Variance Control

Two kinds of variance control or estimation are generally used in addition to that inherent in the Main Factor design:

1. entire designs may be replicated or

2. individual runs may be repeated.

Repeating runs provides information for an estimate of the experimental error, which in turn provides a basis for estimating the significance of a regression and of the fitted coefficients. Replicating a design provides a basis for estimating the effect on variation caused by factors not in the regression model. The variation is calculated at the factor-levels of the design array runs. Subsidiary factors may be selectively included or excluded from the regression to evaluate their relative importance to variation.

The Taguchi approach to inner (subsidiary) designs provides the replicates for a variance analysis under specific conditions of variation provided by the subsidiary design. The usual expectation is that the deliberate sources of variation will be more significant than the uncontrolled variation from unknown sources.

Learn more about the Regression tools in Six Sigma Demystified (2011, McGraw-Hill) by Paul Keller, in his online Regression short course (only $99), or his online Black Belt certification training course ($875).