Factorial Experiment — Definition, Formula & Example
A factorial experiment is a study that changes two or more factors at once and runs every possible combination of their levels. This lets you see whether one factor's effect depends on the level of another.
A design in which factors have levels and experimental units are assigned to all treatment combinations, usually by random assignment. From the cell means you estimate main effects (each factor's effect averaged over the other factors' levels) and interaction effects (how one factor's effect changes across the levels of another). A design with factors at 2 levels each is called a factorial design.
Key Formula
Where:
- = number of treatment combinations (cells) in the design
- = number of levels of factor i
- = number of factors; for example, a 2 × 3 × 4 design has 24 cells
Worked Example
Problem: A physics class tests paper airplanes using two factors: paper weight (light or heavy) and wing shape (narrow or wide). Planes are randomly assigned to each combination, and the response variable is flight distance. The mean distances are: light/narrow 20 ft, light/wide 30 ft, heavy/narrow 24 ft, heavy/wide 26 ft. Find the main effects and decide whether the factors interact.
Step 1: Count the treatment combinations. Each factor has 2 levels, so this is a design.
Step 2: Main effect of wing shape: average each wing level over both paper weights, then subtract. Do the same for paper weight.
Step 3: Check for interaction by finding the wing effect separately at each paper weight, then comparing those two effects.
Answer: Wide wings add 6 ft on average, and paper weight has a main effect of 0 ft. But the wing effect is 10 ft with light paper and only 2 ft with heavy paper. That 8 ft gap shows an interaction. If you tested each factor one at a time, you would never see this.
Why It Matters
Factorial designs are standard in two-way ANOVA in college statistics courses. Engineers use them in quality-improvement work to tune several machine settings at once. For the same precision on main effects, they need fewer runs than separate one-factor studies, and they are the only one of these approaches that can detect interactions.
Common Mistakes
Mistake: Concluding that a factor does not matter because its main effect is zero, as with paper weight in the example.
Correction: Check the interaction first. Paper weight changes how much the wings help, so it does matter. When an interaction is present, interpret each factor's effect at specific levels of the other factor, not just its average.
