Estimação de contrastes de médias de tratamentos, de um experimento em blocos ao acaso, utilizando as análises clássica e espacial / Estimation of treatments means contrasts, in a random blocks model, using the classical and spatial analysis

AUTOR(ES)
DATA DE PUBLICAÇÃO

2008

RESUMO

Not controllable factors is common occur in experiments, they are responsible for the heterogeneity among parcels. Even executing the three experimentation basic principles in the design (repetition, randomization and local control), even so, may have correlation in errors and, therefore, spatial dependence in the area of study. If that autocorrelation structure is detected and if this information is used in statistical analysis, estimates more efficient of contrasts among treatments means are guaranteed, but if this structure is disregarded can prevent that real diferences are detected. In this work, the coordinates of parcels centers in a design of random blocks were observed. The concentration of soil organic carbon is the response variable of this experiment, with the available made at the beginning of the experiment, ie, before the treatments application, therefore, a blank, a year after the treatments application and, again, after a year. Then, the classical and spatial analysis were used to compare the methods of estimation of treatments means contrasts. The method studied for the classical analysis, which considers that the errors are not correlated, was the ordinary least squares. For the analysis, considering the spatial dependence, were used the geostatistical model, where consists in the addition of a random effect with correlation, and the Papadakis model, which consists in the addition of a covariate built from observations in neighbouring. In geostatistical model was verified the spatial dependence through the Akaike and Bayesian or Schwarz criteria of information and the methods tested were the variogram followed by generalized least squares and the maximum likelihood. For the Papadakis model, was tested the significance of covariate referring to the average of residuals among neighbouring parcels and own parcel in the random blocks model and in the completely randomized model, and the test was not significant in any of both cases. Still, the calculus were made for this method, showing that for this data set, this method is not indicated. Using some measures to compare these methods, for these data, the method of estimation of treatments means contrasts which presented the measures of comparison more dispersed was the Papadakis model and the less dispersed was the maximum likelihood. Still, in the confidence intervals, it was observed that in spatial analysis other contrasts di®ered from zero significantly, besides of those which were observed in classical analysis, which concludes that when the autocorrelation of errors is considering, the contrasts are estimated with greater e±ciency.

ASSUNTO(S)

geoestatística geostatistics estatísticas espaciais likelihood. spatial statistics modelos matemáticos least squares verossimilhança. mínimos quadrados data modelling modelagem de dados mathematical models

ACESSO AO ARTIGO

Documentos Relacionados