Robust optimum design of experiments

• Grant number: 14/01818-0
• Support Opportunities: Regular Research Grants
• Start date: May 01, 2014
• End date: April 30, 2016
• Field of knowledge: Physical Sciences and Mathematics - Probability and Statistics - Applied Probability and Statistics
• Principal Investigator: Luzia Aparecida Trinca
• Grantee: Luzia Aparecida Trinca
• Host Institution: Instituto de Biociências (IBB). Universidade Estadual Paulista (UNESP). Campus de Botucatu. Botucatu , SP, Brazil
• Associated researchers: Heloisa Maria de Oliveira

Abstract

Methodologies for the construction of optimal designs bring major contributions to research in experimental areas, especially when there are material restrictions, human resources or runtime limitations. When the experiment involves several factors, the use of full factorial or regular fractions become nonviable. The theory of optimal designs allows searching the most efficient design possible given the practical constraints. Gilmour and Trinca (2012) enlarged the prospects in the area by formulating design criteria that focus on variance of the estimators as well as on estimation of pure error. They introduced a compound criteria able to focus on at least four objectives of the experiment. Compound criteria are very flexible since they can deal in a simple way the optimization of several desirable properties. The purpose of this project is to expand the class of compound criteria for optimal designs for factorial experiments. We recall the 14 desired properties highlighted by Box and Draper (2007) and examine the contributions of the discussant of Gilmour and Trinca (2012). Based on this we intend to formulate criterion that produces optimal designs that are more attractive to experimenters. In particular, in this project we propose to incorporate robustness properties of the design to the presence of influential observations, and hence robustness to the occurrence of loss of observations, to study the relevance of the inclusion of some property to control the imbalance of levels of the factors, as well as to extend the approach to experiments requiring more complex randomization scheme that leads to a mixed model. (AU)

Scientific publications

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

TRINCA, LUZIA A.; GILMOUR, STEVEN G.. Split-Plot and Multi-Stratum Designs for Statistical Inference. TECHNOMETRICS, v. 59, n. 4, p. 446-457, 2017. (14/01818-0)

DA SILVA, MARCELO A.; GILMOUR, STEVEN G.; TRINCA, LUZIA A.. Factorial and response surface designs robust to missing observations. COMPUTATIONAL STATISTICS & DATA ANALYSIS, v. 113, p. 261-272, SEP 2017. (14/01818-0)