ADXFFA: Analyze Fractional Factorial Designs

Macros for the Design and Analysis of Experiments

ADXFFA: Analyze Fractional Factorial Designs

%adxffa(ds, resp, vlst, res, nb)

where

ds gives the name of the SAS data set that contains the design. The design must be in coded form, either as originally constructed or after recoding with the ADXCODE macro.

resp gives the name of the response variable (dependent variable) for analysis.

vlst lists the design factors.

res gives the resolution for the design.

nb gives the number of blocks in the design. If the design does not use blocking, nb can be omitted.

By default, if ds, vlst, res, or nb are omitted, the values from the last design constructed are used.

The ADXFFA macro computes a standard analysis for orthogonal two-level fractional factorial designs. The analysis lists effect estimates and significance levels, and plots the estimates against the quantiles of a normal distribution. The analysis provided by the ADXFFA macro considers only main effects and two-factor interactions. In the plot, the horizontal axis (labeled "EFFEST") is the effect estimate. To interpret the plot, use the values from the horizontal axis to identify effects.

If there are any degrees of freedom available to estimate error in the design, then the significance levels given for the effect estimates are based on tests using the estimated error term. Otherwise, a conservative estimate of error is constructed for each effect by pooling all the other effect estimates. In this situation, the normal plot of effects is a more useful analysis tool than the significance tests. The normal plot is based on the assumption that only a few of the effects will be significant and the rest are all due to noise. The nonsignificant effects should fall roughly on a line, with significant effects deviating from the line. Refer to Box, Hunter, and Hunter (1978) for more information on interpreting normal plots of effects.

For example, to obtain an analysis for a resolution 4 design contained in the DESA data set, with three factors WATER, FERT, and SPECIES and the response variable YIELD, use the following statements:

   %adxgen
   %adxff
   %adxinit
   %adxffa(desa,yield,water fert species,4)

Chapter Contents
Previous
Next
Top

ds	gives the name of the SAS data set that contains the design. The design must be in coded form, either as originally constructed or after recoding with the ADXCODE macro.
resp	gives the name of the response variable (dependent variable) for analysis.
vlst	lists the design factors.
res	gives the resolution for the design.
nb	gives the number of blocks in the design. If the design does not use blocking, nb can be omitted.