On the determination of robust settings in parameter design experiments

In parameter design experiments, it is common to fit a regression model to the original response and then use the model variance induced by the noise variation to determine robust settings of the control factors. We obtain some general characterization of the robust settings in terms of the absolute slopes in the control-by-noise interaction plots. The simplicity and graphical appeal of the latter conditions provide further insight into the choice of robust settings in these experiments.Control-by-noise interaction Transmitted variance model Response model