Critérios compostos para delineamentos ótimos robustos

• Ano de defesa: 2014
• Autor(a) principal: Silva Marcelo Andrade da [UNESP]
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: Biometria; Planejamento experimental; Programação heuristica; Estatistica matematica; Programação (Matemática); Heuristic programming
• Link de acesso: http://hdl.handle.net/11449/110368

In this work we propose the use of a robustness measure to missing data to construct designs for factorial experiments. The robustness property is denoted the H criterion and it is added to a compound design criterion expression. Two versions of the modified exchange algorithm of Fedorov (1972) were implemented computationally for the search of exact optimum designs. Four examples are presented, examples 1, 3 and 4 consider the full second-order model and example 2 considers second-order model excluding the quadratic effects. The examples 1 and 3, in order to preserve good efficiency with respect to other properties, their H efficiency is not high. The results for example 2 showed good performance of the new compound criterion since it produced designs high by efficient for all other properties. In general, the new compound criterion produced more attractive designs than the DP criterion of Gilmour & Trinca (2012) since their leverages were more homogeneous and thus, the designs were more robust to missing data. The designs were also more attractive than those constructed by subsets as in Ahmad & Gilmour (2010)