Compromise designs by augmenting the second-order Rechtschaffner design under a baseline parameterization
Year: 2026
Type: article
Abstract: Two-level factorial experiments under a baseline parameterization model effects relative to a reference configuration, such as comparing treatments to a placebo or modifications to a standard design. The smallest design fitting all main effects and two-factor interactions, called the second-order Rechtschaffner design, leaves no degrees of freedom for residuals and offers no protection against active higher-order interactions. A natural remedy adds a few carefully chosen runs, accepting some bias in exchange for reduced variance. This article studies the resulting variance–bias tradeoff. We give closed-form expressions for both the variance reduction and the bias contribution of any augmentation, valid for any number of factors and runs. Using these formulas we identify the variance-optimal one-run and two-run augmentations explicitly—for two runs, provably for up to twenty-one factors—and describe a canonical family that appears to give the optimum in all cases we examined numerically. The optimal added runs differ from the all-treatment configuration by removing one or two factors at a time, with the choice between regimes governed by an explicit polynomial equation in the number of factors. At the same total run budget, comparisons against replication of the saturated design show variance reductions of approximately eight to thirty-eight percent.
Source: Figshare
Author: Ruwan C. Karunanayaka
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7233547898
FWCI:
Cites:
Cited by:
Related to:
