RESUMO
Este trabalho objetivou avaliar diferentes estruturas da matriz de variâncias e covariâncias residual (Σ), quanto ao ajustamento de dados longitudinais via modelos mistos, em experimentos varietais de cana-de-açúcar. A seleção adequada desta matriz garante a escolha de um modelo mais representativo dos dados. Em cada modelagem, variou-se ainda a suposição associada aos efeitos de tratamentos (variedades), como fixos e aleatórios. Quatro ensaios varietais, conduzidos entre 2005 e 2009, em três localidades do Estado de Goiás, foram considerados. Cada experimento foi delineado em blocos casualizados com três ou quatro repetições. A variável resposta analisada foi toneladas de colmos por hectare (TCH). Para avaliar a qualidade de ajustamento dos modelos, foram utilizados o critério de informação de Akaike (AIC) e o teste da razão de verossimilhanças. Este último foi utilizado apenas para comparar modelos hierárquicos, tomados dois a dois. Constatou-se que as análises pelo modelo univariado clássico de parcelas subdivididas oscilaram entre as piores ou entre aquelas de ajustes apenas medianos. As estruturas da matriz Σ com os melhores ajustamentos variaram entre os ensaios, com destaque para a matriz não-estruturada. Tais resultados revelam que a estrutura de erros independentes, em geral, não se mostra adequada para esse tipo de análise e, também, que não é seguro definir previamente uma estrutura específica de Σ para tais ensaios. Pequenas alterações foram observadas na classificação das estruturas ao se assumirem os efeitos de tratamentos como fixos ou aleatórios; porém, sem efeito importante na classificação das melhores estruturas em cada ensaio.
This study aimed to evaluate different residual structures of variance-covariance matrix (Σ), regarding the fitting of longitudinal data via mixed models in variety trials of sugarcane. The adequate choice of this matrix provides most representative models to the data. In each model was also evaluated the effects of treatments (varieties), either as fixed or as random. Four trials were carried out in three locations in the Goiás State, Brazil, from 2005 to 2009. Each experiment was designed in randomized complete block with three or four repetitions. The response variable analyzed was tons of stalks per hectare (TCH). The goodness of fitting of the different models to the data was assessed by Akaike information criterion (AIC) and by likelihood ratio test (LRT). This last statistic was used only to compare nested models, two by two. It was observed that classic model in split-plot design ranged among the worst or with just median adjustments. The structures of Σ matrix with the best fittings to the data varied among trials, with outstanding for the unstructured matrix. These results show that the structure of independent errors, in general, is not adequate for these analyses, and a prior definition of the co-variance structure can lead to unreliable results for these trials. Small changes were observed in the ranking of these structures by assuming the treatment effects as fixed or random, however, without significant effects on the ranking of the best structures in each trial.
RESUMO
By definition, the genetic effects obtained from a circulant diallel table are random. However, because of the methods of analysis, those effects have been considered as fixed. Two different statistical approaches were applied. One assumed the model to be fixed and obtained solutions through the ordinary least square (OLS) method. The other assumed a mixed model and estimated the fixed effects (BLUE) by generalized least squares (GLS) and the best linear unbiased predictor (BLUP) of the random effects. The goal of this study was to evaluate the consequences when considering these effects as fixed or random, using the coefficient of correlation between the responses of observed and non-observed hybrids. Crossings were made between S1 inbred lines from two maize populations developed at Universidade Federal de Goiás, the UFG-Samambaia "Dent" and UFG-Samambaia "Flint". A circulant inter-group design was applied, and there were five (s = 5) crossings for each parent. The predictions were made using a reduced model. Diallels with different sizes of s (from 2 to 5) were simulated, and the coefficients of correlation were obtained using two different approaches for each size of s. In the first approach, the observed hybrids were included in both the estimation of the genetic parameters and the coefficient of correlation, while in the second a cross-validation process was employed. In this process, the set of hybrids was divided in two groups: one group, comprising 75 percent of the original group, to estimate the genetic parameters, and a second one, consisting of the remaining 25 percent, to validate the predictions. In all cases, a bootstrap process with 200 resamplings was used to generate the empirical distribution of the correlation coefficient. This coefficient showed a decrease as the value of s decreased. The cross-validation method allowed to estimate the bias magnitude in evaluating the correlation coefficient using the same hybrids, to predict the genetic parameters and the correlation evaluation. The bias was shown to be greater when the OLS method was used...