Metrika (1997)45 :253-257
A Relationship between Etticiencies of Marginal
Designs and the Product Design
RAINER SCHWABE 1
Freie Universitfit Berlin 1. Mathematische lnstitut, Arnimallee 2-6, 14 195 Berlin, Germany
WENG KEE WONG
Department of Biostatistics, UCLA, Los Angeles, CA 90095, USA
Abstract: We derive the D- and G-efficienciesof product designs in a multifactor experiment in terms
of the D- and G-efficienciesof the designs in the marginal models.
Key Words and Phrases: Approximate design, D-, G-optimal design, marginal models, interaction,
multi-factor experiments.
1 Introduction
Multi-factor experiments are becoming increasingly common in practice. In this
paper, we study the relationship between the efficiency of a design in a multi-
factor experiment and the marginal efficiencies of each of its factors for a class
of linear models. We focus on the D- and G-efficiencies of product designs since
these designs are intuitively appealing and easy to construct, see for example
bok and Thibodeau (1980), Lau (1988), Lim and Studden (1988), and Wong
(1994). Moreover in the settings under consideration every design is dominated
by the product of its marginals (see Schwabe, 1996b).
All designs considered here are approximate designs or probability measures
defined on a given compact design region T. The models are linear with a known
regression function F(t) and errors are assumed to be independent and homo-
scedastic. Following convention, we measure the worth of the design ~ by its
information matrix:
M(~) = ~ F(t)F(OT~(dt) ,
T
Work supported by grants Ku 719/2 and 477/645/96 of the Deutsche Forschungsgemeinschaft
0026-1335/97/45:3/253 257 $2.50 © 1997 Physica-Verlag, Heidelberg