13/09/2018

TECNOLOGICO DE MONTERREY OUTLINE

• Basic definitions and principles
IN5058 – DESIGN AND ANALYSIS OF
EXPERIMENTS • The advantage of factorials

• The two-factor factorial design

FACTORIAL DESIGNS • The general factorial design

• Fitting response curves and surfaces

Rodrigo E. Peimbert-Garcia, Ph.D.
SNI Candidate – Master Black Belt • Blocking in a factorial design

Montgomery, D. C. (2013). Design and Analysis of Experiments. John Wiley & Sons. Monterrey, Mexico 2

BASIC PRINCIPLES BASIC PRINCIPLES

• Many experiments involve the study of the effects of two • The effect of a factor is defined to be the change in
or more factors. response produced by a change in the level of the factor.

• In general, factorial designs are most efficient for this type • This is frequently called a main effect because it refers to
of experiment.
the primary factors of interest in the experiment. For
example, consider the simple experiment in the Figure
• By a factorial design, we mean that in each complete trial below.
or replicate of the experiment all possible combinations
of the levels of the factors are investigated.

• For example, if there are a levels of factor A and b levels

of factor B, each replicate contains all ab treatment
combinations.
3 4

BASIC PRINCIPLES BASIC PRINCIPLES

• In some experiments, we may find that the difference in • Clearly, the interaction is large in this experiment. These
response between the levels of one factor is not the same ideas may be illustrated below.
at all levels of the other factors.

• When this occurs, there is an interaction between the

factors.

5 6

1
 13/09/2018

BASIC PRINCIPLES BASIC PRINCIPLES

• Two-factor interaction graphs such as these are frequently • There is another way to illustrate the concept of
very useful in interpreting significant interactions and in interaction. Suppose that both of our design factors are
reporting results to non-statistically trained personnel. quantitative (such as temperature, pressure, time, etc.).

• Then a regression model representation of the two-factor

• However, they should not be utilized as the sole factorial experiment could be written as:
technique of data analysis because their interpretation is
subjective and their appearance is often misleading.

• The variables x1 and x2 are coded from -1 to 1.

7 8

BASIC PRINCIPLES BASIC PRINCIPLES

• The estimates of and are one-half the value of the • The parameter is estimated by the average of all four
corresponding main effect; therefore, = 21/2 and responses, or (20 + 40 + 30 + 52)/4= 35.5. Therefore,
/ and. the fitted regression model is:

• The interaction effect in the first example is AB = 1: . . . .

• The parameter estimates obtained in the manner for the

factorial design with all factors at two levels (– and +) turn
out to be least squares estimates.
• so the value of interaction coefficient in the regression
model is 1/2= 0.5.
9 10

BASIC PRINCIPLES BASIC PRINCIPLES

• The interaction coefficient ( 0.5) is small relative to • A three-dimensional graph is called a response surface
the main effect coefficients and . plot.

• We will take this to mean that interaction is small and can

be ignored.

. . .

11 12

2
 13/09/2018

BASIC PRINCIPLES BASIC PRINCIPLES

• Now suppose that the interaction contribution to this • Interaction is a form of curvature in the underlying
experiment was not negligible; that is, the coefficient response surface model for the experiment.
was not small.

• The following figures below show the surface and contour

plots for the model:

. . .

13 14

BASIC PRINCIPLES BASIC PRINCIPLES

• Generally, when an interaction is large, the corresponding • This value is very small, and we are tempted to conclude
main effects have little practical meaning. that there is no effect due to A.

• For the experiment described before, we would estimate • However, when we examine the effects of A at different
the main effect of A to be: levels of factor B, we see that this is not the case. Factor A
has an effect, but it depends on the level of factor B.

• That is, knowledge of the AB interaction is more useful

than knowledge of the main effect.

• A significant interaction will often mask the significance of

main effects.
15 16

ADVANTAGE OF FACTORIALS ADVANTAGE OF FACTORIALS

• Suppose we have two factors A and B, each at two levels. • Because experimental error is present, it is desirable to
We denote the levels of the factors by A-, A+, B-, and B+. take two observations, say, at each treatment
combination and estimate the effects of the factors using
• The effect of changing factor A is given by average responses. Thus, a total of six observations are
A+B- – A-B-, and the effect of changing factor B is given by required.
A-B+ – A-B-.
• Now suppose interaction is present. If the one-factor-at-a-
time design indicated that A-B+ and A+B- gave better
responses than A-B-, a logical conclusion would be that
A+B+ would be even better. However, if interaction is
present, this conclusion may be seriously in error.

17 18

3
 13/09/2018

TWO-FACTOR FACTORIAL DESIGN TWO-FACTOR FACTORIAL DESIGN

• The simplest types of factorial designs involve only two EXAMPLE

factors or sets of treatments. • An engineer is designing a battery for use in a device that will be
subjected to some extreme variations in temperature.

• There are a levels of factor A and b levels of factor B, and

• The only design parameter that he can select at this point is the plate
these are arranged in a factorial design; that is, each material for the battery, and he has three possible choices.
replicate of the experiment contains all ab treatment
combinations. • When the device is manufactured and is shipped to the field, the
engineer has no control over the temperature extremes that the
device will encounter, and he knows from experience that
• In general, there are n replicates. temperature will probably affect the effective battery life.

19 20

TWO-FACTOR FACTORIAL DESIGN TWO-FACTOR FACTORIAL DESIGN

• However, temperature can be controlled in the product development • Life (in hours) Data for the Battery Design Example:
laboratory for the purposes of a test.

• The engineer decides to test all three plate materials at three

temperature levels — 15, 70, and 125°F — because these temperature
levels are consistent with the product end-use environment.

• Because there are two factors at three levels, this design is

sometimes called a 32 factorial design.

• Four batteries are tested at each combination of plate material and • The engineer wants to answer:
temperature, and all 36 tests are run in random order. 1. What effects do material type & temperature have on the life of the battery?
2. Is there a choice of material that would give uniformly long life regardless of
temperature?
21 22

TWO-FACTOR FACTORIAL DESIGN TWO-FACTOR FACTORIAL DESIGN

• Let be the observed response when factor A is at the • In general, a two-factor factorial experiment will appear
ith level (i = 1, 2, . . . , a) and factor B is at the jth level ( j = as in the following Table.
1, 2, . . . , b) for the kth replicate (k = 1, 2, . . . , n).

• The order in which the abn observations are taken is

selected at random so that this design is a completely
randomized design.

23 24

4
 13/09/2018

TWO-FACTOR FACTORIAL DESIGN TWO-FACTOR FACTORIAL DESIGN

• The observations in a factorial experiment can be • Where:

described by a model. – µ is the overall mean effect,

• There are several ways to write the model for a factorial – is the effect of the ith level of the row factor A,
experiment. The effects model is:
– is the effect of the jth level of column factor B,
, ,…
, ,…
, ,… – is the interaction between and ,

– is a random error component.

25 26

TWO-FACTOR FACTORIAL DESIGN TWO-FACTOR FACTORIAL DESIGN

• Both factors are assumed to be fixed, and the treatment • We could also use a regression model.
effects are defined as deviations from the overall mean,
so ∑ and ∑ . • Regression models are particularly useful when one or
more of the factors in the experiment are quantitative.
• Similarly, the interaction effects are fixed and are defined
such that ∑ =∑ . • We will commonly use the effects model.

• Because there are n replicates of the experiment, there • In the two-factor factorial, both row and column factors
are abn total observations. (or treatments), A and B, are of equal interest.

27 28

TWO-FACTOR FACTORIAL DESIGN TWO-FACTOR FACTORIAL DESIGN

• Specifically, we are interested in testing hypotheses about • We are also interested in determining whether row and
the equality of row treatment effects, say: column treatments interact. Thus, we also wish to test:

: ⋯
: ,
:
:
• and the equality of column treatment effects, say:
• We now discuss how these hypotheses are tested using a
: ⋯ two-factor analysis of variance.

29 30

5
 13/09/2018

TWO-FACTOR FACTORIAL DESIGN TWO-FACTOR FACTORIAL DESIGN

• Let yi.. denote the total of all observations under the ith • Expressed mathematically:
level of factor A, y.j. denote the total of all observations
..
under the jth level of factor B, yij. denote the total of all .. .. , ,…,
observations in the ijth cell, and y... denote the grand total
of all the observations.
. .
.. .. , ,…,
• Define .. , . . , . , and … as the corresponding row,
column, cell, and grand averages. . , ,…,
. .
, ,…,

...
... ...
31 32

TWO-FACTOR FACTORIAL DESIGN TWO-FACTOR FACTORIAL DESIGN

• The total corrected sum of squares may be written as: • Notice that the total sum of squares has been partitioned
into: , , , .
…

• This is the fundamental ANOVA equation for the two-

factor factorial.
.. … .. … . .. .. … .

• There must be at least two replicates (n≥ 2) to obtain an

error sum of squares.
.. … .. … . .. .. ...

33 34

TWO-FACTOR FACTORIAL DESIGN TWO-FACTOR FACTORIAL DESIGN

• The number of degrees of freedom associated with each • Each sum of squares divided by its degrees of freedom is a
sum of squares is: mean square. The expected values of the mean squares
are: ∑
Effect Degrees of Freedom
A a-1
B b-1
AB interaction (a - 1)(b - 1) ∑
Error ab(n - 1)
Tota abn - 1
∑ ∑

35 36

6
 13/09/2018

TWO-FACTOR FACTORIAL DESIGN TWO-FACTOR FACTORIAL DESIGN

• If we assume that the effects model is adequate and that • One could write out the individual elements of the
the error terms ijk are normally and independently ANOVA identity:
distributed with constant variance σ2, then each of the
ratios of mean squares , , and is distributed … .. … . . … . .. .. … .

as F with a - 1, b - 1, and (a - 1)(b - 1) numerator degrees

of freedom, respectively, and ab(n - 1) denominator • The total sum of squares is computed as usual by:
degrees of freedom, and the critical region would be the
upper tail of the F distribution.
…

37 38

TWO-FACTOR FACTORIAL DESIGN TWO-FACTOR FACTORIAL DESIGN

• The sum of squares for the main effects are • This sum of squares also contains and .
Therefore, the second step is to compute as
…
..
…
.

…
.. • The second step is to compute SSAB:

• It is convenient to obtain the in two stages. First we

compute the sum of squares between the ab cell totals,
which is called the sum of squares due to “subtotals”:
39 40

TWO-FACTOR FACTORIAL DESIGN

• We may compute by the subtraction as:

• Or:

41