Chapter 9

(AST301) Design and Analysis of Experiments II

Md Rasel Biswas

Md Rasel Biswas Chapter 9 1 / 50

Lecture Outline

1 9. Additional Design and Analysis Topics for Factorial and Fractional

Factorial Designs
9.1 The 3𝑘 Factorial Design
9.2 Confounding in the 3𝑘 Factorial Design
9.3 Fractions of 3𝑘 factorial design

Md Rasel Biswas Chapter 9 2 / 50

 Section 1

9. Additional Design and Analysis Topics for Factorial

and Fractional Factorial Designs

Md Rasel Biswas Chapter 9 3 / 50

 Subsection 1

9.1 The 3𝑘 Factorial Design

Md Rasel Biswas Chapter 9 4 / 50

9.1 The 3𝑘 Factorial Design

The three-level design is written as a 3𝑘 factorial design. It means

that 𝑘 factors are considered, each at 3 levels.
These are (usually) referred to as low, intermediate and high levels.
These levels are numerically expressed as 0, 1, and 2.
One could have considered the digits -1, 0, and +1, but this may be
confusing with respect to the 2-level designs since 0 is reserved for
center points. Therefore, we will use the 0, 1, 2 scheme.
A third level for a continuous factor facilitates investigation of a
quadratic relationship between the response and each of the factors.

Md Rasel Biswas Chapter 9 5 / 50

9.1 The 3𝑘 Factorial Design

Each treatment combination in the 3𝑘 design is denoted by 𝑘 digits,

where the first digit indicates the level of factor 𝐴, the second digit
indicates the level of factor 𝐵, …, and the 𝑘th digit indicates the level
of factor 𝐾 .
For example, in a 32 design, 00 denotes the treatment combination
corresponding to 𝐴 and 𝐵 both at the low level, and 01 denotes the
treatment combination corresponding to 𝐴 at the low level and 𝐵 at
the intermediate level.

Md Rasel Biswas Chapter 9 6 / 50

9.1 The 3𝑘 Factorial Design

In the 3𝑘 system of designs, when the factors are quantitative, we

often denote the low, intermediate, and high levels by −1, 0, and +1,
respectively. This facilitates fitting a regression model relating the
response to the factor levels.
Regression model for analyzing 32 factorial design

𝑦 = 𝛽0 + 𝛽1 𝑥1 + 𝛽2 𝑥2 + 𝛽12 𝑥1 𝑥2 + 𝛽11 𝑥21 + 𝛽22 𝑥2

Md Rasel Biswas Chapter 9 7 / 50

9.1 The 3𝑘 Factorial Design

The 3𝑘 design is useful when curvature of the response surface is

concerned, however, there are more useful methods for examining
curvature of response surface than the 3𝑘 design, which are
▶ response surface design
▶ 2𝑘 design with center points (central composite design)

Md Rasel Biswas Chapter 9 8 / 50

 Subsection 2

The 32 Design

Md Rasel Biswas Chapter 9 9 / 50

The 32 Design

The 32 design has two factors, each at three levels. There are 32 = 9
treatment combinations and there are 8 degrees of freedom between these
treatment combinations:

00 01 02 10 11 12 20 21 22

Main effect of 𝐴: 2 degrees of freedom

Main effect of 𝐵: 2 degrees of freedom
Interaction 𝐴𝐵: 4 degrees of freedom
For 𝑛 replicates, there are 32 𝑛 − 1 total degrees of freedom and 32 (𝑛 − 1)
degrees of freedom for error
The sums of squares for the effects 𝐴, 𝐵, and 𝐴𝐵 can be computed by
usual methods discussed in Chapter 5.

Md Rasel Biswas Chapter 9 10 / 50

The 32 Design
The two degrees of freedom of each main effect can be represented by a
linear and a quadratic component each with a single degrees of freedom.
Suppose that both factors 𝐴 and 𝐵 are quantitative
The two-factor interaction 𝐴𝐵 can be partitioned into two ways:
1 Dividing 𝐴𝐵 into four single-degrees-of-freedom components

𝐴𝐵𝐿×𝐿 , 𝐴𝐵𝐿×𝑄 , 𝐴𝐵𝑄×𝐿 , 𝐴𝐵𝑄×𝑄

This can be done by fitting the terms 𝛽12 𝑥1 𝑥2 ,

𝛽122 𝑥1 𝑥2 2 , 2 , 𝛽112 𝑥21 𝑥2 and,𝛽1122 𝑥2 1 𝑥22 respectively.
2 The second method is based on the orthogonal Latin squares. Two
Latin squares are said to be orthogonal if one square is superimposed
on the other, produce all possible ordered pairs of symbols exactly
once.

Md Rasel Biswas Chapter 9 11 / 50

 Subsection 3

Example 5.5

Md Rasel Biswas Chapter 9 12 / 50

Example 5.5
The effective life of a cutting tool installed in a numerically controlled
machine is thought to be affected by the cutting speed and the tool angle.
Three speeds and three angles are selected, and a 32 factorial experiment
with two replicates is performed. The coded data are shown below

Md Rasel Biswas Chapter 9 13 / 50

Example 5.5

Since the factors are quantitative, and both factors have three levels, a
second-order model such as
𝑦 = 𝛽0 + 𝛽1 𝑥1 + 𝛽2 𝑥2 + 𝛽12 𝑥1 𝑥2 + 𝛽11 𝑥2 1 + 𝛽22 𝑥22 + 𝛽122 𝑥1 𝑥22 +
𝛽112 𝑥2 1 𝑥2 + 𝛽1122 𝑥21 𝑥22

can be fit to the data.

From the data we obtain sum of squares as

𝐴𝐵𝐿×𝐿 = 8, 𝐴𝐵𝐿×𝑄 = 42.67, 𝐴𝐵𝑄×𝐿 = 2.67 and 𝐴𝐵𝑄×𝑄 = 8

That is 𝑆𝑆 𝐴𝐵 = 8 + 42.67 + 2.67 + 8 = 61.34.

Md Rasel Biswas Chapter 9 14 / 50

Example 5.5

The second method is based on orthogonal Latin squares. This method

does not require that the factors be quantitative, and it is usually
associated with the case where all factors are qualitative.
The two factors 𝐴 and 𝐵 correspond to the rows and columns,
respectively, of a 3 × 3 Latin square.
For the above example, two particular 3 × 3 Latin squares are shown
superimposed on the cell totals below:

Md Rasel Biswas Chapter 9 15 / 50

Example 5.5
Treatment combination totals from Example 5.5 with two orthogonal Latin
squares superimposed

So, SS from first table is known as 𝐴𝐵 component of the interaction, and

SS from the second table is known as 𝐴𝐵2 component of the interaction
Md Rasel Biswas Chapter 9 16 / 50
Example 5.5

Total of the letters from the first table: 𝑄 = 18, 𝑅 = −2, 𝑆 = 8. Sum of
squares between these totals is 33.3 with 2 df.
Total of the letters from the second table: 𝑄 = 0, 𝑅 = 6, 𝑆 = 18. Sum
of squares between these totals is 28 with 2 df.
𝑆𝑆𝐴𝐵 = 33.3 + 28 = 61.3 with 4 df
It can also be shown that 𝐴𝐵2 = (𝐴𝐵2 )2 = 𝐴2 𝐵4 = 𝐴2 𝐵

Md Rasel Biswas Chapter 9 17 / 50

Example 5.5
There is another way of calculating the 𝐴𝐵 and 𝐴𝐵2 components of the
interaction:
Add the data by diagonals downward from the left to right and the totals
are
−3 + 4 − 1 = 0 ⎫
}
−3 + 10 − 1 = 6
⎬ ⇒ sum of squares of these totals = 28 = 𝐴𝐵2
5 + 2 + 11 = 18 }
⎭

Add the data by diagonals downward from the right to left and the totals
are
5+4−1 =8 ⎫
}
−3 + 2 − 1 = −2 ⇒ sum of squares of these totals = 33.3 = 𝐴𝐵
⎬
}
−3 + 11 + 10 = 18 ⎭

Md Rasel Biswas Chapter 9 18 / 50

Example 5.5

Yates called these components of interaction as the 𝐼 and 𝐽 components

of interaction:

𝐼(𝐴𝐵) =𝐴𝐵2
𝐽 (𝐴𝐵) =𝐴𝐵

Md Rasel Biswas Chapter 9 19 / 50

 Subsection 4

The 33 design

Md Rasel Biswas Chapter 9 20 / 50

The 33 design

The 33 design has three factors (𝐴, 𝐵, 𝐶), each has three levels and there
33 = 27 treatment combinations
Among 26 degrees of freedom, 6 for the main effects, 12 for the two-factor
interactions, and 8 for the three-factor interaction
For 𝑛 replicates, there are 33 𝑛 − 1 total degrees of freedom and 33 (𝑛 − 1)
degrees of freedom for error

Md Rasel Biswas Chapter 9 21 / 50

The 33 design

Sums of squares can be calculated using the standard methods, in addition

to that, if the factors are quantitative
Main effects can be partitioned into linear (𝐿) and quadratic (𝑄)
components, each with one degrees of freedom
Two-factor interactions can be partitioned into components
corresponding to 𝐿 × 𝐿, 𝐿 × 𝑄, 𝑄 × 𝐿, 𝑄 × 𝑄
Three-factor interaction can be partitioned into components
corresponding to 𝐿 × 𝐿 × 𝐿, 𝐿 × 𝐿 × 𝑄, etc.

Md Rasel Biswas Chapter 9 22 / 50

The 33 design

The sum of squares corresponding to the two-factor interaction has 𝐼 and

𝐽 components, e.g. 𝐼(𝐴𝐵) = 𝐴𝐵2 and 𝐽 (𝐴𝐵) = 𝐴𝐵
The sum of squares corresponding to a three-factor interaction has 𝑊 , 𝑋 ,
𝑌 , and 𝑍 components, e.g.

𝑊 (𝐴𝐵𝐶) = 𝐴𝐵2 𝐶 2 , 𝑋(𝐴𝐵𝐶) = 𝐴𝐵𝐶 2 ,

𝑌 (𝐴𝐵𝐶) = 𝐴𝐵2 𝐶, 𝑍(𝐴𝐵𝐶) = 𝐴𝐵𝐶

Md Rasel Biswas Chapter 9 23 / 50

 Subsection 5

The general 3𝑘 design

Md Rasel Biswas Chapter 9 24 / 50

The general 3𝑘 design

The 3𝑘 factorial design has 𝑘 factors, each has 3 levels, so there are 3𝑘
treatment combinations with 3𝑘 − 1 degrees of freedom between them
There are
𝑘 main effects, each has 2 degrees of freedom
(𝑘2) two-factor interactions, each has 22 = 4 degrees of freedom
one 𝑘-factor interaction, which has 2𝑘 degrees of freedom
If there are 𝑛 replications then total degrees of freedom will be 𝑛3𝑘 − 1
and error degrees of freedom is 3𝑘 (𝑛 − 1)
Sums of squares for different effects are computed by usual method of
factorial design

Md Rasel Biswas Chapter 9 25 / 50

The general 3𝑘 design
Any ℎ-factor interaction has 2ℎ−1 orthogonal components, each of which
has 2 degrees of freedom
E.g. The four-factor interaction 𝐴𝐵𝐶𝐷 has 24−1 = 8 orthogonal
components, which are

𝐴𝐵𝐶𝐷2 𝐴𝐵𝐶 2 𝐷 𝐴𝐵2 𝐶𝐷 𝐴𝐵𝐶𝐷

𝐴𝐵𝐶 2 𝐷2 𝐴𝐵2 𝐶𝐷2 𝐴𝐵2 𝐶 2 𝐷 𝐴𝐵2 𝐶 2 𝐷2

Note that only exponent allowed for the first letter is 1 and if the exponent
of the first letter is not 1 then the entire expression must be squared, e.g.

𝐴2 𝐵𝐶𝐷 = (𝐴2 𝐵𝐶𝐷)2 = 𝐴4 𝐵2 𝐶 2 𝐷2 = 𝐴𝐵2 𝐶 2 𝐷2

These interaction components have no physical meaning but they are

useful in constructing more complex designs
Md Rasel Biswas Chapter 9 26 / 50
 Subsection 6

9.2 Confounding in the 3𝑘 Factorial Design

Md Rasel Biswas Chapter 9 27 / 50

9.2 Confounding in the 3𝑘 Factorial Design

Even when a single replicate of the 3𝑘 design is considered, the design

requires so many runs that it is unlikely that all 3𝑘 runs can be made
under uniform conditions.
Thus, confounding in blocks is often necessary. The 3𝑘 design may be
confounded in 3𝑝 incomplete blocks, where 𝑝 < 𝑘.
Thus, these designs may be confounded in three blocks, nine blocks, and
so on.

Md Rasel Biswas Chapter 9 28 / 50

The 3𝑘 factorial design in three blocks

The three blocks have two degrees of freedom, so there must be two
degrees of freedom confounded with blocks
In the 3𝑘 factorial design
each main effect has two degrees of freedom
each two-factor interaction has four degrees of freedom, which can be
decomposed into two components (e.g. 𝐴𝐵 and 𝐴𝐵2 ), each of which
has two degrees of freedom
each three-factor interaction has eight degrees of freedom, which can
be decomposed into four components (e.g. 𝐴𝐵𝐶 , 𝐴𝐵𝐶 2 , 𝐴𝐵2 𝐶 ,
and 𝐴𝐵2 𝐶 2 ), each of which has two degrees of freedom
and so on
So it is convenient to confound an interaction component with blocks

Md Rasel Biswas Chapter 9 29 / 50

The 3𝑘 factorial design in three blocks

The general procedure to construct a defining contrast

𝐿 = 𝛼1 𝑥1 + 𝛼2 𝑥2 + ⋯ + 𝛼𝑘 𝑥𝑘 ,

where
𝛼𝑖 represents the exponent on the 𝑖th factor in the effect to be
confounded
𝑥𝑖 is the level of the 𝑖th factor in a particular treatment combination,
𝑥𝑖 can take values either 0 or 1 or 2
For 3𝑘 design, 𝛼𝑖 = 0, 1, or 2 with first nonzero 𝛼𝑖 is unity
𝐿 can take on only the values 0, 1, or 2 (mod 3) and the treatment
combinations satisfying 𝐿 = 0 (mod 3) constitute the principal block,
which always contains the treatment combination 00 … 0

Md Rasel Biswas Chapter 9 30 / 50

The 3𝑘 factorial design in three blocks
Suppose goal is to construct a 32 factorial design in 3 blocks, one of
the components of 𝐴𝐵 interaction (𝐴𝐵 or 𝐴𝐵2 ) needs to be
confounded
If 𝐴𝐵2 is chosen for confounding with blocks, then the corresponding
defining contrast

𝐿 = 𝑥1 + 2𝑥2

The value of 𝐿 (mod 3) of each treatment combination

00 ∶𝐿 = 1(0) + 2(0) = 0 (mod3) 12 ∶ 𝐿 = 1(1) + 2(2) = 2 (mod3)

01 ∶𝐿 = 1(0) + 2(1) = 2 (mod3) 20 ∶ 𝐿 = 1(2) + 2(0) = 2 (mod3)
02 ∶𝐿 = 1(0) + 2(2) = 1 (mod3) 21 ∶ 𝐿 = 1(2) + 2(1) = 1 (mod3)
10 ∶𝐿 = 1(1) + 2(0) = 1 (mod3) 22 ∶ 𝐿 = 1(2) + 2(2) = 0 (mod3)
11 ∶𝐿 = 1(1) + 2(1) = 0 (mod3)

Md Rasel Biswas Chapter 9 31 / 50

The 3𝑘 factorial design in three blocks

Assignment of treatment combinations to blocks

Note that the treatment combinations of principal block form a group with
respect to addition (mod 3), i.e. addition of pair of treatment combination
will lead to a treatment combination of the principal block. E.g.

00 + 11 = 11 11 + 22 = 00 22 + 00 = 22

Md Rasel Biswas Chapter 9 32 / 50

The 3𝑘 factorial design in three blocks
Assignment of treatment combinations to blocks

Note that the treatment combinations in the other two blocks may be
generated by adding (mod 3) any element of the new block to the
elements of principal block. E.g.

Block 2 ∶ 00 + 02 = 02 11 + 02 = 10 22 + 02 = 21
Block 3 ∶ 00 + 20 = 20 11 + 20 = 01 22 + 20 = 12

Md Rasel Biswas Chapter 9 33 / 50

Example 9.2

The statistical analysis of the 32 design confounded in three blocks is

illustrated by using the following data

We find that

Md Rasel Biswas Chapter 9 34 / 50

Example 9.2
SS for main effects and interaction:

replication chara block

172 + 112 + 12 72
𝑆𝑆𝐴 ∶ − = 131.56
3 9
22 + 2 2 + 3 2 72
𝑆𝑆𝐵 ∶ − = 0.22
3 9
42 + 32 + 02 72
𝑆𝑆𝐴𝐵 ∶ − = 2.89
3 9
02 + 02 + 72 72
𝑆𝑆𝐴𝐵2 ∶ − = 10.89
3 9

Md Rasel Biswas Chapter 9 35 / 50

Example 9.2
2
The I or 𝐴𝐵 component of the 𝐴𝐵 interaction may be found by
computing the sum of squares between the left-to-right diagonal totals in
2
the above layout. 𝑆𝑆 Blocks is exactly equal to the 𝐴𝐵 component of
interaction.
The analysis of variance is shown below. Because there is only one
replicate, no formal tests can be performed. It is not a good idea to use
the AB component of interaction as an estimate of error.

Md Rasel Biswas Chapter 9 36 / 50

Example 9.2

Slightly complicated example: The 33 design confounded in three blocks

of nine runs each. Suppose the 𝐴𝐵2 𝐶 2 component of the three-factor
interaction will be confounded with blocks, the defining contrast is

𝐿 = 𝑥1 + 2𝑥2 + 2𝑥3

It is clear that the treatment combinations 000, 110 and 101 are elements
of principal block, other six elements of principal block can be obtained
from these elements:

110 + 110 = 220 101 + 101 = 202 110 + 101 = 211

220 + 101 = 021 202 + 220 = 122 202 + 110 = 012

Md Rasel Biswas Chapter 9 37 / 50

Example 9.2
Treatment combinations of the principal block of the 33 design with
𝐴𝐵2 𝐶 2 component confounded in three blocks.

Treatment combinations of other two blocks can easily be obtained by

adding a treatment combination (which is not in the principal block) to
each element of the principal block
Md Rasel Biswas Chapter 9 38 / 50
Example 9.2

Md Rasel Biswas Chapter 9 39 / 50

The 3𝑘 factorial design in nine blocks

If 3𝑘 design is confounded in 9 blocks then 8 degrees of freedom is

confounded with blocks.
We need to choose two interaction components (4 degrees of
freedom), which will result two more components will be confounded
(4 degrees of freedom)
E.g. if 𝑃 and 𝑄 are two components that are selected for
confounding then 𝑃 𝑄 and 𝑃 𝑄2 will also be confounded with blocks
There will be two defining contrasts

𝐿1 = 𝛼1 𝑥1 + 𝛼2 𝑥2 + ⋯ + 𝛼𝑘 𝑥𝑘
𝐿2 = 𝛽1 𝑥1 + 𝛽2 𝑥2 + ⋯ + 𝛽𝑘 𝑥𝑘

9 blocks can be constructed based on the values of 𝐿1 and 𝐿2 or

using group-theoretic property of the principal block

Md Rasel Biswas Chapter 9 40 / 50

The 3𝑘 factorial design in nine blocks

Consider the 34 factorial design confounded in nine blocks of nine runs

each.
2
Suppose we choose to confound 𝐴𝐵𝐶 and 𝐴𝐵 𝐷2 . Their generalized
interactions
2
(𝐴𝐵𝐶) (𝐴𝐵2 𝐷2 ) = 𝐴2 𝐵3 𝐶𝐷2 = (𝐴2 𝐵3 𝐶𝐷2 ) = 𝐴𝐶 2 𝐷
2 2 2
(𝐴𝐵𝐶) (𝐴𝐵2 𝐷2 ) = 𝐴3 𝐵5 𝐶𝐷4 = 𝐵2 𝐶𝐷 = (𝐵2 𝐶𝐷) = 𝐵𝐶 𝐷2

are also confounded with blocks. The defining contrasts for 𝐴𝐵𝐶 and
2
𝐴𝐵 𝐷2 are
𝐿1 = 𝑥1 + 𝑥2 + 𝑥3
𝐿2 = 𝑥1 + 2𝑥2 + 2𝑥4

Md Rasel Biswas Chapter 9 41 / 50

The 3𝑘 factorial design in nine blocks

Md Rasel Biswas Chapter 9 42 / 50

 Subsection 7

9.3 Fractions of 3𝑘 factorial design

Md Rasel Biswas Chapter 9 43 / 50

The one-third fraction of the 3𝑘 factorial design
The largest fraction of the 3𝑘 design is a one-third fraction contains
3𝑘−1 runs, which is known as 3𝑘−1 fractional design
To construct a 3𝑘−1 fractional factorial design, select an interaction
component (which has 2 degrees of freedom) and partition the full 3𝑘
runs into three blocks.
Each of the three resulting blocks is a 3𝑘−1 fractional design, and any
one of the blocks may be selected for use
If 𝐴𝐵𝛼2 𝐶 𝛼3 ⋯ 𝐾 𝛼𝑘 is the component of interaction used to define
the blocks, then the defining relation is

𝐼 = 𝐴𝐵𝛼2 𝐶 𝛼3 ⋯ 𝐾 𝛼𝑘

The alias structure for an effect can be obtained by multiplying the

effect by both 𝐼 and 𝐼 2 , e.g. alias structure of the main effect 𝐴 are
𝐴𝐼 and 𝐴𝐼 2 .
Md Rasel Biswas Chapter 9 44 / 50
The one-third fraction of the 3𝑘 factorial design

Example: To construct a one-third fraction of the 33 design, we may

consider any of the interaction components as defining relation.
Since there are four interaction components, which are 𝐴𝐵𝐶 , 𝐴𝐵𝐶 2 ,
𝐴𝐵2 𝐶 , 𝐴𝐵2 𝐶 2 , there will be 12 possible one-third fraction of the 33
design and the defining contrast to obtain the designs

𝑥1 + 𝛼2 𝑥2 + 𝛼3 𝑥3 = 𝑢 (mod 3)

where 𝛼 = 1, 2 and 𝑢 = 0, 1, or 2.
If we select the component 𝐴𝐵2 𝐶 2 , the 33−1 design will contain
exactly 9 treatment combinations that must satisfy

𝑥1 + 2𝑥2 + 2𝑥3 = 𝑢, (mod 3)

where 𝑢 = 0, 1, or 2.

Md Rasel Biswas Chapter 9 45 / 50

The one-third fraction of the 3𝑘 factorial design
The alias structure with the defining relation 𝐼 = 𝐴𝐵2 𝐶 2 :

The four effects that are actually estimated from the eight degrees of
freedom are: 𝐴 + 𝐵𝐶 + 𝐴𝐵𝐶 , 𝐵 + 𝐵𝐶 2 + 𝐴𝐵𝐶 2 , 𝐶 + 𝐴𝐵2 + 𝐴𝐵2 𝐶 ,
𝐴𝐵 + 𝐴𝐶 + 𝐵𝐶 2

Md Rasel Biswas Chapter 9 46 / 50

The one-third fraction of the 3𝑘 factorial design

The treatment combinations in a 3𝑘−1 design with the defining relation

𝐼 = 𝐴𝐵𝛼2 𝐶 𝛼3 ⋯ 𝐾 𝛼𝑘

can be constructed as:

Write down the 3𝑘−1 runs for a full factorial design with 𝑘 − 1 factors,
with the usual 0, 1, 2 notation.
The 𝑘th factor is introduced by equating its levels 𝑥𝑘 to the
appropriate component of the highest order interaction
𝐴𝐵𝛼2 ⋯ (𝐾 − 1)𝛼𝑘−1 through the relationship

𝑥𝑘 = 𝛽1 𝑥1 + ⋯ + 𝛽𝑘−1 𝑥𝑘−1 ,

where 𝛽𝑖 = (3 − 𝛼𝑘 )𝛼𝑖 (mod 3) for 1 ≤ 𝑖 ≤ 𝑘 − 1. This yields a

design of the highest possible resolution

Md Rasel Biswas Chapter 9 47 / 50

The one-third fraction of the 3𝑘 factorial design
Consider a one-third fraction of the 34 factorial design with 𝐼 = 𝐴𝐵2 𝐶𝐷,
i.e. 𝑥1 + 2𝑥2 + 𝑥3 + 𝑥4 = 𝑢 (mod 3).
First construct the 33 design in factor 𝐴, 𝐵, and 𝐶 :

Levels for the factor 𝐷 can be obtained from

𝑥4 = 𝛽1 𝑥1 + 𝛽2 𝑥2 + 𝛽3 𝑥3 , 𝛽𝑖 = (3 − 𝛼4 )𝛼𝑖 , 𝑖 = 1, 2, 3
Md Rasel Biswas Chapter 9 48 / 50
The one-third fraction of the 3𝑘 factorial design
Defining relation: 𝑥1 + 2𝑥2 + 𝑥3 + 𝑥4 = 𝑢 (mod 3)
⇒ 𝛼1 = 1, 𝛼2 = 2, 𝛼3 = 1, 𝛼4 = 1 Levels for factor D:
𝑥4 = 𝛽1 𝑥1 + 𝛽2 𝑥2 + 𝛽3 𝑥3 , where 𝛽1 = (3 − 𝛼4 )𝛼1 = 2, 𝛽2 = 1, and 𝛽3 = 2
⇒ 𝑥4 = 2𝑥1 + 𝑥2 + 2𝑥3

As an example, alias structure for the effect 𝐴 is

𝐴(𝐴𝐵2 𝐶𝐷) = 𝐴2 𝐵2 𝐶𝐷 = 𝐴𝐵𝐶 2 𝐷2 , 𝐴(𝐴𝐵2 𝐶𝐷)2 = 𝐵𝐶 2 𝐷2
Md Rasel Biswas Chapter 9 49 / 50
The general 3𝑘−𝑝 fractional factorial design

The 3𝑘−𝑝 fractional factorial design is th (1/3)𝑝 fraction of the 3𝑘

design for 𝑝 < 𝑘, where the fraction contains 3𝑘−𝑝 runs.
The construction of 3𝑘−𝑝 includes the selection of 𝑝 components of
interaction first, which is then used to partition 3𝑘 treatment
combinations into 3𝑝 blocks. Each block is a 3𝑘−𝑝 fractional factorial
design.
The defining relation 𝐼 of any fraction consists of 𝑝 initially chosen
components and their (3𝑝 − 2𝑝 − 1)/2 generalized interactions.
Alias structure of an effect can be obtained by multiplying the effect
by the defining relation 𝐼 and 𝐼 2
The 3𝑘−𝑝 fractional factorial design can also be obtained by writing
down the full 3𝑘−𝑝 design first in (𝑘 − 𝑝) factors and then levels of
additional 𝑝 factors are obtained from the corresponding defining
contrasts.

Md Rasel Biswas Chapter 9 50 / 50