In robust design experiments the objective is somewhat different compared to classical experimentation. What traditionally have been regarded as nuisance factors which ought to be held constant or nullified by blocking or randomization, are now exploited to improve the robustness of a system to variation outside and within the system. Instead of focusing only on location effect modeling, emphasis has been directed also to factors affecting the variance of the data, known as dispersion effect modeling. This thesis directs focus to three aspects of the parameter design step of robust design engineering: identifying design by environmental interaction effects, identifying dispersion effects, and improving robustness by reducing error transmitted through the system components. In detail, the following topics are considered.

Regarding design by environmental interaction effects, the 12 run Plackett-Burman design is considered for use in situations where interaction effects are present or can be expected. It appears that the 12 run design provides an appealing compromise to the two-level 8 and 16 run designs and holds properties which enable interaction effect identification even if the experimentis saturated. Moreover, a tentative strategy for follow-up experimentation is also considered for those cases in which the initial experiment does not provide a basis for unequivocal conclusions.

Regarding dispersion effect identification, a new method is proposed which facilitates the use of common hypothesis testing principles for judging the importance of dispersion effects from unreplicated experiments. The method is introduced in a 2^k-p design context and generalized to other multilevel experimental designs and to the inner and outer array set-up. The use of normal probability plotting is also considered for dispersion effect identification, and the effects of non-normality are investigated in a study using simulation.

Regarding error transmission from the system components, the Wheatstone bridge application, originally used by Taguchi, is revisited to illustrate and bridge the gap between engineering and statistical approaches for solving the parameter design problem.

Overall, the thesis consists of a framework and eight appended papers addressing the topics overviewed above.