Pragmatic Fractional Factorial Experiment

A standard fractional factorial design is typically used in industrial or laboratory settings with high control over factor levels. The 'pragmatic' modifier means the design is implemented in real-world conditions — clinical settings, schools, communities — where participant heterogeneity, delivery variation, and logistical constraints are accepted rather than eliminated. The statistical framework is identical; the difference is in implementation philosophy and the emphasis on real-world feasibility.

MOST (Multiphase Optimization Strategy) is a framework developed by Collins et al. that uses a pragmatic fractional factorial experiment as its optimization phase. After identifying active components via the fractional factorial experiment, MOST proceeds to a confirmatory randomized controlled trial testing the optimized intervention. The pragmatic fractional factorial experiment is thus a preparatory step that makes the eventual RCT more efficient by ensuring only effective components are included.

Resolution determines which effects are aliased. Resolution III designs confound main effects with two-factor interactions — acceptable only if you are confident interactions are negligible. Resolution IV designs alias two-factor interactions with each other but keep main effects clean — a safer choice when interactions are plausible. Resolution V or higher allows estimation of all two-factor interactions. Start by listing which interactions you cannot afford to confound and choose the minimum resolution that keeps those effects estimable within your feasible number of runs.

Sample size depends on the number of conditions in the chosen fraction, the expected effect size for individual components, and the desired power. Because each participant contributes information about all components simultaneously, the design is efficient — but each condition still needs an adequate cell size. Collins (2018) recommends powering for the smallest practically important component effect. Simulation-based power analyses using the alias structure of your specific design are advisable when interactions are expected.

No — the fractional factorial design requires that each component can be independently varied across conditions. If two components always appear together or are logically inseparable, they cannot be treated as independent factors. In that case, collapse them into a single factor with distinct levels, or use a different design (e.g., a dismantling study or a full factorial over the separable factors only).