Um estudo do teste não paramétrico de Kohli aplicado em Conjoint Analysis

• Ano de defesa: 2011
• Autor(a) principal: Mendes, André
• Orientador(a): Silva, Carlos Henrique Osório
• Banca de defesa: Lopes, Jaques Silveira , Nascimento, Moysés
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Viçosa
• Programa de Pós-Graduação: Mestrado em Estatística Aplicada e Biometria
• Departamento: Estatística Aplicada e Biometria
• País: BR
• Palavras-chave em Português: Estatísticas; Teste não paramétrico; Kohli; Conjoint Analysis
• Palavras-chave em Inglês: Statistics; Nonparametric test; Kohli; Conjoint Analysis
• Área do conhecimento CNPq: CNPQ::CIENCIAS AGRARIAS
• Link de acesso: http://locus.ufv.br/handle/123456789/4042

We conducted a consumers preference study with simulated data in order to compare Kohli’s (1988) nonparametric test, called h test, for assessing attributes significance in Conjoint Analysis (CA), with the usual ANOVA F test. We simulated preference rates given by 48 consumers to eight treatments formed by a full factorial combination scheme of 3 attributes (A, B and C) with two levels each. Our main goal was to try to understand the theoretical basis for the h test. Thus, we considered an additive CA model with no interaction and defined four scenarios with distinct Relative Importances (RI) for the attributes (and consequently for the range of the part-worths, PW): Scenario 1 – RIA = 60%, RIB = 30% and RIC = 10%; Scenario 2 – RIA = 40%, RIB = 40% and RIC = 20%; Scenario 3 – RIA = 35%, RIB = 35% and RIC = 30% and Scenario 4 – RIA = 5%, RIB = 45% and RIC = 50%. For each scenario we also generated the random error values of the CA model from two distinct probability distribution models, both with zero mean and with standard deviation equal to sigma (σ): the normal distribution and a non-normal U shaped distribution. In addition, for each distribution we also investigated the following sigma values (σ = 1,5; 2,0; 2,5; 3,0; 3,5 and 4.0). Results did not allow us to relate significance of an attribute by Kohli’s h test neither to (i) magnitude of the RI value, nor to (ii) range of PW’s in comparison to the σ value. Even under non normal data the h test did not give understandable results (in a practical sense). We concluded that the h test should not be recommended.