Statistics - Design Of Experiment

DOE stands for Design Of Experiment. It became a popular experimental tool only recently - in the 1990's, but was developed in the first part of the last century by Sir R. A. Fisher. A Japanese engineer, G. Taguchi, popularized it as a practical solution rather than a mathematical curiosity. Using his techniques, DOE became a useful statistical method for maximizing the knowledge gained from experimental data.

What is it?

As the name implies, it is a statistical discipline that begins with the design of the experiment. Factors and their extreme values are selected and tested in mixed trials. Powerful statistical techniques then sleuth out the individual factor contributions and produce a mathematical model of the experiment.

In the literature about it, there are many conjectures like: had Newton used DOE, he would saved years from pondering the form of the equation for force on objects, gravity, etc., and who knows how much further he would have advanced... They then give a simple example of how Newton could have applied it and gotten after 4 measurements the equation fundamental equation F=ma from the DOE methods themselves.

DOE methods can tell which measured variables are significant to affect the measured results and which are not based only on extreme values for the variables. This is "Screening". When the extreme values and others in the middle are used, the DOE method produces a mathematical model that shows how each factor affects the others. This is a "Modeling" experiment. Coefficients in the models relate to how sensitive the results are to changes in the variables. The simplest experiments use linear and mixed terms similar to partial results obtained in standard analysis of variance. The difference is, the terms are all strung together for all variables in the trials; no need to pair up factors. This means that instead of running 2 to the n trials for n factors, you need many fewer. Using DOE can save lots of time!