Design Strategies in Fitting a Nonlinear Model

Year: 2014       Vol.: 63       No.: 1      

Authors: Michael Van Supranes

Abstract:

Estimation of parameters in a nonlinear model depends on the distribution of data points along various levels of curvature in the function to be estimated. Using Monte Carlo simulation, an optimal allocation procedure for building stratified designs was derived. The optimal allocation procedure conforms well to a proportionality property, directly relating the number of observations with the total curvature and measure or length of the domain. The proportionality property can be used to easily construct an allocation procedure that is near the optimal. Stratification results were applied and explored on uniform designs. Simulation results show that strategic stratification can improve the prediction accuracy of uniform designs.

Keywords: Stratification, Experimental Designs, Spline Regression, Monte Carlo Simulation

Download this article: