Fractional Factorial
A fractional factorial design is a statistical experiment that runs only a carefully chosen fraction of all possible factor combinations from a full factorial design. It enables screening of many variables with fewer test runs, less time, and lower cost while estimating main effects and selected two-factor interactions. Some effects become confounded or aliased, so it is mainly used for early-stage screening and optimization rather than final confirmation.
On a shop floor or in a warehouse, fractional factorial designs let operations teams test multiple controllable factors without shutting down the process for exhaustive trials. A materials group might explore changes to reorder policy, supplier lot size, and receiving inspection intensity in a single reduced run matrix to see which combination most reduces backorder rate and WIP starvation. A warehouse team could screen slotting method, pick path, and replenishment trigger against pick time and mispicks. Procurement might test order cadence, minimum order quantity, and approved vendor count to understand service level and carrying cost. Because only a fraction of combinations is run, these experiments are affordable and minimally disruptive. The result is a ranked set of dominant drivers that can guide larger optimization trials. This makes fractional factorials a practical first step for process improvement, allowing teams to separate signal from noise while spending limited time and resources.
- Confounding a critical interaction with a main effect: A low-resolution design can make a real interaction such as supplier lead time by reorder point look like a single-factor effect, so the team optimizes the wrong lever and creates stockouts or excess inventory.
- Using the design on an unstable process: If demand spikes, receiving delays, or machine downtime occur during the test, the reduced design misattributes variation to the studied factors rather than the instability, yielding false conclusions.
- Treating a screening design as a final operating policy: Because fractional factorials only identify important factors, deploying the result as a finished standard without follow-up runs can lock in a bad replenishment rule and cause bottlenecked receiving, poor slot utilization, or persistent shortages.
When is a fractional factorial design preferable to a full factorial in a supply or warehousing experiment?
When there are many controllable factors and the objective is to screen for dominant drivers quickly, especially if full coverage would be too costly, slow, or operationally disruptive to normal material flow.
What does design resolution mean in a fractional factorial?
Resolution indicates the shortest word in the defining relation and describes which effects are aliased. Higher resolution provides better separation of main effects and lower-order interactions, making conclusions more trustworthy.
What is the main statistical limitation of a fractional factorial design?
Only a subset of combinations is run, so some effects cannot be distinguished from each other. Interpretation relies on the assumption that higher-order interactions are negligible, and confirmation runs are often needed.