Abstract
The approach considered in this chapter is probably the most common for designing experiments in nonlinear situations. It consists in optimizing a scalar function of the asymptotic covariance matrix of the estimator and thus relies on asymptotic normality, as considered in Chaps. 3 and 4. Design based on more accurate characterizations of the precision of the estimation will be considered in Chap. 6. Additionally to asymptotic normality, the approach also supposes that the asymptotic covariance matrix takes the form of the inverse of an information matrix, as it is the case, for instance, for weighted LS with optimum weights, see Sect. 3.1.3, or maximum likelihood estimation, see Sect. 4.2.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
Another interpretation is that locally optimum design is based on the Cramér–Rao bound and the Fisher information matrix; see Sect. 4.4.2. Note, however, that the Cramér–Rao inequality gives a lower bound, whereas an upper bound would be more suitable, but unfortunately is not available.
- 2.
When necessary, we shall extend the definition of a concave criterion Φ( ⋅) over the whole set \(\mathbb{M}\) of symmetric p ×p matrices by allowing Φ( ⋅) to take the value − ∞. All the criteria considered are proper functions; that is, they satisfy Φ(M) < ∞ for all \(\mathbf{M} \in \mathbb{M}\), and their effective domain \(\mathrm{dom}(\Phi ) =\{ \mathbf{M} \in \mathbb{M} : \Phi (\mathbf{M}) > -\infty \}\) is non empty.
- 3.
The reparameterization may also be nonlinear, with \(\mathbf{A} = \partial \beta /{\partial \theta }^{\top }\vert _{{\theta }^{0}}\), where θ 0 denotes the nominal value for θ used in locally optimum design.
- 4.
Up to a change of sign, since Kiefer considered criteria that should be minimized.
- 5.
From Cauchy–Schwarz inequality, ρ F (M) ≥ p with equality if and only if M is proportional to the identity matrix.
- 6.
For any \(\mathbf{M}_{1},\,\mathbf{M}_{2} \in \mathbb{M}\), M 1 ≽ M 2 (M 1 ≻ M 2) if and only if \(\mathbf{M}_{1} -\mathbf{M}_{2} \in {\mathbb{M}}^{\geq }\) (\({\mathbb{M}}^{>}\)).
- 7.
For any \(\mathbf{M}_{1},\,\mathbf{M}_{2} \in \mathbb{M}\), M 1 is better that M 2 according to Schur’s ordering if and only if \(\Phi _{E_{k}}(\mathbf{M}_{1}) \geq \Phi _{E_{k}}(\mathbf{M}_{2})\) for k = 1, …, p, see (5.12), with strict inequality for one k at least.
- 8.
It should be stressed here that neither positive homogeneity nor concavity do necessarily matter for the optimization of a differentiable criterion. Positive homogeneity is important when we wish to compare criteria through their efficiency. Equivalence with a concave criterion is crucial for preventing the existence of local maxima, but concavity itself is required in some particular situations only (typically, when the criterion is not differentiable). One may refer, e.g., to Avriel [2003, Chap. 6], for extensions of the notion of convexity.
- 9.
Notice that it implies that log[ − Φ c ( ⋅)] is convex on \(\mathbb{M}_{\mathbf{c}}^{\geq }\), which is a stronger results than Φ c ( ⋅) being concave on \(\mathbb{M}_{\mathbf{c}}^{\geq }\).
- 10.
Note that log[Φ q, I + (M)] is therefore concave on \({\mathbb{M}}^{>}\) for q ∈ ( − 1, ∞), so that log[trace(M − q)]1 ∕ q is convex, which is stronger than the convexity results mentioned, e.g., in [Kiefer, 1974] for [trace(M − q)]1 ∕ q.
- 11.
The now classical denomination “equivalence theorem” can be considered as a tribute to the work of Kiefer and Wolfowitz [1960]; we use this term throughout the book although “Necessary and Sufficient condition for optimality” would be appropriate too.
- 12.
ξ = ξ 1 ⊗ ξ 2 is the joint design measure on \(\mathcal{X} = \mathcal{X}_{1} \times \mathcal{X}_{2}\) with ξ(dx, dy) given by the product ξ 1(dx) ×ξ 2(dy) of the marginal measures, respectively, on \(\mathcal{X}_{1}\) and \(\mathcal{X}_{2}\).
- 13.
We have m = p(p + 1) ∕ 2 when the design criterion Φ( ⋅) is strictly isotonic since an optimum design matrix M ∗ (unique if Φ( ⋅) is strictly concave or is equivalent to a strictly concave criterion) necessarily lies on the boundary of \(\mathcal{M}_{\theta }(\Xi )\). Indeed, when M is in the interior of \(\mathcal{M}_{\theta }(\Xi )\), there exists α > 1 such that \(\alpha \mathbf{M} \in \mathcal{M}_{\theta }(\Xi )\) and Φ(α M) > Φ(M). From Caratheodory’s theorem, M ∗ can then be written as the linear combination of p(p + 1) ∕ 2 elements of \(\mathcal{M}_{\theta }(\mathcal{X})\).
- 14.
References
Alexéev, V., E. Galéev, and V. Tikhomirov (1987). Recueil de Problèmes d’Optimisation. Moscou: MIR.
Atwood, C. (1969). Optimal and efficient designs of experiments. Ann. Math. Statist. 40(5), 1570–1602.
Atwood, C. (1980). Convergent design sequences for sufficiently regular optimality criteria, II singular case. Ann. Statist. 8(4), 894–912.
Avriel, M. (2003). Nonlinear Programming. Analysis and Methods. New York: Dover. [Originally published by Prentice Hall, 1976].
Bloomfield, P. and G. Watson (1975). The inefficiency of least squares. Biometrika 62(1), 121–128.
Box, G. and H. Lucas (1959). Design of experiments in nonlinear situations. Biometrika 46, 77–90.
Chaloner, K. and I. Verdinelli (1995). Bayesian experimental design: a review. Statist. Sci. 10(3), 273–304.
Cook, D. and V. Fedorov (1995). Constrained optimization of experimental design (invited discussion paper). Statistics 26, 129–178.
Cook, D. and W. Wong (1994). On the equivalence between constrained and compound optimal designs. J. Amer. Statist. Assoc. 89(426), 687–692.
Cook, R., D. Hawkins, and S. Weisberg (1993). Exact iterative computation of the robust multivariate minimum volume ellipsoid estimator. Statist. Probab. Lett. 16, 213–218.
Cramér, H. (1946). Mathematical Methods of Statistics. Princeton, N.J.: Princeton University Press.
Davies, L. (1992). The asymptotics of Rousseeuw’s minimum volume ellipsoid estimator. Ann. Statist. 20(4), 1828–1843.
de la Garza, A. (1954). Spacing of information in polynomial regression. Ann. Math. Statist. 25, 123–130.
Dem’yanov, V. and V. Malozemov (1974). Introduction to Minimax. New York: Dover.
Dette, H. (1993). Elfving’s theorem for D-optimality. Ann. Statist. 21(2), 753–766.
Dette, H. and V. Melas (2011). A note on de la Garza phenomenon for locally optimal designs. Ann. Statist. 39(2), 1266–1281.
Dette, H. and W. Studden (1993). Geometry of E-optimality. Ann. Statist. 21(1), 416–433.
Dragalin, V. and V. Fedorov (2006). Adaptive designs for dose-finding based on efficacy-toxicity response. J. Stat. Plann. Inference 136, 1800–1823.
Elfving, G. (1952). Optimum allocation in linear regression. Ann. Math. Statist. 23, 255–262.
Fedorov, V. (1971). The design of experiments in the multiresponse case. Theory Probab. Appl. 16(2), 323–332.
Fedorov, V. (1972). Theory of Optimal Experiments. New York: Academic Press.
Fedorov, V. (1989). Optimal design with bounded density: optimization algorithms of the exchange type. J. Stat. Plann. Inference 22, 1–13.
Fedorov, V. and P. Hackl (1997). Model-Oriented Design of Experiments. Berlin: Springer.
Fedorov, V. and A. Pázman (1968). Design of physical experiments. Fortschritte der Physik 16, 325–355.
Fellman, J. (1974). On the allocation of linear observations. Comment. Phys. Math. 44, 27–78.
Galil, Z. and J. Kiefer (1977). Comparison of simplex designs for quadratic mixture models. Technometrics 19(4), 445–453.
Hadjihassan, S., L. Pronzato, E. Walter, and I. Vuchkov (1997). Robust design for quality improvement by ellipsoidal bounding. In G. Yin and Q. Zhang (Eds.), Proc. 1996 AMS–SIAM Summer Seminar, Math. of Stochastic Manufacturing Systems, Lectures in Applied Math., vol. 33, Providence, Rhode Island, pp. 127–138. American Math. Soc.
Harman, R. (2004a). Lower bounds on efficiency ratios based on ϕ p -optimal designs. In A. Di Bucchianico, H. Läuter, and H. Wynn (Eds.), mODa’7 – Advances in Model–Oriented Design and Analysis, Proc. 7th Int. Workshop, Heeze (Netherlands), Heidelberg, pp. 89–96. Physica Verlag.
Harman, R. (2004b). Minimal efficiency of designs under the class of orthogonally invariant information criteria. Metrika 60, 137–153.
Harman, R. and T. Jurík (2008). Computing c-optimal experimental designs using the simplex method of linear programming. Comput. Statist. Data Anal. 53, 247–254.
Harman, R. and M. Trnovská (2009). Approximate D-optimal designs of experiments on the convex hull of a finite set of information matrices. Math. Slovaca 59(5), 693–704.
Harville, D. (1997). Matrix Algebra from a Statistician’s Perspective. Heidelberg: Springer.
Karlin, S. and W. Studden (1966). Optimal experimental designs. Ann. Math. Statist. 37, 783–815.
Khachiyan, L. (1979). A polynomial algorithm in linear programming. Doklady Akademïa Nauk SSSR 244, 1093–1096. [English transl. Soviet Math. Doklady, 20, 191–194].
Khachiyan, L. (1996). Rounding of polytopes in the real number model of computation. Math. Oper. Res. 21(2), 307–320.
Khachiyan, L. and M. Todd (1993). On the complexity of approximating the maximal inscribed ellipsoid for a polytope. Math. Programming A61(2), 137–159.
Kiefer, J. (1962). Two more criteria equivalent to D-optimality of designs. Ann. Math. Statist. 33(2), 792–796.
Kiefer, J. (1974). General equivalence theory for optimum designs (approximate theory). Ann. Statist. 2(5), 849–879.
Kiefer, J. and J. Wolfowitz (1960). The equivalence of two extremum problems. Canad. J. Math. 12, 363–366.
Liu, S. and H. Neudecker (1997). Kantorovich inequalities and efficiency comparisons for several classes of estimators in linear models. Statistica Neerlandica 51(3), 345–355.
López-Fidalgo, J. and J. Rodríguez-Díaz (1998). Characteristic polynomial criteria in optimal experimental design. In A. Atkinson, L. Pronzato, and H. Wynn (Eds.), Advances in Model–Oriented Data Analysis and Experimental Design, Proc. MODA’5, Marseilles, June 22–26, 1998, pp. 31–38. Heidelberg: Physica Verlag.
López-Fidalgo, J., B. Torsney, and R. Ardanuy (1998). MV-optimisation in weighted linear regression. In A. Atkinson, L. Pronzato, and H. Wynn (Eds.), Advances in Model–Oriented Data Analysis and Experimental Design, Proc. MODA’5, Marseilles, June 22–26, 1998, pp. 39–50. Heidelberg: Physica Verlag.
Magnus, J. and H. Neudecker (1999). Matrix Differential Calculus, with Applications in Statistics and Econometrics. New York: Wiley.
Mikulecká, J. (1983). On a hybrid experimental design. Kybernetika 19(1), 1–14.
Müller, C. and A. Pázman (1998). Applications of necessary and sufficient conditions for maximin efficient designs. Metrika 48, 1–19.
Pázman, A. (1986). Foundations of Optimum Experimental Design. Dordrecht (co-pub. VEDA, Bratislava): Reidel (Kluwer group).
Pázman, A. (2001). Concentration sets, Elfving sets and norms in optimum design. In A. Atkinson, B. Bogacka, and A. Zhigljavsky (Eds.), Optimum Design 2000, Chapter 10, pp. 101–112. Dordrecht: Kluwer.
Pázman, A. (2002a). Optimal design of nonlinear experiments with parameter constraints. Metrika 56, 113–130.
Pečarić, J., S. Puntanen, and G. Styan (1996). Some further matrix extensions of the Cauchy-Schwarz and Kantorovich inequalities, with some statistical applications. Linear Algebra Appl. 237/238, 455–476.
Pilz, J. (1983). Bayesian Estimation and Experimental Design in Linear Regression Models, Volume 55. Leipzig: Teubner-Texte zur Mathematik. [Also published by Wiley, New York, 1991].
Polak, E. (1987). Optimization. Algorithms and Consistent Approximations. New York: Springer.
Pronzato, L. (2004). A minimax equivalence theorem for optimum bounded design measures. Statist. Probab. Lett. 68, 325–331.
Pronzato, L. (2006). On the sequential construction of optimum bounded designs. J. Stat. Plann. Inference 136, 2783–2804.
Pronzato, L. (2009b). On the regularization of singular c-optimal designs. Math. Slovaca 59(5), 611–626.
Pronzato, L. (2010b). Penalized optimal designs for dose-finding. J. Stat. Plann. Inference 140, 283–296.
Pronzato, L., C.-Y. Huang, and E. Walter (1991). Nonsequential T-optimal design for model discrimination: new algorithms. In A. Pázman and J. Volaufová (Eds.), Proc. ProbaStat’91, Bratislava, pp. 130–136.
Pronzato, L. and E. Walter (1994). Minimal-volume ellipsoids. Internat. J. Adapt. Control Signal Process. 8, 15–30.
Pronzato, L. and E. Walter (1996). Volume-optimal inner and outer ellipsoids. In M. Milanese, J.-P. Norton, H. Piet-Lahanier, and E. Walter (Eds.), Bounding Approaches to System Identification, pp. 119–138. London: Plenum.
Pronzato, L., H. Wynn, and A. Zhigljavsky (2000). Dynamical Search. Boca Raton: Chapman & Hall/CRC.
Pronzato, L., H. Wynn, and A. Zhigljavsky (2005). Kantorovich-type inequalities for operators via D-optimal design theory. Linear Algebra Appl. 410, 160–169.
Pukelsheim, F. (1993). Optimal Experimental Design. New York: Wiley.
Raynaud, H.-F., L. Pronzato, and E. Walter (2000). Robust identification and control based on ellipsoidal parametric uncertainty descriptions. Eur. J. Control 6(3), 245–257.
Rényi, A. (1961). On measures of entropy and information. In Proc. 4th Berkeley Symp. on Math. Statist. and Prob., pp. 547–561.
Rockafellar, R. (1970). Convex Analysis. Princeton: Princeton Univ. Press.
Sahm, M. and R. Schwabe (2001). A note on optimal bounded designs. In A. Atkinson, B. Bogacka, and A. Zhigljavsky (Eds.), Optimum Design 2000, Chapter 13, pp. 131–140. Dordrecht: Kluwer.
Schwabe, R. (1995). Designing experiments for additive nonlinear models. In C. Kitsos and W. Müller (Eds.), MODA4 – Advances in Model-Oriented Data Analysis, Spetses (Greece), june 1995, pp. 77–85. Heidelberg: Physica Verlag.
Schwabe, R. (1996). Optimum Designs for Multi-Factor Models. New York: Springer.
Schwabe, R. (1997). Maximin efficient designs. Another view at D-optimality. Statist. Probab. Lett. 35, 109–114.
Schwabe, R. and W. Wong (1999). Efficiency bounds for product designs in linear models. Ann. Inst. Statist. Math. 51, 723–730.
Shor, N. and O. Berezovski (1992). New algorithms for constructing optimal circumscribed and inscribed ellipsoids. Optim. Methods Softw. 1, 283–299.
Sibson, R. (1972). Discussion on a paper by H.P. Wynn. J. Roy. Statist. Soc. B34, 181–183.
Sibson, R. (1974). D A -optimality and duality. Progress in Stat. Colloq. Math. Soc. Janos Bolyai 9, 677–692.
Silvey, S. (1980). Optimal Design. London: Chapman & Hall.
Silvey, S. and D. Titterington (1973). A geometric approach to optimal design theory. Biometrika 60(1), 21–32.
Studden, W. (1971). Elfving’s theorem and optimal designs for quadratic loss. Ann. Math. Statist. 42, 1613–1621.
Sun, P. and R. Freund (2004). Computation of minimum-volume covering ellipsoids. Oper. Res. 52(5), 690–706.
Tarasov, S., L. Khachiyan, and I. Erlich (1988). The method of inscribed ellipsoids. Soviet Math. Dokl. 37(1), 226–230.
Titterington, D. (1975). Optimal design: some geometrical aspects of D-optimality. Biometrika 62(2), 313–320.
Titterington, D. (1978). Estimation of correlation coefficients by ellipsoidal trimming. J. Roy. Statist. Soc. C27(3), 227–234.
Todd, M. and E. Yildirim (2007). On Khachiyan’s algorithm for the computation of minimum volume enclosing ellipsoids. Discrete Appl. Math. 155, 1731–1744.
Torsney, B. (1986). Moment inequalities via optimal design theory. Linear Algebra Appl. 82, 237–253.
Vandenberghe, L., S. Boyd, and S.-P. Wu (1998). Determinant maximisation with linear matrix inequality constraints. SIAM J. Matrix Anal. Appl. 19(2), 499–533.
Welzl, E. (1991). Smallest enclosing disks (balls and ellipsoids). In New Results and New Trends in Computer Science, Lecture Notes in Comput. Sci., vol. 555, pp. 359–370. Heidelberg: Springer.
Wynn, H. (1977). Optimum designs for finite populations sampling. In S. Gupta and D. Moore (Eds.), Statistical Decision Theory and Related Topics II, pp. 471–478. New York: Academic Press.
Wynn, H. (1982). Optimum submeasures with applications to finite population sampling. In S. Gupta and J. Berger (Eds.), Statistical Decision Theory and Related Topics III. Proc. 3rd Purdue Symp., vol. 2, pp. 485–495. New York: Academic Press.
Yang, M. (2010). On de la Garza phenomenon. Ann. Statist. 38(4), 2499–2524.
Yang, M. and J. Stufken (2009). Support points of locally optimal designs for nonlinear models with two parameters. Ann. Statist. 37(1), 518–541.
Zhang, Y. and L. Gao (2003). On numerical solution of the maximum volume ellipsoid problem. SIAM J. Optim. 14(1), 53–76.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2013 Springer Science+Business Media New York
About this chapter
Cite this chapter
Pronzato, L., Pázman, A. (2013). Local Optimality Criteria Based on Asymptotic Normality. In: Design of Experiments in Nonlinear Models. Lecture Notes in Statistics, vol 212. Springer, New York, NY. https://doi.org/10.1007/978-1-4614-6363-4_5
Download citation
DOI: https://doi.org/10.1007/978-1-4614-6363-4_5
Published:
Publisher Name: Springer, New York, NY
Print ISBN: 978-1-4614-6362-7
Online ISBN: 978-1-4614-6363-4
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)