Details of the FACTEX Procedure |
MODEL Statement
- MODEL model-specification <MINABS
<(d)>> ;
-
You use the MODEL statement to provide the model for the
construction of the factorial design. The model can
be specified either directly by specifying
the effects to be estimated
with the ESTIMATE= option
or indirectly by specifying the resolution of the design
with the RESOLUTION= option.
If you create a
fractional factorial design or if you create a design that involves blocking,
the MODEL statement is required.
The two model-specifications
are described as follows:
- ESTIMATE=(effects) <option>
-
identifies the effects that you want to
estimate with the design.
To specify effects, simply list the names of main
effects, and join terms in interactions with
asterisks. The effects listed must be enclosed within parentheses.
See "Specifying Effects in the MODEL Statement"
for
details. You can use EST or E for the keyword ESTIMATE.
After the ESTIMATE= option,
you can specify the following option:
- NONNEGLIGIBLE=(nonnegligible-effects)
- identifies nonnegligible effects.
These are the effects whose magnitudes are unknown, but you do
not necessarily want to estimate them with the design.
If you do not want certain effects to be aliased with ESTIMATE= effects,
then list them in the NONNEGLIGIBLE= effects.
The nonnegligible-effects listed must be enclosed within parentheses.
You can use NONNEG or N for the keyword NONNEGLIGIBLE.
For example, suppose that you want to construct a fraction of a 24
design in order to estimate the main effects of the four factors.
To specify the model, simply list the main effects with the EFFECTS=
option, since these are the effects of interest.
Furthermore, if you consider the two-factor interactions to be significant
but are not interested in estimating them, then list these interactions
with the NONNEGLIGIBLE= option.
See
Example 15.8 for an example using the ESTIMATE= option.
See Chapter 16,"Theory of Orthogonal Designs"
for details on how the FACTEX procedure
interprets the model and derives an
appropriate confounding scheme.
- RESOLUTION=r
- RESOLUTION=MAXIMUM
-
specifies the resolution of the design.
The resolution number r must be
a positive integer greater than or equal to 3.
The interpretation of r is as follows:
- If r is odd,
then the effects of interest are taken to be those of order
(r-1)/2 or less.
- If r is even,
then the effects of interest are taken to be those of order
(r-2)/2 or less, and the nonnegligible effects are
taken to be those of order r/2 or less.
If you specify RESOLUTION=MAXIMUM, the FACTEX procedure searches
for a design with the highest resolution that satisfies
the SIZE statement requirements.
You can use RES or R for the keyword RESOLUTION and MAX for MAXIMUM.
For more on design resolution,
see "Resolution" .
For an example of model specification using the
RESOLUTION=r option, see "Example of a Half-Fraction Factorial Design"
.
For an example of the RESOLUTION=MAX option, see
"Example of a Full Factorial Design in Two Blocks" .
- MINABS <(d)>
-
requests a search for a design that has
minimum aberration.
Specifying (d) immediately after the MINABS option
requests a search for a minimum aberration design involving
interactions up to order d.
The default
value for d is the same as for the ALIASING option in the
EXAMINE statement. See "Minimum Aberration"
for more information.
For an example of the MINABS option, see Example 15.11.
Examples of the MODEL Statement
Suppose you specify a design with the
following FACTORS
statement, where the number of factors f can be replaced
with a number:
factors x1-x
f;
Then Table 15.6 lists equivalent ways to specify common models.
Table 15.6: Equivalent of Model Specifications
RES= option
|
EST= and NONNEG= options
|
model res=3 | model est=(x1-x f); |
model res=4 | model est=(x1-x f) nonneg=(x1|x2|x3| ...|x f@2); |
model res=5 | model est=(x1|x2|x3| ...|x f@2); |
The resolution specification is more concise than the effects
specification and is also more
efficient in an algorithmic sense. To decrease
the time required to find a design, particularly for designs with a
large number of factors,
you should specify your model using the RESOLUTION= option
rather than listing the effects.
For more information
on interpreting the resolution number, see
"Resolution" .
Copyright © 1999 by SAS Institute Inc., Cary, NC, USA. All rights reserved.