Analysis of Data from Designed Experiments
Wim P. Krijnen, Ernst C. Wit · 2022
Sometimes the analyst is involved not only in the analysis of the data, but also in deciding what data to collect, i.e., the so-called design of the experiment. In principle one should aim to vary the explanatory factors in all possible ways. However, if a large number of factor is investigated, a so- called full-factorial design may require too too many runs to be practically feasible. In such cases, fractional designs that focus only on main effects and low-order interactions may be an efficient alternative. The analysis of such designed experiments focuses on three fundamental techniques. An analysis of variance is the key method for disentangling blocking, main as well as interaction effects. An analysis of a continuous response surface provides a way for finding the setting at which some chemical yield reaches its optimum. Finally, mixed effect models are used for describing the dependence of some chemical quantity on various factors, while dealing with experimental nuisance effects, such as the different operators that happened to be involved in the execution of the experiment. Examples from chemical engineering practice are analyzed and discussed in detail.