SED navigation bar go to NIST home page SED Home Page SED Contacts SED Projects SED Products and Publications Search SED Pages
Invited Session: Recent Advances in Design & Analysis of Experiments

Invited Session: Recent Advances in Design & Analysis of Experiments

Organizer: Don X. Sun, Bell Laboratories
Session Chair: Dennis Lin, Pennsylvania State Univ.

Multipanel Conditioning: Modeling Data from Designed Experiments

William S. Cleveland
Statistical Models & Methods Research Dept., Bell Laboratories

Montserrat Fuentes
Dept. of Statistics, Univ. of Chicago

Multipanel conditioning, a feature of Trellis display, is a visualization tool for studying the dependence of a response on two or more explanatory variables. For data from designed experiments, multipanel conditioning can be used to display the raw data, partial fits, partial residuals, fits, and residuals. This is particularly effective for ferreting out interactions, two-factor and higher order, and goes well beyond the standard, but limited, interaction plots used to show the two-factor interactions of a fitted model. Multipanel conditioning helps the data analyst to build parsimonious models for experimental data in place of standard models associated with the analysis of variance which often employ more parameters than necessary, losing both sensitivity to detecting effects and simplicity.

[William S. Cleveland, Bell Laboratories, 600 Mountain Ave., Murray Hill, NJ 07974 USA;]


Click on for black and white version or on for color version.


Interaction Between Practical Experimentation & Design Theory: Some Recent Advances and Experience

Jeff Wu
Dept. of Statistics, Univ. of Michigan

Practical experimentation often provides feedback that stimulates the development of new theory in design, which in turn provides better tools for experimenters. In this talk I will use three recent examples from my research to illustrate and support this scientific paradigm. 1.Theory of minimum aberration design(based on a novel use of the McWilliams' identity in coding theory) for optimal factor assignment in 2^n-k fractional factorial design. 2.A theory and framework for optimal blocking schemes in full and fractional factorial design, which allows blocking to be chosen in an optimal manner. 3.A hidden projection property for non-geometric designs(eg, 12-run Plackett- Burman design) which explains why interactions in non-geometric designs with complex aliasings can often be estimated. Traditionally this type of designs are used only for factor screening. This property enables the experimenters to use the same designs to study interactions without incurring additional cost.

[Jeff Wu, Dept. of Statistics, Univ. of Michigan, Ann Arbor, MI 48109 USA; ]

Date created: 6/5/2001
Last updated: 6/21/2001
Please email comments on this WWW page to