Sensitivity Analysis with Fractional Factorial Design
Sensitivity Analysis Using Fractional Factorial Experimental Design · Also known as: FFD sensitivity analysis, fractional factorial sensitivity screening, SA-FFD, screening design sensitivity analysis
Sensitivity analysis with fractional factorial design (SA-FFD) is an experimental screening method that uses a carefully chosen fraction of all possible factor combinations to identify which input variables most strongly influence a system's output. By running only 2^(k-p) experiments instead of a full 2^k factorial, it makes sensitivity ranking feasible when many factors are present. The approach is widely used in engineering, product development, simulation modeling, and process optimization.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use SA-FFD when you have many (typically 5–30) candidate input factors and need to identify the vital few that drive output variation before investing in full optimization. It is particularly appropriate in engineering design, process engineering, simulation model calibration, and quality engineering where full factorial experiments are too expensive. Avoid it when interaction effects among many factors are expected to be as important as main effects and the design resolution would hopelessly confound them — in that case, use a higher-resolution or full factorial design. Also avoid SA-FFD when factor levels are continuous and the response is strongly nonlinear; response surface methodology or space-filling designs are more appropriate for optimization in that setting.
Strengths & limitations
- Drastically reduces the number of experimental runs needed to screen many factors — a 2^(10-4) design screens 10 factors in only 64 runs instead of 1,024.
- Provides a principled, quantitative ranking of factor importance that is easy to communicate to engineers and decision-makers.
- Designs are orthogonal, ensuring that main-effect estimates are uncorrelated and unambiguous at the chosen resolution.
- Directly feeds into follow-up optimization: factors identified as insensitive can be fixed, narrowing the optimization space.
- Well-supported by commercial software (JMP, Minitab, Design-Expert) and open-source tools (R package FrF2, Python pyDOE).
- Lower-resolution designs (III or IV) confound some interaction effects with main effects or with each other, so interaction estimates may be biased.
- Assumes a linear (or at most linear-plus-interaction) model; pronounced curvature in the response surface is not captured without adding center points or augmenting to a response surface design.
- Requires factor levels to be set and held at specified values, which may not be feasible for factors that are difficult to control precisely.
- Sensitivity rankings are local to the chosen factor ranges; changing those ranges can change which factors appear important.
Frequently asked
How do I choose the resolution of my fractional factorial design?
Resolution determines which effects are confounded. Resolution III is the minimum for screening main effects when interactions are assumed negligible; Resolution IV ensures main effects are not confounded with two-factor interactions but two-factor interactions are confounded with each other; Resolution V allows clean estimation of all main effects and two-factor interactions. As a practical rule, use at least Resolution IV when you suspect some interactions may be important, and use Resolution V if budget permits and interactions are critical to the study objectives.
What is the difference between SA-FFD and a Plackett-Burman design?
Both are screening designs for many factors, but they differ in structure. Fractional factorial designs are 2^(k-p) designs with clear aliasing structure defined by design generators, making it straightforward to understand which effects are confounded. Plackett-Burman designs are non-geometric designs that can screen k factors in k+1 runs (when k+1 is a multiple of 4), achieving greater efficiency but with complex, partial aliasing of interactions spread across many main-effect estimates — which makes interaction diagnosis harder. FFD is preferred when interaction structure needs to be understood; Plackett-Burman is preferred for very aggressive screening with many factors.
Can I apply SA-FFD to a computer simulation model?
Yes, and this is one of the most common applications. Computer experiments have no measurement noise, so effect estimates from a fractional factorial are exact (up to model linearity assumptions) with no need for replication. However, if the simulation is stochastic, replicate each design point across multiple random seeds to separate factor effects from simulation variance. If the response surface is highly nonlinear, consider augmenting with center points or switching to a space-filling design.
How is a sensitivity index from FFD related to Sobol indices?
FFD sensitivity rankings (based on main effect magnitudes) are analogous to first-order Sobol sensitivity indices — both quantify how much each factor individually explains output variance. However, Sobol indices are derived from variance decomposition over the entire factor space and handle nonlinear and non-monotonic responses; FFD effects assume linearity within the tested range. SA-FFD is much cheaper computationally and suffices for screening, while Sobol analysis is preferred for detailed global sensitivity analysis after key factors have been identified.
What sample size do I need?
The number of runs is determined by the design, not by a traditional power calculation. For k factors at two levels, common fractions are 2^(k-p): for example, 16 runs for 5–8 factors (Resolution IV or V), 32 runs for 6–10 factors, 64 runs for 10–15 factors. Adding 3–5 center points per design allows curvature detection without greatly inflating run count. If budget allows, add replicate center points to obtain an independent error estimate for significance testing.
Sources
- Box, G. E. P., Hunter, J. S., & Hunter, W. G. (2005). Statistics for Experimenters: Design, Innovation, and Discovery (2nd ed.). Wiley-Interscience. ISBN: 978-0471718130
- Saltelli, A., Ratto, M., Andres, T., Campolongo, F., Cariboni, J., Gatelli, D., Saisana, M., & Tarantola, S. (2008). Global Sensitivity Analysis: The Primer. Wiley. ISBN: 978-0470059975
How to cite this page
ScholarGate. (2026, June 3). Sensitivity Analysis Using Fractional Factorial Experimental Design. ScholarGate. https://scholargate.app/en/experimental-design/sensitivity-analysis-with-fractional-factorial-design
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Response Surface MethodologyExperimental design↔ compare