Eficiência de delineamentos fatoriais com dois fatores de níveis qualitativos em blocos incompletos

Abstract

This work aims to implement a single swap algorithm using the A-optimality criterion to find optimal or near optimal designs to factorial experiments with two factors ofqualitative leveis. Experimental situations were studied in which optimal designs for models with the absence and the presence of interactions were searched. The resulting designs were compared. The efficiency loss for using optimal designs in the absence of interaction to estimate a model with interaction was calculated, and vice-versa. Some resulting designs were also compared to designs built from the basic orthogonalplans of square lattices. The more robust situation was in general to plan to the presence of interactions. Some resulting designs were also compared to designs constructed from the basic orthogonalplans ofsquare and rectangular lattices. Designs resulting from the search algorithm were always A-optimal in this situation. The more robust situation was, in general, to plan for the absence ofinteractions. The algorithm could not find some ofthe resolvable designs (e.g. 5x5 treatments in 20 blocks ofsize 5 for the model without interaction). Ifthough, the designs resulting from the search were also A-optimum. In most ofthe situations the algorithm leads to optimal or near-optimal designs, showing that this is a very promising approach to design incomplete block factorial experiments.