Principles of DOE
Several techniques enable planned tests to produce efficient and objective results. It is not always possible to faithfully include each technique in a design, but the more tests rely on these elements, the more trustworthy results they provide.
Randomization: Randomization is vital tool for eliminating confounds or extraneous factors (those factors that we are not controlling or observing), and potential influences that are outside of the evaluators control. Randomizing the order of test runs, the allocation of experimental units to conditions, test resources to experimental units, and so on, helps to eliminate bias and enables causal inference. Many statistical analyses assume that observed units are independent, and randomization usually validates this assumption.
Replication: Replication involves independently repeating each run or test point. Replication allows the evaluator to estimate experimental error, or variability within each condition, and provides a more stable, precise estimate of the factors effects for each condition. Replication helps the analyst separate a true effect from experimental error or random variation.
Blocking: Blocking is a technique to reduce variability among conditions that is not due to the experimental factors, or independent variables. An evaluator can standardize conditions under which the test is conducted by blocking conditions within levels of the “nuisance” factor. This provides a way to keep variance from nuisance variables separate from that of key factors.