Design of Experiments glossary
Short, precise definitions of the terms you will meet when planning and analyzing experiments.
- Aliasing
- When two or more effects cannot be told apart because the design varies them in the same pattern.
- ANOVA
- Analysis of variance: splitting the variation in the data into parts explained by the model and by error.
- Box-Behnken design
- A response-surface design that runs at the midpoints of the edges of the factor cube.
- Central composite design
- A response-surface design: a factorial plus axial and centre points, able to fit quadratic models.
- Centre point
- A run with every factor at its midpoint, usually replicated.
- Factor
- An input you deliberately change in an experiment, such as temperature or pressure.
- Fractional factorial design
- A design that runs a carefully chosen fraction of a full factorial.
- Full factorial design
- A design that runs every combination of the factor levels.
- Interaction
- When the effect of one factor depends on the level of another.
- Lack of fit
- A test of whether the model's form is too simple, compared with pure replicate error.
- Level
- A specific setting of a factor. Two-level designs use a low and a high setting.
- Predicted R²
- How well the model predicts runs it was not fitted to, based on leave-one-out errors.
- Pure error
- Variation among replicated runs: the noise the model could never explain.
- Randomization
- Running the experiments in random order to protect against hidden trends.
- Residual
- The difference between a measured response and the model's fitted value.
- Resolution
- How seriously a fractional design confounds effects: higher resolution means less severe aliasing.
- Response
- The measured outcome you want to improve, such as yield, strength or purity.
- Response surface methodology
- Fitting a quadratic model to find the factor settings that optimize the response.