Next: Design of
simulation study
Up: Statistical aspects
of Multi-Criteria-Optimisation
Previous: Derringer/Suich case
Sensitivity to modelling errors: a simulation case study
What is the consequence of estimating the underlying functions for the estimation of Xopt?
Design of Derringer/Suich: chemical data
Four
targets Y1 to
Y4, three factors
X1 to
X3.
Data were generated using a
central-composite design with 20 experiments.
Fitted typical second-order
optimisation models 
for each of the targets.
Targets Y1 and
Y2 are of the type "the more the
better".
Targets Y3 and
Y4 are real target value problems.
All desirability functions in the original paper were linear.
Derringer/Suich got the following models for the targets
Yi:
Dipl.-Stat. Detlef Steuer
1999-10-04