or confounded with four factor or 4 level interactions, the
resolution of the experiment would be 4+1= 5
2. If some main effects are confounded with 2 factor, the resolution would be 2+1=3

Note:

1. Resolution 2 design don’t exist as it will imply confounding of main effects

2. Resolution 3, 4 and 5 are popular
3. Resolution codes are written in Roman number, Example Resolution III, IV and V.

Resolution III- Designs

 Used for screening out insignificant factors when number of factors in an experiment is
large typically more than 5.
 Here main effects are aliased with 2 factor interactions.

14 | P a g e
Resolution IV- Designs

 No main effects are aliased with ant other main effects or 2-factor interactions. But some 2 factor
interactions are aliased with other 2 factor interactions and main effects are aliased with three
factors interactions.

Resolution V- Designs

 Two factor interactions are aliased with three factor interactions and
 Main effects are aliased with four factor interactions. However, no main factors or two factor
interactions aliased with other main effects or two two factor interactions.
 This is applied when we need in two factor interaction and want to develop a regression model.

3.8. Surface Response Design

15 | P a g e