Design of Experiments (DOE) is a tool to model a continuous output variable (Y) with continuous or discrete input/process variables (Xs). It describes a way to set up the eyperimental data collection (experimental plan) and to analyze the results from the conducted experiments (DOE analysis).

Main purposes

  • Screening – Reducing the number of Xs
  • Focusing – Verifying and quantifying significant X-Y relationships
  • Optimizing – Determining the best settings for the Xs

Main outputs

  • Model – an equation of the form: 𝑌=𝑏0+𝑏1𝑋1+𝑏2𝑥2+𝑏3𝑥1𝑥2…
  • P-values of terms – how significant are the factors (Xs)?
  • R-sq, S unexplained – how much of the observed variation does the model explain, what portion remains unexplained?

The experimental plan is set up in a way

  • To get as much information as possible from all the variables included in the design (each factor will be tested equally often on each level)
  • This allows for checking for main effects and interactions of factors
    • Main effect: effect of one variable independent of others (e.g. for the Y “braking distance of a car”, a main effect may be the X “speed”)
    • Interaction effect: effect depends on the influence of two variables at the same time (e.g. for the Y “braking distance” an interaction of “tire type” and “street conditions” may be important: one type of tire is better on wet streets, another type of tire is better on dry streets – so you can’t generally say one type of tire is better, it depends)