Design of Experiments Terminology can be daunting! Here’s an easy glossary to reference when working Design of Experiments Terminology questions.
Alias structures are used in fractional factorial designs to determine which combinations of factors and levels will be tested.
Each factor appears the same number of times.
Blocking involves recognizing uncontrolled factors in an experiment – for example, gender and age in a medical study – and ensuring as wide a spread as possible across these nuisance factors. Read more about blocking.
A classification factor is an element that cannot be specified or set by the experiment designer, but can be used in sample selection. The sex of a subject is an example of a classification factor.
Confounding occurs when we can’t be sure which factors – or combinations of factors – are affecting a result. Blocking can help to minimize confounding. Read more about confounding.
An experimental factor is one that can be modified and set by the person designing the experiment.
Experiments can be abbreviated numerically.
2^5 means that there are 5 factors at 2 levels.
Factors are elements in your experiment – whether in your control or outside of it – that affect the outcomes. Read more about factors.
Experiment design that tests only a subset of the possible factor-level combinations. Read more about Fractional factorial.
Experiment design that tests the full set of possible factor-level combinations. Read more about full factorial.
Factor interactions occur when multiple factors affect the output of an experiment.
Every treatment factor in an experiment will have at least two levels. These might be minimum and maximum values, discrete groups (for example, ‘male’ and ‘female’), or designated ranges (for example, ages 25-34; 35-44; etc).
A nuisance factor is the opposite of a treatment factor – an element that is of no interest for the experiment, but needs to be considered anyway in case it skews results.
A qualitative factor is a treatment factor that contains a number of categories.
A quantitative factor is a treatment factor that can be set to a specific level as required.
Randomization is essentially conducting trials in no pre-determined order. There’s no reason to believe conducting trials randomly will cost less (in fact, it may very well cost more) but it helps overcome (or avoid) bias. It also means that noise factors are equally as likely to be present in all trials.
Replicates are multiple experimental runs of the same factors and experiments. You do this so you can get a sense of the variability in the data set. Good expansive article on Replicates and Repeats here from MiniTab.
Replication means more trials, thus more costs but it only allows you to study whatever interactions the original trials already allowed you to study. It also allows for the estimation of experimentation error.
Resolution tells us the confounding we will see between factors.
A treatment factor is an element that is of interest to you in your experiment, and that you will be manipulating in order to test your hypothesis.
Six Sigma Black Belt Certification Design of Experiments Terminology Questions:
Question: Which of the following best describes a controlled variable whose influence on a response is being studied? Taken from (ASQ sample Black Belt exam.)
Six Sigma Green Belt Certification Factorial Design Questions:
Question: The following chart was developed by a six sigma team to measure reactions on two different products:
In this chart, temperature represents the
Question: Which of the following graphs represents a factorial experiment with the strongest interaction?