Design of Experiments Guide > Nonlinear Designs
Publication date: 08/13/2020

Nonlinear Designs

When the goal of your experiment is to fit a model that is nonlinear in the unknown parameters, use a nonlinear design to place design points in areas that are key to fitting the nonlinear model. Although you could use an orthogonal design that is optimal for a linear model, such designs, in general, do not place design points in locations that minimize the uncertainty (or maximize the precision) of the estimates of the fitted parameters.

The efficiency of a nonlinear design depends on the values of the unknown parameters. This creates a circular problem in that to find the best design, you need to know the parameters in advance. JMP uses a Bayesian approach to construct a nonlinear design that maximizes the average efficiency over specified ranges of the values of the parameter. To properly specify these ranges, you must have some insight about the system of interest.

Nonlinear designs offer these advantages compared to designs for linear models:

Predictions using a well-chosen model are likely to be good over a wider range of factor settings.

It is possible to model response surfaces with complex curvature and with asymptotic behavior.

Figure 23.1 Design Points for a Nonlinear Model 


Overview of Nonlinear Designs

Examples of Nonlinear Designs

Create a Nonlinear Design with No Prior Data
Augment a Design Using Prior Data
Create a Design for a Binomial Response

Nonlinear Design Launch Window

Nonlinear Design Window

Design Generation
Make Table or Augment Table

Nonlinear Design Options

Statistical Details for Nonlinear Designs

Nonlinear Models
Radial-Spherical Integration of the Optimality Criterion
Finding the Optimal Design
Want more information? Have questions? Get answers in the JMP User Community (