FREDERICK PHOA - Institute of Statistical Science, Academia Sinica, Taipei
Nowadays, research centers and industrial firms often ask their statistical consultants the following question before conducting experiments: How can we set up a good experiment? The definition of good is broad, but it can generally be interpreted as "spending the least amount of resources and gaining maximum amount of information". The aim of this talk is to try to provide some hints to this question. This talk is briefly divided into three parts. The first part introduces some recently developed screening methods in order to remove unimportant effects from the experiments. The second part discusses why nonregular fractional factorial designs are preferred over the traditional regular fractional factorial designs from several different viewpoints. The third part introduces the construction method for a new class of designs called Quaternary-code designs. This class of designs has many nice properties such that it can fulfill the good requirement when an experiment is set up. Some ongoing researches will be introduced at the end of this talk.