Computing A-optimal and E-optimal designs for regression models via semidefinite programming

In semidefinite programming (SDP), we minimize a linear objective function subject to a linear matrix being positive semidefinite. A powerful program, SeDuMi, has been developed in MATLAB to solve SDP problems. In this article, we show in detail how to formulate A-optimal and E-optimal design problems as SDP problems and solve them by SeDuMi. This technique can be used to construct approximate A-optimal and E-optimal designs for all linear and nonlinear regression models with discrete design spaces. In addition, the results on discrete design spaces provide useful guidance for finding optimal designs on any continuous design space, and a convergence result is derived. Moreover, restrictions in the designs can be easily incorporated in the SDP problems and solved by SeDuMi. Several representative examples and one MATLAB program are given.     

Some robust designs for polynomial regression models

The author identifies static optimal designs for polynomial regression models with or without intercept. His optimality criterion is an average between the D‐optimality criterion for the estimation of low‐degree terms and the D₈‐optimality criterion for testing the significance of higher degree terms. His work relies on classical results concerning canonical moments and the theory of continued fractions.     

Robust regression designs for approximate polynomial models

In this article, we consider robust designs for approximate polynomial regression models, by applying the theory of canonical moments. The design criterion, first given in Liu and Wiens (J. Statist. Planning Inference 64 (1997) 369), is to maximize the determinant of the information matrix subject to a side condition of bounding the bias arising from model misspecification. We give a new proof of, and extend, the main theorem in Liu and Wiens (op. cit.); in so doing we shed new light on the structure of this problem. New designs, with the further property of minimizing the generalized variance of the additional regression coefficients when an enlarged model is fitted, are derived and assessed. These provide additional robustness against uncertainty regarding the proper degree of the fitted polynomial response.     

Computing E-optimal polynomial regression designs

We consider the problem of computing E-optimal designs in heteroscedastic polynomial regression with not necessarily strictly positive efficiency function. Based on a relation between E- and c-optimal designs, a reasonable candidate for E-optimality is obtained from equioscillating weighted polynomials. Optimality of that candidate is easily checked, at least numerically. Moreover, nonoptimality of that design has some interesting consequences, e.g. on the support, which might be helpful to obtain the optimal design also in this case.For computing the candidate numerically we propose a procedure based on Remez's second algorithm. Convergence of that procedure is verified, extending a result of Studden and Tsay (1976). Numerical examples are presented for some efficiency functions.     

E-optimal designs for polynomial regression without intercept

We give all E-optimal designs for the mean parameter vector in polynomial regression of degree d without intercept in one real variable. The derivation is based on interplays between E-optimal design problems in the present and in certain heteroscedastic polynomial setups with intercept. Thereby the optimal supports are found to be related to the alternation points of the Chebyshev polynomials of the first kind, but the structure of optimal designs essentially depends on the regression degree being odd or even. In the former case the E-optimal designs are precisely the (infinitely many) scalar optimal designs, where the scalar parameter system refers to the Chebyshev coefficients, while for even d there is exactly one optimal design. In both cases explicit formulae for the corresponding optimal weights are obtained. Remarks on extending the results to E-optimality for subparameters of the mean vector (in heteroscdastic setups) are also given.     

Locally and maximin optimal designs for multi-factor nonlinear models

This paper considers the search for locally and maximin optimal designs for multi-factor nonlinear models from optimal designs for sub-models of a lower dimension. In particular, sufficient conditions are given so that maximin D-optimal designs for additive multi-factor nonlinear models can be built from maximin D-optimal designs for their sub-models with a single factor. Some examples of application are models involving exponential decay in several variables.     

D-optimum designs in multi-factor models with heteroscedastic errors

This paper considers the construction of D-optimum designs for Kronecker product and additive regression models when the errors are heteroscedastic. Sufficient conditions are given so that D-optimum designs for the multi-factor models can be built from D-optimum designs for their sub-models with a single factor. A robustness study is included to investigate how design efficiencies change when the efficiency functions are miss-specified.     

Discrete M-robust designs for regression models

M-robust designs are defined and constructed for misspecified linear regression models with possibly autocorrelated errors on a discrete design space. These designs minimize the mean-squared errors if linear regression models are correct with uncorrelated errors, subject to two robust constraints which control the change of the bias and the change of variance under model departures. Simulated annealing algorithm is applied to construct M-robust designs. Examples are given to show M-robust designs and compare them with minimax robust designs.     

D-optimal designs for Poisson regression models

Most of the current research on optimal experimental designs for generalized linear models focuses on logistic regression models. In this paper, D-optimal designs for Poisson regression models are discussed. For the one-variable first-order Poisson regression model, it has been found that the D-optimal design, in terms of effective dose levels, is independent of the model parameters. However, it is not the case for more complicated models. We investigate how the D-optimal designs depend on the model parameters for the one-variable second-order model and two-variable interaction model. The performance of some “standard” designs that appeal to practitioners is also studied.     

Optimal designs for beta-binomial logistic regression models

Optimal designs for a logistic regression model with over-dispersion introduced by a beta-binomial distribution are characterized. Designs are defined by a set of design points and design weights as usual but, in addition, the experimenter must also make a choice of a sub-sampling design specifying the distribution of observations on sample sizes. In an earlier work it has been shown that Ds-optimal sampling designs for estimation of the parameters of the beta-binomial distribution are supported on at most two design points. This admits a simplified approach using single sample sizes. Linear predictor values for Ds-optimal designs using a common sample size are tabulated for different levels of over-dispersion and choice of subsets of parameters.     

Bayesian D-optimal designs for exponential regression models

We consider the Bayesian D-optimal design problem for exponential growth models with one, two or three parameters. For the one-parameter model conditions on the shape of the density of the prior distribution and on the range of its support are given guaranteeing that a one-point design is also Bayesian D-optimal within the class of all designs. In the case of two parameters the best two-point designs are determined and for special prior distributions it is proved that these designs are Bayesian D-optimal. Finally, the exponential growth model with three parameters is investigated. The best three-point designs are characterized by a nonlinear equation. The global optimality of these designs cannot be proved analytically and it is demonstrated that these designs are also Bayesian D-optimal within the class of all designs if gamma-distributions are used as prior distributions.     

Optimal multi-criteria designs for Fourier regression models

Riccomagno, Schwabe and Wynn (RSW) (1997) have given a necessary and sufficient condition for obtaining a complete Fourier regression model with a design based on lattice points that is D-optimal. However, in practice, the number of factors to be considered may be large, or the experimental data may be restricted or not homogeneous. To address these difficulties we extend the results of RSW to obtain a sufficient condition for an incomplete interaction Fourier model design based on lattice points that is D-, A-, E- and G-optimal. We also propose an algorithm for finding such optimal designs that requires fewer design points than those obtained using RSW's generators when the underlying model is a complete interaction model.     

Optimal discrimination designs for exponential regression models

We investigate optimal designs for discriminating between exponential regression models of different complexity, which are widely used in the biological sciences; see, e.g., Landaw [1995. Robust sampling designs for compartmental models under large prior eigenvalue uncertainties. Math. Comput. Biomed. Appl. 181–187] or Gibaldi and Perrier [1982. Pharmacokinetics. Marcel Dekker, New York]. We discuss different approaches for the construction of appropriate optimality criteria, and find sharper upper bounds on the number of support points of locally optimal discrimination designs than those given by Caratheodory's Theorem. These results greatly facilitate the numerical construction of optimal designs. Various examples of optimal designs are then presented and compared to different other designs. Moreover, to protect the experiment against misspecifications of the nonlinear model parameters, we adapt the design criteria such that the resulting designs are robust with respect to such misspecifications and, again, provide several examples, which demonstrate the advantages of our approach.     

A-optimal designs for heteroscedastic multifactor regression models

Abstract

This paper searches for A-optimal designs for Kronecker product and additive regression models when the errors are heteroscedastic. Sufficient conditions are given so that A-optimal designs for the multifactor models can be built from A-optimal designs for their sub-models with a single factor. The results of an efficiency study carried out to check the adequacy of the products of optimal designs for uni-factor marginal models when these are used to estimate different multi-factor models are also reported.     

Restricted minimax robust designs for misspecified regression models

The authors propose and explore new regression designs. Within a particular parametric class, these designs are minimax robust against bias caused by model misspecification while attaining reasonable levels of efficiency as well. The introduction of this restricted class of designs is motivated by a desire to avoid the mathematical and numerical intractability found in the unrestricted minimax theory. Robustness is provided against a family of model departures sufficiently broad that the minimax design measures are necessarily absolutely continuous. Examples of implementation involve approximate polynomial and second order multiple regression.