Influence of local values on cD-optimal designs for logistic models
Abstract
When experiments are designed, it is uncommon to use criteria to determine the treatments and number of replicates that should be conducted to properly estimate the parameters in the model under study. This is mainly due to a lack of knowledge of said criteria and, in many other cases, to the difficulty of interpreting them. Optimal designs try to overcome this difficulty by providing optimal experimental conditions and factor levels whose response should be evaluated in order to improve the quality of the statistical inference at a lower cost. Said designs also use optimality criteria, which are functions of the Fisher-information matrix. One of the most common estimation problems in nonlinear models is specifying local values for the parameters of the model, which is necessary to minimice the optimality criterion (King & Wong, 2000) [1]. This work examines the robustness of optimal designs obtained from logistic models when perturbations in the local values of the parameters are considered. The objective is to provide researchers with a wide range of action to select local values while guaranteeing that the efficiency of the resulting optimal designs is not considerably compromised compared to reference values. For that purpose, based on the data of an example, the efficiency of each resulting design was contrasted with the efficiency of unperturbed values. Furthermore, cD-optimal designs were created to estimate the logit variance. The impact of said perturbations on local cD-optimal designs resulted in an efficiency of approximately 70%, with a range of 0.04 of perturbation above the reference value.
References
[2] J. Kiefer, “Optimum Experimental Designs,” J. R. Stat. Soc., vol. 21, no. 2, pp. 272–319, 1959.
[3] K. Smith, “On the Standard Deviations of Adjusted and Interpolated Values of an Observed Polynomial Function and its Constants and the Guidance they give Towards a Proper Choice of the Distribution of Observations,” Biometrika, vol. 12, no. 1, p. 1, Nov. 1918.
[4] V. I. López and R. Ramos, “Una introducción a los diseños óptimos,” Rev. Colomb. Estadística, vol. 30, no. 1, pp. 37–51, 2007.
[5] T. Mathew and B. K. Sinha, “Optimal designs for binary data under logistic regression,” J. Stat. Plan. Inference, vol. 93, no. 1–2, pp. 295–307, Feb. 2001.
[6] P. Parsa Maram and H. Jafari, “Bayesian D-optimal design for logistic regression model with exponential distribution for random intercept,” J. Stat. Comput. Simul., vol. 86, no. 10, pp. 1856–1868, Jul. 2016.
[7] L. M. Haines, G. Kabera, P. Ndlovu, and T. E. O’Brien, “D-optimal Designs for Logistic Regression in Two Variables,” in mODa 8 - Advances in Model-Oriented Design and Analysis, Heidelberg: Physica-Verlag HD, pp. 91–98.
[8] M. A. Heise and R. H. Myers, “Optimal Designs for Bivariate Logistic Regression,” Biometrics, vol. 52, no. 2, p. 613, Jun. 1996.
[9] T. E. O’Brien and G. M. Funk, “A Gentle Introduction to Optimal Design for Regression Models,” Am. Stat., vol. 57, no. 4, pp. 265–267, Nov. 2003.
[10] V. V. Fedorov and P. Hackl, Model-Oriented Design of Experiments, 1st ed., vol. 125. New York, NY: Springer New York, 1997.
[11] S. D. Silvey, Optimal Design. Dordrecht: Springer Netherlands, 1980.
[12] A. Atkinson, A. Donev, and R. Tobias, Optimum Experimental Designs with SAS. OUP UK, 2007.
[13] J. Kiefer and J. Wolfowitz, “The equivalence of two extremum problems,” Can. J. Math., vol. 12, pp. 363–366, Jan. 1960.
[14] P. Whittle, “Some general points in the theory of optimal experimental design,” R. Stat. Soc., vol. 35, no. 1, pp. 123–130, 1973.
[15] C. F. Restrepo and V. I. López Ríos, “Elección de la constante de ponderación en diseños cD- óptimos compuestos,” Ing. y Cienc., vol. 9, no. 18, pp. 107–130, 2013.
[16] H. Chernoff, “Locally optimal designs for estimating parameters,” Ann. Math. Stat., vol. 24, no. 4, pp. 586–602, 1953.
[17] I. Ford, B. Torsney, and C. F. . Wu, “The Use of a canonical Form in the Construction of Locally Optimal Designs For Nonlinear Problems,” R. Stat. Siciety, vol. 54, no. 2, pp. 569–583, 1992.
[18] H. Dette, V. B. Melas, and A. Pepelyshev, “Optimal designs for a Class of Nonlinear Regression Models,” Ann. Stat., vol. 32, no. 5, pp. 2114–2167, 2004.
[19] H. Dette, V. B. Melas, and W. K. Wong, “Optimal Design for Goodness-of-Fit of the Michaelis–Menten Enzyme Kinetic Function,” J. Am. Stat. Assoc., vol. 100, no. 472, pp. 1370–1381, Dec. 2005.
[20] D. W. Hosmer and S. Lemeshow, Applied Logistic Regression, 2nd ed. John Wiley & Sons Inc, 2000.
[21] R. D. C. Team, “R: A Language and Environment for Statistical Computing,” 2011. [Online]. Available: https://www.gbif.org/tool/81287/r-a-language-and-environment-for-statistical-computing