Traditional response surface methodology focuses on modeling responses using parametric models with designs chosen to balance cost with adequate estimation of parameters and prediction in the design space. Using nonparametric smoothing to approximate the response surface offers both opportunities as well as problems. This article explores some conditions under which these methods can be appropriately used to increase the flexibility of surfaces modeled. The Box and Draper (1987) printing ink study is considered to illustrate the methods.
Anderson-Cook, Christine M. and Prewitt, Kathryn
"Some Guidelines For Using Nonparametric Methods For Modeling Data From Response Surface Designs,"
Journal of Modern Applied Statistical Methods:
1, Article 12.
Available at: http://digitalcommons.wayne.edu/jmasm/vol4/iss1/12