What Is a Screening Design? Definition and Uses

A screening design is a small, deliberately economical experiment whose only job is to sort a long list of candidate factors into the vital few that drive a process and the trivial many that do not. It optimizes nothing. It buys clarity cheaply, so the expensive experiments that follow are spent on factors that deserve them.

How it works

Screening designs achieve their economy through heavy fractionation. Low-resolution fractional factorials and Plackett–Burman designs — a family published in 1946 — can assess many factors in remarkably few runs: eleven factors in twelve runs, seven in eight. The trade is explicit. Main effects are estimated cleanly enough to rank the factors, but interactions are entangled and largely out of reach. That is acceptable because of the question being asked: screening asks who matters, not how, and not in combination with what.

The strategy leans on effect sparsity — the expectation that only a minority of candidates will prove important — and experience across industries keeps rewarding that bet. The typical sequence runs screen, then characterize the survivors with a higher-resolution factorial, then optimize with response surface methods. Each stage narrows the field and raises the stakes, which is exactly the order in which experimental money should be spent.

A worked example

A laboratory struggles with inconsistent turnaround on a routine assay and lists eight plausible drivers, from incubation temperature to batch size to instrument warm-up. Eight factors would need 256 runs in full factorial; the lab screens them in twelve. The analysis ranks three factors far above the rest. The other five are set to convenient values and dropped from the investigation, and the follow-up factorial on the surviving three — sixteen runs, fully characterized — finds the interaction that becomes the eventual fix. Total cost: under thirty runs to untangle an eight-variable problem. Just as important is what the screen prevents — weeks of meetings arguing over five factors that, on the evidence of twelve runs, never deserved the attention.

Common mistakes

• Optimizing from screening results — the design was never built to locate a best setting
• Trusting main effects blindly when a low-resolution alias structure could be distorting them
• Leaving a cheap-to-vary factor off the list to save runs the design would have absorbed anyway
• Skipping the confirmation step on the factors you are about to discard

Screening spends a little to find out where spending a lot will be worth it.

Screening strategy — when to fractionate, and how to read what a small design can and cannot say — is taught in our Black Belt program ($499, about 60 hours, with a 150-question proctored exam and lifetime access to the material). It is the phase of experimentation where judgment matters most, which is why it sits at the highest belt.