Analyzing Two-Level Fractional Factorial Experiments for Possible Dispersion Effects.

摘要

Unreplicated fractional factorial designs are frequently employed as screening designs when it is believed that a condition of effect sparisity will ensure that only a few of the possible effects are likely to be large. After considering the concept of effect sparisity as a justification for the use of unreplicated fractional designs, we discuss the situation where factors may influence not only the location but also the dispersion of the data. The aliasing of location and dispersion effects is explored and methods for identifying an appropriate location - dispersion model are considered.