The Necessity of the Sparsity Assumption in Unreplicated Screening Experiments

Abstract

Designed experiments that are orthogonal for the main effects are also D-optimal for the  main effects model. Despite these strengths, it can be very difficult to separate the active effects from the negligible ones when the number of active effects exceeds half the number of runs. This talk shows why this is true by demonstration using many different analytical approaches for unreplicated screening experiments. The results underscore the value of replication for providing a model independent estimate of error when the number of active effects is large.

2015 QPRC