What Is Blocking in Design of Experiments?

Blocking is the technique of grouping experimental runs by a known nuisance variable — raw-material batch, day, machine, operator — so that its influence can be measured and mathematically removed rather than left to contaminate the results. The experimenters’ maxim puts it in eight words: block what you can, randomize what you cannot.

How it works

A block is a set of runs performed under homogeneous conditions — the same batch, the same day, the same machine. The design assigns factor combinations across blocks in a balanced way, and run order is still randomized within each block. In the analysis, differences between blocks are calculated and set aside, so the comparisons that matter — factor high versus factor low — are made within matched conditions. The nuisance variable still varies; it simply no longer gets to masquerade as a factor effect.

Blocking differs from randomization in what it assumes. Randomization defends against influences you cannot name. Blocking handles the ones you can name but cannot hold constant — and converts them from threats into measured quantities. In practice the two always work together: a sound experiment blocks on the nuisances it knows about and randomizes against the ones it does not. The paired comparison taught in basic statistics is the smallest possible blocked experiment — two treatments applied to the same unit, with the unit itself serving as the block.

A worked example

A bakery supplier is testing two factors in a dough process, but the eight-run experiment needs two days, and flour arrives fresh each morning. Rather than hope the deliveries are identical, the team blocks on day: four runs per day, chosen so each day contains a balanced set of factor combinations, randomized within the day. The analysis later shows the second day ran noticeably different overall — a difference the blocking isolates in a single number, leaving the factor comparisons clean. Without the block, that day effect would have landed on whichever factor was unlucky. Balance is what makes it work: had one day held mostly high-temperature runs, day and temperature would have stayed partly tangled despite the block.

Common mistakes

• Splitting an experiment across days, machines, or batches without blocking on the split
• Blocking on a variable and then forgetting to account for it in the analysis
• Treating a block difference as a nuisance only — a large day effect is itself a clue worth chasing
• Confusing blocking with holding a variable constant, which narrows every conclusion to that one condition

A block turns a known nuisance into a measured number instead of a hidden bias.

Blocking belongs to the design toolkit taught in our Black Belt program ($499, about 60 hours, ending in a 150-question proctored exam). Green Belt ($299) prepares the ground with hypothesis testing — the paired comparison, which Green Belts learn, is blocking in miniature.