Dispersion DoE (Dual Response)

A dispersion DoE (dual response) has at least two true replicates at every design point. From each point two quantities are computed: the mean $\bar{y}$ and the log-variance $\ln (s^2)$. Both are separately regressed against the factors.

• Goal: find factor settings that minimise process variability — independent of the target response
• Robustness — the classical Taguchi question ("which setting is insensitive to noise?") within a clean factorial structure
• Precondition: true, independent replicates per run — not multiple measurements on the same specimen (pseudo-replicates capture only measurement variation, not process variation)