Experimental Design
| Introduction | Experiment Design | Model Optimization | Links |


Gosset manual

Yin-Carter paper

R based review of regression

more regression notes

Problems with stepwise regression

Experimental Design

assumes you know the neighborhood of the optimal parameters and answers the question of how to look around the neighborhood. This requires you assume something about the experimental model. GOSSET is used in this application.

GOSSET = minimum prediction experimental variance (assumes quadratic model can be as complex as desired)

INFAC is a similar software package, but is not open sourced or maintained. INFAC = Incomplete Factorial Experiment Design. It seems similar to a linear model used in GOSSET.


Model Optimization

uses the coefficients and parameters from your experiments to estimate what the most likely model is. This is done using the Akaike information criterion stepwise regression. This is a maximum information algorithm to estimate what the best model is given experimental results. A Monte Carlo search is then employed to find the maximum yield using the model.