Design of experiments

Design of experiments is the systematic planning of experiments to estimate how controlled input factors affect measured outcomes while managing effort and uncertainty.

Experimental factor: A controllable or observed variable whose levels are examined for effects on an experimental outcome. Factors are the inputs whose effects the design is built to estimate.
Treatment combination: A specific joint setting of the factors in an experiment. Each experimental run applies a defined combination of factor levels.
Screening experiment: An experiment designed to identify a small set of influential factors among many candidates. Screening designs prioritize learning which inputs deserve detailed follow-up.
Observational study: A study that records variables without assigning experimental conditions to participants or units. Unlike a designed experiment, it cannot randomize factor settings to control confounding.
Response variable: A measured outcome used to assess the effects of experimental conditions. A design links chosen factor settings to measured responses.
Interaction (statistics): A change in one factor's effect depending on the level of another factor. Designs with suitable combinations can detect effects that single-factor comparisons miss.
Response surface methodology: A collection of experimental and statistical methods for modeling and optimizing a response. It guides experiments near promising settings to locate useful operating conditions.
One-factor-at-a-time method: An approach that changes one input at a time while holding other inputs fixed. It can miss interactions that factorial designs estimate by varying factors together.
Randomization: The assignment of experimental conditions by chance to reduce bias from uncontrolled influences. Random assignment protects comparisons from systematic differences among experimental units.
Analysis of variance: A statistical method that partitions response variation among modeled sources. It evaluates whether designed factor effects explain response variation beyond experimental error.

Linked from 16 pages