The usefulness of Bayesian optimal designs for discrete choice experiments

Recently, the use of Bayesian optimal designs for discrete choice experiments, also called stated choice experiments or conjoint choice experiments, has gained much attention, stimulating the development of Bayesian choice design algorithms. Characteristic for the Bayesian design strategy is that it incorporates the available information about people's preferences for various product attributes in the choice design. This is in contrast with the linear design methodology, which is also used in discrete choice design and which depends for any claims of optimality on the unrealistic assumption that people have no preference for any of the attribute levels. Although linear design principles have often been used to construct discrete choice experiments, we show using an extensive case study that the resulting utility‐neutral optimal designs are not competitive with Bayesian optimal designs for estimation purposes. Copyright © 2011 John Wiley & Sons, Ltd.     

An improved two‐stage variance balance approach for constructing partial profile designs for discrete choice experiments

In many discrete choice experiments set up for product innovation, the number of attributes is large, which results in a substantial cognitive burden for the respondents. To reduce the cognitive burden in such cases, Green suggested in the early '70s the use of partial profiles that vary only the levels of a subset of the attributes. In this paper, we present two new methods for constructing Bayesian ‐optimal partial profile designs for estimating main‐effects models. They involve alternative generalizations of Green's approach that makes use of balanced incomplete block designs and take into account the fact that attributes may have differing numbers of levels. We refer to our methods as variance balance I and II because they vary an attribute with a larger number of levels more often than an attribute with fewer levels to stabilize the variances of the individual part‐worth estimates. The two variance balance methods differ in the way attributes with differing numbers of levels are weighted. Both methods provide statistically more efficient partial profile designs for differing numbers of attribute levels than another generalization of Green's approach that does not weight the attributes. This method is called attribute balance. We show results from an actual experiment in software development demonstrating the usefulness of our methods. Copyright © 2014 John Wiley & Sons, Ltd.     

Properties of optimal designs of regression experiments in singular cases

A test for optimality using the pseudoinverse matrix is given for the case when the matrix of the design is degenerate.Translated from Staticheskie Metody, pp. 151–153, 1978.     

Flexible sliced designs for computer experiments

Sliced Latin hypercube designs are popularly adopted for computer experiments with qualitative factors. Previous constructions require the sizes of different slices to be identical. Here we construct sliced designs with flexible sizes of slices. Besides achieving desirable one-dimensional uniformity, flexible sliced designs (FSDs) constructed in this paper accommodate arbitrary sizes for different slices and cover ordinary sliced Latin hypercube designs as special cases. The sampling properties of FSDs are derived and a central limit theorem is established. It shows that any linear combination of the sample means from different models on slices follows an asymptotic normal distribution. Some simulations compare FSDs with other sliced designs in collective evaluations of multiple computer models.     

Application-driven sequential designs for simulation experiments: Kriging metamodelling

This paper proposes a novel method to select an experimental design for interpolation in simulation. Although the paper focuses on Kriging in deterministic simulation, the method also applies to other types of metamodels (besides Kriging), and to stochastic simulation. The paper focuses on simulations that require much computer time, so it is important to select a design with a small number of observations. The proposed method is therefore sequential. The novelty of the method is that it accounts for the specific input/output function of the particular simulation model at hand; that is, the method is application-driven or customized. This customization is achieved through cross-validation and jackknifing. The new method is tested through two academic applications, which demonstrate that the method indeed gives better results than either sequential designs based on an approximate Kriging prediction variance formula or designs with prefixed sample sizes.     

Uniform designs for mixture-amount experiments and for mixture experiments under order restrictions

With order statistics of the uniform distribution on [0, l], exponential and beta distributions, a stochastic representation is obtained for the uniform distribution over various domains, where A-type domains are closely associated with reliability growth analysis, order restricted statistical inference and isotonic regression theory, V-type domains are connected with the mixture-amount experiments, and T-type domains are well related to mixture experiments. With these stochastic representations, the corresponding uniform distribution and number-theoretic nets can be generated. This approach seems to be new and is called order statistics method. Some examples on reliability growth analysis and experimental design are presented. This work was partially supported by a Hong Kong UGC-RGC grant, the Statistics Research and Consultancy Centre of HK-BU, and the Chinese Academy of Sciences.     

Constructions of optimal packing designs

Let v and k be positive integers. A (v, k, 1)-packing design is an ordered pair (V, B) where V is a v-set and B is a collection of k-subsets of V (called blocks) such that every 2-subset of V occurs in at most one block of B. The packing problem is mainly to determine the packing number P(k, v), that is, the maximum number of blocks in such a packing design. It is well known that P(k, v) ≤ ⌊v⌊(v − 1)/(k − 1)⌋/k⌋ = J(k, v) where ⌊×⌋ denotes the greatest integer y such that y ≤ x. A (v, k, 1)-packing design having J(k, v) blocks is said to be optimal. In this article, we develop some general constructions to obtain optimal packing designs. As an application, we show that P(5, v) = J(5, v) if v ≡ 7, 11 or 15 (mod 20), with the exception of v ∈ {11, 15} and the possible exception of v ∈ {27, 47, 51, 67, 87, 135, 187, 231, 251, 291}. © 1998 John Wiley & Sons, Inc. J Combin Designs 6: 245–260, 1998     

Robustness of optimal designs for correlated random variables

Suppose that Y = (Y_i) is a normal random vector with mean Xb and covariance σ²I_n, where b is a p-dimensional vector (b_j), X = (X_ij) is an n × p matrix with X_ij ∈ {−1, 1}; this corresponds to a factorial design with −1, 1 representing low or high level respectively, or corresponds to a weighing design with −1, 1 representing an object j with weight b_j placed on the left and right of a chemical balance respectively. E-optimal designs Z are chosen that are robust in the sense that they remain E-optimal when the covariance of Y_i, Y_i′ is ρ > 0 for i ≠ i′. Within a smaller class of designs similar results are obtained with respect to a general class of optimality criteria which include the A- and D-criteria.     

On optimal kernel choice for deconvolution

In this note we show that, from a conventional viewpoint, there are particularly close parallels between optimal-kernel-choice problems in non-parametric deconvolution, and their better-understood counterparts in density estimation and regression. However, other aspects of these problems are distinctly different, and this property leads us to conclude that “optimal” kernels do not give satisfactory performance when applied to deconvolution. This unexpected result stems from the fact that standard side conditions, which are used to ensure that the familiar kernel-choice problem has a unique solution, do not have statistically beneficial implications for deconvolution estimators. In consequence, certain “sub-optimal” kernels produce estimators that enjoy both greater efficiency and greater visual smoothness.     

Circular neighbor-balanced designs universally optimal for total effects

In many experiments, the performance of a subject may be affected by some previous treatments applied to it apart from the current treatment. This motivates the studies of the residual effects of the treatments in a block design. This paper shows that a circular block design neighbor-balanced at distances up toγ≤k - 1, where k is the block size, is universally optimal for total effects under the linear models containing the neighbor effects at distances up toγamong the class of all circular binary block designs. Some combinatorial approaches to constructing these circular block designs neighbor-balanced at distances up to k - 1 are provided.     

Constrained optimal discrimination designs for Fourier regression models

In this article, the problem of constructing efficient discrimination designs in a Fourier regression model is considered. We propose designs which maximize the power of the F-test, which discriminates between the two highest order models, subject to the constraints that the tests that discriminate between lower order models have at least some given relative power. A complete solution is presented in terms of the canonical moments of the optimal designs, and for the special case of equal constraints even more specific formulae are available.     

Economical experiments: Bayesian efficient experimental design

We propose and implement a Bayesian optimal design procedure. Our procedure takes as its primitives a class of parametric models of strategic behavior, a class of games (experimental designs), and priors on the behavioral parameters. We select the experimental design that maximizes the information from the experiment. We sequentially sample with the given design and models until only one of the models has viable posterior odds. A model which has low posterior odds in a small collection of models will have an even lower posterior odds when compared to a larger class, and hence we can dismiss it. The procedure can be used sequentially by introducing new models and comparing them to the models that survived earlier rounds of experiments. The emphasis is not on running as many experiments as possible, but rather on choosing experimental designs to distinguish between models in the shortest possible time period. We illustrate this procedure with a simple experimental game with one-sided incomplete information.We acknowledge the financial support from NSF grant #SES-9223701 to the California Institute of Technology. We also acknowledge the research assistance of Eugene Grayver who wrote the software for the experiments.     

A method of construction of optimal experimental designs for the circle

An optimal design concentrated at a minimum number of points is considered for a polynomial regression experiment on the sphere. For the case of a polynomial experiment on the circle it is possible to explicitly find the parameters of an optimal minimal design if any optimal design is known.Translated from Zapiski Nauchnykh Seminarov Leningradskogo Otdeleniya Matematicheskogo Instituta im. V. A. Steklova AN SSSR, Vol. 53, pp. 107–117, 1975.     

Sensitivity analysis for discrete optimal control problems

We present local sensitivity analysis for discrete optimal control problems with varying endpoints in the case when the customary regularity of boundary conditions can be violated. We study the behavior of the optimal solutions subject to parametric perturbations of the problem.     

Bayesian cooperative choice of strategies

This paper provides a formal framework within which to study cooperative behavior in the presence of incomplete information, and shows how far the known results in the static cooperative game theory can readily be applied to the proposed framework. The new concepts of Bayesian society, Bayesian strong equilibrium and Bayesian incentive compatible strong equilibrium are introduced and studied.     

Theory of optimal blocking for fractional factorial split-plot designs

The issue of optimal blocking for fractional factorial split-plot (FFSP) designs is considered under the two criteria of minimum aberration and maximum estimation capacity. The criteria of minimum secondary aberration (MSA) and maximum secondary estimation capacity (MSEC) are developed for discriminating among rival nonisomorphic blcoked FFSP designs. A general rule for identifying MSA or MSEC blocked FFSP designs through their blocked consulting designs is established.