Factorial Design
A factorial design is a design of experiments in which two or more factors are studied simultaneously, with every combination of their levels included. This allows estimation of both main effects and interactions. Full factorial designs require 2^k runs for k two-level factors, while fractional factorial designs test a subset to reduce run count.
In a manufacturing or warehousing setting, factorial design helps teams test how operational inputs jointly drive outcomes such as throughput, defects, lead time, stockout risk, picking accuracy, receiving delay, or material availability. Typical factors include reorder point, safety stock, supplier lead time, batch size, barcode scan compliance, forklift availability, and receiving dock staffing; responses include fill rate, line stoppage minutes, inventory record accuracy, and dock-to-stock time. The key advantage is that factorial experiments expose interaction effects that one-factor-at-a-time testing misses. For example, higher safety stock may improve service only when supplier lead time variability is high, or extra receiving labor reduces congestion only when inbound volume exceeds a threshold. A supply-chain team defines the objective, selects factor levels, runs a planned matrix of scenarios, and analyzes results with ANOVA, effect plots, or regression to identify the inputs that materially drive performance.
What is the difference between a factor and a level?
A factor is an input variable, and a level is a specific setting of that factor.
Why use factorial design instead of one-factor-at-a-time testing?
Factorial design estimates both individual factor effects and interactions, which is essential when inventory, procurement, and warehousing variables influence each other. One-factor-at-a-time testing misses these interaction effects and can lead to incorrect conclusions.
When is fractional factorial design preferred?
When the number of factors is too large for a full factorial and the goal is screening—identifying the few most important drivers with fewer runs. NIST notes full factorial designs are not recommended for five or more factors.