• :

    For both science and industrial systems, the 2-level full and fractional factorial designs play a particularly important (and efficient) role in the screening/sensitivity analysis problem: ferreting out dominant factors (and interactions) that affect a response of interest.  They also serve as efficient gateway designs for the optimization of such systems, and for the determination of how robust our conclusions are about any particular system factor.  The purpose of this talk is to present two methodologies for the analysis of data from 2-level factorials.  The first methodology is a structured 10-step EDA analysis protocol which serves as the basis for a routine and thorough visual analysis in that it provides information not only about important factors, but also about good settings, and goodness of models. The second methodology is a related EDA procedure (ordered block plot) which visually highlights how robust our conclusions are about a particular factor and additionally provides insight into the (sometimes-)nuanced nature of 2- and 3-term interactions.  All of these “deliverables” are critical in appreciating what makes a system “tic”, and what direction subsequent researcher efforts should go in the immediate future.  In a NIST-context, these tools have been routinely applied to a wide variety of problems over the years.  We illustrate these two methodologies by application to a selected set of problems including HRTEMs, Internet Traffic Congestion, and Cell Imaging Devices.

  • : James J. Filliben
  • : National Institute of Standards and Technology
  • : James J. Filliben
  • : experimental_design
  • : intermediate
  • : james.filliben@nist.gov
EDA Methodologies for the Analysis of 2-Level Factorial Designs