Fitting a response surface in an incomplete block design: Restricted likelihood and Bayesian analysis aplied to a manuring simulation in citrus

Claudiney Nunes de LIMA[1]

Júlio Sílvio de Sousa BUENO FILHO1

§    ABSTRACT: Finding the combination of nutrient levels that yield maximum production is an interesting problem in manuring assays. The aim of this work was to illustrate the restrict likelihood analysis (REML) and Bayesian analysis (GS, or via Gibbs Sampler) of a partially balanced incomplete block design. For this illustration, production data for the first year of a citrus culture were simulated, using combinations of levels for two nutrients. Model fitting and statistical properties were worked out trough REML and GS analysis of a mixed model. Results from both methodologies were equivalent, indicating that they are good tools for mixed models fitting that should be recommended for response surface estimation.

§    KEYWORDS: Gibbs sampling; incomplete block designs; manuring; REML.



[1]Departamento de Ciências Exatas, Universidade Federal de Lavras, Caixa Postal 3037, CEP: 37200-000, Lavras, MG, Brasil, E-mail: pleudiney@yahoo.com.br / julio@stout.ufla.br