statrefs home‎ > ‎Main‎ > ‎Methods‎ > ‎Design of Experiments‎ > ‎

Response Optimization

Where is the process mean the best?  (highest?  lowest?)

  • Consider using a contour plot of the process mean based on the DOE model.

Where is the process variation (std dev) the smallest?

  • Can generate a contour plot of the expected process variance based on the DOE model.

Consider using the mean and std dev to calculate an estimated Process Performance Index value.

  • Can estimate std dev by dividing the 95% Prediction Interval by four.  (It represents two standard deviations on each side of the regression mean.)
  • Can use the point estimate of the regression mean.

Where is the slope of the response the flattest?

  • We generally want the process to operate in a region where changes (controlled or uncontrollable) in process inputs do not cause a large shift in the process output.
  • Looking at the response slope w.r.t. changes in process inputs, look at the minimum amount of control that is available for each process input (e.g. how tightly can temp, pressure, other be controlled?) and check to see how far this input variation might move the response along the response contour.


Optimizing more than one response variable or process output.

[more information to be added later]

Various software packages offer methods for finding process settings that optimize across several response variables. 
  • Statistica DOE Profiler
  • SAS
  • R
  • Excel - can use the optimizer add-on utility in combination with the model equations from other software packages.