A Bayesian design criterion for the dual objective of model discrimination and parameter estimation
Two objectives of much experimentation in manufacturing are (i) to establish the form of an adequate mathematical model for the system in question and (ii) to obtain precise parameter estimates. Most experimental design research examines these issues separately, although Hill, Hunter and Wichern (1968) and Borth (1975) tackled both jointly. The new joint criterion presented here makes 2 contributions. One, the authors recognize that model and parameter entropy measures are defined up to an arbitrary scale, then recalibrate both to give each an equal weighting. Two, They simplify computations for th enormal linear model by identifying an appproxiamtion that leads to a closed form.
Copyright INSEAD. All rights reserved