Published
Locally D-Optimal Designs with Heteroscedasticity: A Comparison between Two Methodologies
Diseños D-óptimos locales con heterocedasticidad: una comparación entre dos metodologías
DOI:
https://doi.org/10.15446/rce.v37n1.44360Keywords:
D-efficiency, D-optimal design, Box-Cox transformations, Heteroscedasticity. (en)D-eficiencia, Diseños D-óptimos, heterocedasticidad, transformación de Box-Cox (es)
Downloads
The classic theory of optimal experimental designs assumes that the errors of the model are independent and have a normal distribution with constant variance. However, the assumption of homogeneity of variance is not always satisfied. For example when the variability of the response is a function of the mean, it is probably that a heterogeneity model be more adequate than a homogeneous one. To solve this problem there are two methods: The first one consists of incorporating a function which models the error variance in the model, the second one is to apply some of the Box-Cox transformations to both sides on the nonlinear regression model to achieve a homoscedastic model (Carroll & Ruppert 1988, Chapter 4). In both cases it is possible to find the optimal design but the problem becomes more complex because it is necessary to find an expression for the Fisher information matrix of the model. In this paper we present the two mentioned methodologies for the D-optimality criteria and we show a result which is useful to find D- optimal designs for heteroscedastic models when the variance of the response is a function of the mean. Then we apply both methods with an example, where the model is nonlinear and the variance is not constant. Finally we find the D-optimal designs with each methodology, calculate the efficiencies and evaluate the goodness of fit of the obtained designs via simulations.
La teoría clásica de los diseños experimentales óptimos supone que los errores del modelo son independientes y tienen una distribución normal con varianza constante. Sin embargo, el supuesto de homogeneidad de varianza no siempre se satisface. Por ejemplo, cuando la variabilidad de la respuesta es una función de la media, es probable que un modelo heterocedástico sea más adecuado que uno homogéneo. Para solucionar este problema hay dos métodos: el primero consiste en incorporar una función que modele la varianza del error en el modelo; el segundo consiste en aplicar alguna de las transformaciones de Box-Cox en el modelo de regresión no lineal (Carroll & Ruppert 1988, Capítulo 4). En ambos casos es posible hallar el diseño óptimo, pero el problema se vuelve más complejo porque es necesario encontrar una expresión de la matriz de información de Fisher del modelo. En este artículo se presentan las dos metodologías mencionadas para el criterio D-optimalidad y se muestra un resultado que es útil para encontrar diseños D-óptimos para modelos heterocedásticos cuando la varianza de la respuesta es una función de la media. Luego, se aplican ambos métodos en un ejemplo donde el modelo es no lineal y la varianza no constante. Finalmente se encuentra el diseño D-óptimo con cada metodología, se calculan las eficiencias y se evalúa la bondad del ajuste de los diseños obtenidos a través de simulaciones.
https://doi.org/10.15446/rce.v37n1.44360
1Universidad Nacional de Colombia, Facultad de Ciencias, Escuela de Estadística, Medellín, Colombia. MSc in Statistics. Email: jagaviriab@unal.edu.co
2Universidad Nacional de Colombia, Facultad de Ciencias, Escuela de Estadística, Medellín, Colombia. Associate professor. Email: vilopez@unal.edu.co
The classic theory of optimal experimental designs assumes that the errors of the model are independent and have a normal distribution with constant variance. However, the assumption of homogeneity of variance is not always satisfied. For example when the variability of the response is a function of the mean, it is probably that a heterogeneity model be more adequate than a homogeneous one. To solve this problem there are two methods: The first one consists of incorporating a function which models the error variance in the model, the second one is to apply some of the Box-Cox transformations to both sides on the nonlinear regression model to achieve a homoscedastic model (R.J. Carroll & D. Ruppert 1988, Chapter 4). In both cases it is possible to find the optimal design but the problem becomes more complex because it is necessary to find an expression for the Fisher information matrix of the model. In this paper we present the two mentioned methodologies for the D-optimality criteria and we show a result which is useful to find D-optimal designs for heteroscedastic models when the variance of the response is a function of the mean. Then we apply both methods with an example, where the model is nonlinear and the variance is not constant. Finally we find the D-optimal designs with each methodology, calculate the efficiencies and evaluate the goodness of fit of the obtained designs via simulations.
Key words: D-efficiency, D-optimal design, Box-Cox transformations, \linebreak Heteroscedasticity.
La teoría clásica de los diseños experimentales óptimos supone que los errores del modelo son independientes y tienen una distribución normal con varianza constante. Sin embargo, el supuesto de homogeneidad de varianza no siempre se satisface. Por ejemplo, cuando la variabilidad de la respuesta es una función de la media, es probable que un modelo heterocedástico sea más adecuado que uno homogéneo. Para solucionar este problema hay dos métodos: el primero consiste en incorporar una función que modele la varianza del error en el modelo; el segundo consiste en aplicar alguna de las transformaciones de Box-Cox en el modelo de regresión no lineal (R.J. Carroll & D. Ruppert 1988, Capítulo 4). En ambos casos es posible hallar el diseño óptimo, pero el problema se vuelve más complejo porque es necesario encontrar una expresión de la matriz de información de Fisher del modelo. En este artículo se presentan las dos metodologías mencionadas para el criterio D-optimalidad y se muestra un resultado que es útil para encontrar diseños D-óptimos para modelos heterocedásticos cuando la varianza de la respuesta es una función de la media. Luego, se aplican ambos métodos en un ejemplo donde el modelo es no lineal y la varianza no constante. Finalmente se encuentra el diseño D-óptimo con cada metodología, se calculan las eficiencias y se evalúa la bondad del ajuste de los diseños obtenidos a través de simulaciones.
Palabras clave: D-eficiencia, Diseños D-óptimos, heterocedasticidad, transformación de Box-Cox.
Texto completo disponible en PDF
References
1. A. C. Atkinson,, A. N. Donev, & R. D Tobias, (2007), Optimum Experimental Designs with SAS, Oxford Science Publications, New York.
2. A. C. Atkinson, & R. D. Cook, (1995), 'D-Optimum Designs for Heteroscedastic Linear Models', Journal of the American Statistical Association 90(429), 204-212.
3. A. C. Atkinson, & R. D. Cook, (1997), 'Designing for a Response Transformation Parameter', Journal of the Royal Statistical Society. Series B (Methodological) 59(1), 111-124.
4. Anthony C. Atkinson, (2003), 'Horwitz's Rule, Transforming Both Sides and the Design of Experiments for Mechanistic Models', Journal of the Royal Statistical Society. Series C (Applied Statistics) 52(3), pp. 261-278.
5. C. Ritz, & J. Streibig, (2008), Nonlinear Regression with R, Springer, New York.
6. D. Downing,, V. Fedorov, & S. Leonov, (2001), Extracting Information from the Variance Function: Optimal Design, Springer, Austria.
7. D. M. Bates, & D. G. Watts, (1988), Nonlinear Regression Analysis and its Applications, John Wiley and Sons, New York.
8. G. E. P. Box, & D. R. Cox, (1964), 'An Analysis of Transformations', Journal of the Royal Statistical Society. Series B (Methodological) 26(2), pp. 211-252.
9. G. Seber, & C. Wild, (1989), Nonlinear Regression, John Wiley, New York.
10. H. Dette, & K. Wong, (1999), 'Optimal Designs When the Variance Is a Function of the Mean', Biometrics 55(3), 925-929.
11. J. Kiefer, (1959), 'Optimum Experimental Designs', Journal of the Royal Statistical Society. Series B (Methodological) 21(2), pp. 272-319.
12. J. Kiefer, & J Wolfowitz, (1960), 'The equivalence of two extremum problems', Canadian Journal of Mathematics 12(5), pp. 363-365.
13. R Development Core Team, (2013), R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0. *http://www.R-project.org
14. R.J. Carroll, & D. Ruppert, (1988), Transformation and Weighting in Regression, Chapter 4, Taylor & Francis.
15. S. Huet,, A. Bouvier,, M. Poursat, & E. Jolivet, (2004), Statistical Tools for Nonlinear Regression: A Practical Guide With S-PLUS and R Examples , Springer-Verlag, New York.
16. T.E O'Brien, & G. M Funk, (2003), 'A gentle introduction to optimal design for regression models', Journal of the American Statistical Association 57(4), 265-267.
17. V López, & R Ramos, (2007), 'Introducción a los diseños óptimos', Revista Colombiana de Estadística 30(1), 37-51.
18. Weng Kee Wong, & R. Dennis Cook, (1993), 'Heteroscedastic G-Optimal Designs', Journal of the Royal Statistical Society. Series B (Methodological) 55(4), pp. 871-880.
Este artículo se puede citar en LaTeX utilizando la siguiente referencia bibliográfica de BibTeX:
@ARTICLE{RCEv37n1a07,
AUTHOR = {Gaviria, Jaime Andrés and López-Ríos, Víctor Ignacio},
TITLE = {{Locally D-Optimal Designs with Heteroscedasticity: A Comparison between Two Methodologies}},
JOURNAL = {Revista Colombiana de Estadística},
YEAR = {2014},
volume = {37},
number = {1},
pages = {95-110}
}
References
Atkinson, A. C. (2003), ‘Horwitz’s rule, transforming both sides and the design of experiments for mechanistic models’, Journal of the Royal Statistical Society. Series C (Applied Statistics) 52(3), pp. 261–278.
Atkinson, A. C. & Cook, R. D. (1995), ‘D-optimum designs for heteroscedastic linear models’, Journal of the American Statistical Association 90(429), 204–212.
Atkinson, A. C. & Cook, R. D. (1997), ‘Designing for a response transformation parameter’, Journal of the Royal Statistical Society. Series B (Methodological) 59(1), 111–124.
Atkinson, A. C., Donev, A. N. & Tobias, R. D. (2007), Optimum Experimental Designs with SAS, Oxford Science Publications, New York.
Bates, D. M. & Watts, D. G. (1988), Nonlinear Regression Analysis and its Applications, John Wiley and Sons, New York.
Box, G. E. P. & Cox, D. R. (1964), ‘An analysis of transformations’, Journal of the Royal Statistical Society. Series B (Methodological) 26(2), pp. 211–252.
Carroll, R. & Ruppert, D. (1988), Transformation and Weighting in Regression, Chapter 4, Taylor & Francis.
Dette, H. & Wong, K. (1999), ‘Optimal Designs When the Variance Is a Function of the Mean’, Biometrics 55(3), 925–929.
Downing, D., Fedorov, V. & Leonov, S. (2001), Extracting Information from the Variance Function: Optimal Design, Springer, Austria.
Huet, S., Bouvier, A., Poursat, M. & Jolivet, E. (2004), Statistical Tools for Nonlinear Regression: A Practical Guide With S-PLUS and R Examples , Springer-Verlag, New York.
Kiefer, J. (1959), ‘Optimum experimental designs’, Journal of the Royal Statistical Society. Series B (Methodological) 21(2), pp. 272–319.
Kiefer, J. & Wolfowitz, J. (1960), ‘The equivalence of two extremum problems’, Canadian Journal of Mathematics 12(5), pp. 363–365.
López, V. & Ramos, R. (2007), ‘Introducción a los diseños óptimos’, Revista Colombiana de Estadística 30(1), 37–51.
O’Brien, T. & Funk, G. M. (2003), ‘A gentle introduction to optimal design for regression models’, Journal of the American Statistical Association 57(4), 265–267.
R Development Core Team (2013), R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0.
Ritz, C. & Streibig, J. (2008), Nonlinear Regression with R, Springer, New York.
Seber, G. & Wild, C. (1989), Nonlinear Regression, John Wiley, New York.
Wong, W. K. & Cook, R. D. (1993), ‘Heteroscedastic G-optimal designs’, Journal of the Royal Statistical Society. Series B (Methodological) 55(4), pp. 871–880.
How to Cite
APA
ACM
ACS
ABNT
Chicago
Harvard
IEEE
MLA
Turabian
Vancouver
Download Citation
CrossRef Cited-by
1. Xiao Zhang, Gang Shen. (2024). On efficiency of locally D -optimal designs under heteroscedasticity and non-Gaussianity . Journal of Applied Statistics, 51(16), p.3292. https://doi.org/10.1080/02664763.2024.2346822.
2. Catalina Patiño-Bustamante, Víctor López-Ríos. (2020). Diseños Dπ-óptimos para modelos no lineales heteroscedásticos: un estudio de robustez. Ingeniería y Ciencia, 16(31), p.77. https://doi.org/10.17230/ingciencia.16.31.4.
3. Sebastian Kaminski, André Bardow, Kai Leonhard. (2016). The trade-off between experimental effort and accuracy for determination of PCP-SAFT parameters. Fluid Phase Equilibria, 428, p.182. https://doi.org/10.1016/j.fluid.2016.06.004.
Dimensions
PlumX
Article abstract page views
Downloads
License
Copyright (c) 2014 Revista Colombiana de Estadística

This work is licensed under a Creative Commons Attribution 4.0 International License.
- Authors retain copyright and grant the journal right of first publication with the work simultaneously licensed under a Creative Commons Attribution License that allows others to share the work with an acknowledgement of the work's authorship and initial publication in this journal.
- Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the journal's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgement of its initial publication in this journal.
- Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work (See The Effect of Open Access).






