Escolha dos níveis nutricionais na determinação do nível-ótimo e no ajuste de modelos estatísticos utilizados em ensaios dose-resposta

Carregando...
Imagem de Miniatura

Data

2010-02-24

Autores

Souza, Fernando Augusto de [UNESP]

Título da Revista

ISSN da Revista

Título de Volume

Editor

Universidade Estadual Paulista (Unesp)

Resumo

Este trabalho avaliou a influência da heterocedasticidade e dos níveis nutricionais (número e posição) utilizados em ensaios dose-resposta, na estimativa do nível-ótimo e no ajuste dos modelos, além de verificar o quão informativas são as estatísticas utilizadas para avaliar a precisão do ajuste (R², R² ajustado, CV e SQD). Utilizaram-se dados dos experimentos realizados por Nascimento et al. (2007) e Siqueira (2009) e dados simulados. Constatou-se que, quando os níveis estiveram distribuídos próximos do verdadeiro requerimento, os modelos com platô proporcionaram resultados mais confiáveis. Já os modelos quadrático e exponencial se mostraram mais adequados para situações no qual os níveis estão mais dispersos em relação ao verdadeiro requerimento. A heterocedasticidade não interferiu na estimativa do nível-ótimo, porém influenciou no ajuste dos modelos e proporcionou pequenas mudanças nos parâmetros das equações obtidas. O coeficiente de determinação (ajustado e não ajustado) foi diretamente influenciado pela definição do nível mais próximo do ótimo e dos níveis extremos, enquanto que, o coeficiente de variação e a soma dos quadrados dos desvios, pelos níveis iniciais e pelo nível próximo do ótimo. A soma dos quadrados dos desvios demonstrou ser mais sensível, pois seu valor apresentou pequenas variações entre os modelos nas diferentes situações, que as outras estatísticas não detectaram. Ressalta-se a importância de se estabelecer corretamente o intervalo dos níveis estudados para que a dispersão dos valores do nível-ótimo estimado seja minimizada e que o ajuste seja satisfatório, independente do modelo utilizado
This work evaluated the influence of heteroskedasticity and the nutritional levels (number and position) used in dose-response trials to estimating the optimal-level and the adjustment of the models, also check how informative are the statistics used to evaluate the accuracy of fit (R ², R ² adjusted, CV and SQD). The data used in this experiment are from Nascimento et al. (2007) and’ Siqueira (2009) trials and simulated data. It was found that when levels were distributed close to the real requirement, the models with plateau have provided more reliable results. Since the quadratic and exponential models were more suitable for situations in which the levels are more dispersed about the real requirement. The heteroskedasticity did not affect the estimate of the level-optimal, but influenced the adjustment of the models and provided small changes in the parameters of the equations obtained. The coefficient of determination (adjusted and unadjusted) was directly influenced by the definition of the level closest to the optimum and extreme levels, while the coefficient of variation and the sum of squares of deviations were influenced by the initial levels and the level close to the optimum. The sum of squares of deviations was more sensitive, because its value showed small variations between models in different situations that the other statistics did not detect. Emphasized the importance to precisely define the range of levels studied to the dispersion of the obtained optimal-level is minimized and the fit is satisfactory regardless of the model

Descrição

Palavras-chave

Análise de regressão, Ensaios dose-resposta, Heterocedasticidade, Níveis nutricionais, LRP, Regression analysis, Dose-response trials, Heteroskedasticity, Nutritional levels

Como citar

SOUZA, Fernando Augusto de. Escolha dos níveis nutricionais na determinação do nível-ótimo e no ajuste de modelos estatísticos utilizados em ensaios dose-resposta. 2010. vii, 68 f. Dissertação (mestrado) - Universidade Estadual Paulista, Faculdade de Ciências Agrárias e Veterinárias, 2010.