Pure error

noise estimated from replicated runs

The variation between runs made at exactly the same settings. Because the settings are the same, all of it is noise, so it is the cleanest base for testing whether an effect is real. In a 2^k design with r replicates it has 2^k · (r − 1) degrees of freedom.

Calculate it

DOE: 2-Level Factorial Design: 2^k factorial: run plan, effects, Pareto, interaction plots and best settings.

Related terms

• Replicate: a complete repeat of the design
• Lenth's pseudo standard error: significance of effects without replicates