What Is a Full Factorial Design? Definition and Uses

A full factorial design is an experiment that tests every possible combination of the chosen factor levels. Two levels of three factors means 2 × 2 × 2, or eight runs; add a fourth factor and the count doubles to sixteen. In exchange for that thoroughness, the design gives up nothing: every main effect and every interaction, of every order, can be estimated cleanly.

How it works

The workhorse is the two-level factorial. Each factor gets a low setting and a high setting, and the design matrix pairs them so the experiment is balanced: each factor spends half its runs at each level, within every combination of the other factors. That balance — statisticians call it orthogonality — is what lets one experiment answer many questions at once. The effect of temperature can be computed as if temperature were the only thing varied, even though everything varied together.

The cost is exponential growth. The run count doubles with every factor added: eight runs for three factors, thirty-two for five, one hundred twenty-eight for seven. Full factorials therefore shine when the factor list is short — typically two to four — or when runs are cheap. When the list is long, experimenters switch to fractional factorials and accept a known, chosen loss of information instead.

A worked example

An injection-molding shop sees too many warped parts. The team picks three factors — mold temperature, injection pressure, and cooling time — each at a sensible low and high. The full factorial takes eight runs; the team replicates it for sixteen, in random order, measuring warp on every part. Analysis shows cooling time has the largest effect, pressure a modest one, and a temperature-by-cooling-time interaction: longer cooling only helps when mold temperature is low. The team sets the process accordingly, and the confirmation runs hold.

Practical pointers

• Reserve full factorials for a short list of factors — screen a long list down first
• Set low and high levels boldly enough to produce a detectable difference, but inside safe operating limits
• Replicate the design, or add center points, so effects can be judged against real process noise
• Randomize run order even when it is inconvenient — especially when it is inconvenient
• Analyze interactions before declaring any single factor’s effect, because interactions change the story

Eight well-chosen runs can settle an argument that years of tinkering never will.

Two-level factorials are the entry point to formal experimentation, and our Black Belt program ($499, about 60 hours of material) covers them from design matrix to analysis, alongside the fractional and response surface designs built on top of them. The exam — 150 proctored questions, 70% to pass, one free retake — treats factorials as core material, because in practice they are.