Experimental Design: Hypercubes and Blocking
Wednesday 4 June, 10:20--12:10
- Moderator: Nancy BELUNIS
- Merck \& Co.
- Shaowei CHENG, C.F.J. WU
- Department of Statistics, University of Michigan, Ann Arbor, MI, USA
Blocking is commonly used to
reduce the systematic variation of experiments.
In fractional factorial design, blocking can be
implemented by using blocking factor. For $2^{n-k}$
fractional factorial design in $2^p$ blocks,
Sun, Wu and Chen (1995), Sitter, Chen and Feder (1997)
study the confounding pattern between blocking factors and
treatment factors.
Their criteria for selecting good blocking schemes
are primarily based on the word lengths
of the defining contrast subgroup.
Extensions of their work to three-level designs
are complicated by the choice of
different contrasts for the main effects and
interactions.
Two different criteria are considered, based
on the word lengths of the defining contrast subgroup,
or on the (linear, quadratic) system of contrasts.
A collection of optimal blocking schemes for
three-level designs will be presented.
[Shaowei Cheng, Department of Statistics,
University of Michigan, Ann Arbor, Michigan, swcheng@stat.lsa.umich.edu]
- Arthur G. HOLMS
- Private Consultant, Cleveland, OH, USA
Bisgaard (1994a, 1994b) investigated the reduction of effective
resolution level of two-level fractional factorial experiments caused
by blocking. He did not consider experiments such as the expansible
sequences of Addelman (1969). Addelman did not consider the
confounding due to blocking. This article considers the block
confounding accompanying expansible sequences. Strategies are
described for designing experiments as expansible sequences of
orthogonal blocks under conditions of crossed classification block
effects of differing types. Rules are given for identifying at each
stage of the expansible options what aliased sets of full model
coefficients are confounded with what block effect parameters. This
principles are illustrated with a sequence, one stage of which
resembles the $2^{6-2}$ experiment of Bisgaard (1994a).
[Arthur G. Holms, 9303 Newkirk Drive, Cleveland, OH 44130-4166]
- William LI
- University of Minnesota, Minneapolis, MN, USA
- Agus SUDJIANTO
- Ford Motor Company, Dearborn, MI, USA
Using a fatigue-life project in an automotive company, this
paper shows a new algorithm for constructing Optimal Latin Hypercube.
The proposed algorithm makes use of special structures of Latin
Hypercubes.
It enjoys faster implementing time and higher efficiency
compared with other Latin Hypercube generating methods
[William Li, 1008 Sixth Street SE, Minneapolis, MN 55414, wli@csom.umn.edu]
URL: http://x4-135.csom.umn.edu
- Kenny Q. YE
- Department of Statistics, University of Michigan, Ann Arbor,
MI, USA
Designing computer experiments has drawn much attention as the
applications of computer models in industry have been rapidly increasing
in the past decade. In particular, Latin hypercubes have been frequently
used in conducting computer experiments. A class of orthogonal Latin
hypercubes which preserve the orthogonality among columns is proposed.
Applying an orthogonal Latin hypercube design to a computer experiment
benefits the data analysis in two ways. First, it retains the
orthogonality of traditional experimental designs. The estimates of
linear effects of all factors are not only uncorrelated with each other,
but
also uncorrelated with the estimates of all quadratic effects and bilinear
interactions. Second, it can facilitate non-parametric fitting procedures,
since we can select good space filling designs within the class of
orthogonal
Latin hypercubes according to some criteria. Also presented is an
application of orthogonal Latin hypercube designs to a cooling system
computer experiment at the ACD division of Ford Motor Co.
[Kenny Q. Ye, Department of Statistics,
University of Michigan, Ann Arbor, MI 48104, USA, qye@stat.lsa.umich.edu]
Back to JRC program
Back to JRC Home Page