Special Topics: Deleting Inactive Factors :: Getting Started with the SAS(R) 9.2 ADX Interface for Design of Experiments

Special Topics

Deleting Inactive Factors

After you run and analyze a two-level full factorial or fractional factorial experiment, you might discover that some factors are not statistically significant and have no bearing on the outcome of the response variables. In such cases you can drop the inactive factors permanently from the experiment and project the design in the subset of the remaining factors. This greatly increases the power of your design with no additional resources. Suppose the original design is an unreplicated full factorial design and factors are discarded. Then the projected design becomes a $2^{k-r}$ replicated factorial design with replicates.

For example, let's suppose the original design is a $2^{7-2}$ quarter-fraction factorial, Resolution 4, design and you find after fitting a model and analyzing the data that two of the factors are not important and these two factors are not involved in any significant interactions. So you choose to delete those two factors and project it into a full factorial design. The augmented design can now estimate all two-factor interactions. Montgomery (1997) suggests that the conclusions drawn from designs of this type of augmentation should be considered tentative and subject to further analysis.

Top of Page