Response surface designs

 Response Surface (method) objective: The experiment is designed

to allow us to estimate interaction and even quadratic effects, and
therefore give us an idea of the (local) shape of the response surface
we are investigating. For this reason, they are termed response
surface method (RSM) designs. RSM designs are used to:
o Find improved or optimal process settings
o Troubleshoot process problems and weak points
o Make a product or process more robust against external and
non-controllable influences. "Robust" means relatively
insensitive to these influences.

Respons Earlier, we described the response surface method (RSM) objective. Under
e surface some circumstances, a model involving only main effects and interactions
models may be appropriate to describe a response surface when
may
involve 1. Analysis of the results revealed no evidence of "pure quadratic"
just main curvature in the response of interest (i.e., the response at the center
effects approximately equals the average of the responses at the factorial
and runs).
interacti 2. The design matrix originally used included the limits of the factor
ons or settings available to run the process.
they may
also
have
quadrati
c and
possibly
cubic
terms to
account
for
curvatur
e

Equation In other circumstances, a complete description of the process behavior

s for might require a quadratic or cubic model:
quadrati
c and Quadratic
cubic y^=β0+β1x1+β2x2+β3x3+β12x1x2+β13x1x3+β23x2x3+β11x21+β22x22+β33x23
models
 Cubic
y^=quadratic
model+β123x1x2x3+β112x21x2+β113x21x3+β122x1x22+β133x1x23+β223x22x3+β
233x2x23+β111x31+β222x32+β333x33+

These are the full models, with all possible terms, rarely would all of the
terms be needed in an application.

Quadrati If the experimenter has defined factor limits appropriately and/or taken
c models advantage of all the tools available in multiple regression analysis
almost (transformations of responses and factors, for example), then finding an
always industrial process that requires a third-order model is highly unusual.
sufficient Therefore, we will only focus on designs that are useful for fitting
for quadratic models. As we will see, these designs often provide lack of fit
industria detection that will help determine when a higher-order model is needed.
l
applicati
ons

General Figures 3.9 to 3.12 identify the general quadratic surface types that an
quadrati investigator might encounter
c surface
types

FIGURE 3.9 A Response FIGURE 3.10 A Response

Surface "Peak" Surface "Hillside"

FIGURE 3.11 A Response FIGURE 3.12 A Response

Surface "Rising Ridge" Surface "Saddle"

Factor Levels for Higher-Order Designs

Possible Figures 3.13 through 3.15 illustrate possible behaviors of responses as

behavior functions of factor settings. In each case, assume the value of the response
s of increases from the bottom of the figure to the top and that the factor
response settings increase from left to right.
s as
functions
of factor
settings

FIGURE 3.13 FIGURE 3.14 FIGURE 3.15

Linear Function Quadratic Function Cubic Function

A two- If a response behaves as in Figure 3.13, the design matrix to quantify that
level behavior need only contain factors with two levels -- low and high. This
experime model is a basic assumption of simple two-level factorial and fractional
nt with factorial designs. If a response behaves as in Figure 3.14, the minimum
center number of levels required for a factor to quantify that behavior is three.
points One might logically assume that adding center points to a two-level design
can would satisfy that requirement, but the arrangement of the treatments in
detect, such a matrix confounds all quadratic effects with each other. While a two-
but not level design with center points cannot estimate individual pure quadratic
fit, effects, it can detect them effectively.
quadrati
c effects

Three- A solution to creating a design matrix that permits the estimation of simple
level curvature as shown in Figure 3.14 would be to use a three-level factorial
factorial design. Table 3.21 explores that possibility.
design

Four- Finally, in more complex cases such as illustrated in Figure 3.15, the
level design matrix must contain at least four levels of each factor to
factorial characterize the behavior of the response adequately.
design

3-level TABLE 3.21 Three-level Factorial Designs

factorial Number Treatment Combinations
designs of Factors 3k Factorial
can fit
quadrati
c models 2 9
but they 3 27
require 4 81
many
5 243
runs
when 6 729
there are
more
than 4
factors

Fraction Two-level factorial designs quickly become too large for practical
al application as the number of factors investigated increases. This problem
factorial was the motivation for creating 'fractional factorial' designs. Table 3.21
designs shows that the number of runs required for a 3k factorial becomes
created unacceptable even more quickly than for 2k designs. The last column in
to avoid Table 3.21 shows the number of terms present in a quadratic model for
such a each case.
large
number
of runs

Number With only a modest number of factors, the number of runs is very large,
of runs even an order of magnitude greater than the number of parameters to be
large estimated when k isn't small. For example, the absolute minimum number
even for of runs required to estimate all the terms present in a four-factor quadratic
modest model is 15: the intercept term, 4 main effects, 6 two-factor interactions,
number and 4 quadratic terms.
of
factors The corresponding 3k design for k = 4 requires 81 runs.

Complex Considering a fractional factorial at three levels is a logical step, given the
alias success of fractional designs when applied to two-level designs.
structure Unfortunately, the alias structure for the three-level fractional factorial
and lack designs is considerably more complex and harder to define than in the
of two-level case.
rotatabil
ity for 3- Additionally, the three-level factorial designs suffer a major flaw in their
level lack of 'rotatability.'
fractiona
l
factorial
designs

Rotatability of Designs

"Rotatab In a rotatable design, the variance of the predicted values of y is a function

ility" is a of the distance of a point from the center of the design and is not a
desirable function of the direction the point lies from the center. Before a study
property begins, little or no knowledge may exist about the region that contains the
not optimum response. Therefore, the experimental design matrix should not
present bias an investigation in any direction.
in 3-level
factorial
designs
Contours In a rotatable design, the contours associated with the variance of the
of predicted values are concentric circles. Figures 3.16 and 3.17 (adapted
variance from Box and Draper, `Empirical Model Building and Response Surfaces,'
of page 485) illustrate a three-dimensional plot and contour plot, respectively,
predicte of the `information function' associated with a 32 design.
d values
are
concentr
ic circles

Informat Each figure clearly shows that the information content of the design is not
ion only a function of the distance from the center of the design space, but also
function a function of direction.

Graphs Figures 3.18 and 3.19 are the corresponding graphs of the information
of the function for a rotatable quadratic design. In each of these figures, the value
informati of the information function depends only on the distance of a point from
on the center of the space.
function
for a
rotatable
quadrati
c design

FIGURE 3.16 Three-

FIGURE 3.17
Dimensional Illustration for
Contour Map of the Information
the Information Function of a
Function for a 32 Design
32 Design

FIGURE 3.18 Three-

Dimensional Illustration of the FIGURE 3.19 Contour Map of the
Information Function for a Information Function for a Rotatable
Rotatable Quadratic Design Quadratic Design for Two Factors
for Two Factors
Central Composite Designs (CCD)
Box-Wilson Central Composite Designs

CCD designs A Box-Wilson Central Composite Design, commonly called 'a

start with a central composite design,' contains an imbedded factorial or
factorial or fractional factorial design with center points that is augmented
fractional with a group of 'star points' that allow estimation of curvature. If
factorial the distance from the center of the design space to a factorial
design (with point is ±1 unit for each factor, the distance from the center of
center points) the design space to a star point is |α| > 1. The precise value
and add "star" of α depends on certain properties desired for the design and on
points to the number of factors involved.
estimate
curvature Similarly, the number of centerpoint runs the design is to
contain also depends on certain properties required for the
design.

Diagram of
central
composite
design
generation for
two factors

FIGURE 3.20: Generation of a Central Composite

Design for Two Factors

A CCD design A central composite design always contains twice as many star
with k factors points as there are factors in the design. The star points
has 2k star represent new extreme values (low and high) for each factor in
points the design. Table 3.22 summarizes the properties of the three
varieties of central composite designs. Figure 3.21 illustrates the
relationships among these varieties.

Description of TABLE 3.22: Central Composite Designs

3 types of CCD
designs, which Central
depend on Composite
where the star Design Type Terminology Comments
points are
placed CCC designs are the original form of
the central composite design. The
star points are at some
distance from the center based
on the properties desired for the
design and the number of factors in
the design. The star points establish
new extremes for the low and high
Circumscribed CCC settings for all factors. Figure 5
illustrates a CCC design. These
designs have circular, spherical, or
hyperspherical symmetry and
require 5 levels for each factor.
Augmenting an existing factorial or
resolution V fractional factorial
design with star points can produce
this design.

For those situations in which the

limits specified for factor settings
are truly limits, the CCI design uses
the factor settings as the star points
and creates a factorial or fractional
factorial design within those limits
Inscribed CCI
(in other words, a CCI design is a
scaled down CCC design with each
factor level of the CCC design
divided by α to generate the CCI
design). This design also requires 5
levels of each factor.

In this design the star points are at

the center of each face of the
factorial space, so α = ± 1. This
variety requires 3 levels of each
Face Centered CCF
factor. Augmenting an existing
factorial or resolution V design with
appropriate star points can also
produce this design.
 Pictorial
representation
of where the
star points are
placed for the
3 types of CCD
designs

FIGURE 3.21: Comparison of the Three Types of Central

Composite Designs

Historical Design

Definition and Purpose

The purpose of a historical research design is to collect, verify, and synthesize evidence from
the past to establish facts that defend or refute a hypothesis. It uses secondary sources and
a variety of primary documentary evidence, such as, diaries, official records, reports, archives,
and non-textual information [maps, pictures, audio and visual recordings]. The limitation is
that the sources must be both authentic and valid.

What do these studies tell you?

1. The historical research design is unobtrusive; the act of research does not affect the results
of the study.
2. The historical approach is well suited for trend analysis.
3. Historical records can add important contextual background required to more fully
understand and interpret a research problem.
4. There is often no possibility of researcher-subject interaction that could affect the findings.
5.Historical sources can be used over and over to study different research problems or to
replicate a previous study.
What these studies don't tell you?

1. The ability to fulfill the aims of your research are directly related to the amount and quality
of documentation available to understand the research problem.
2. Since historical research relies on data from the past, there is no way to manipulate it to
control for contemporary contexts.
3. Interpreting historical sources can be very time consuming.
4. The sources of historical materials must be archived consistently to ensure access. This
may especially challenging for digital or online-only sources.
5. Original authors bring their own perspectives and biases to the interpretation of past events
and these biases are more difficult to ascertain in historical resources.
6. Due to the lack of control over external variables, historical research is very weak with
regard to the demands of internal validity.
7. It is rare that the entirety of historical documentation needed to fully address a research
problem is available for interpretation, therefore, gaps need to be acknowledged.