In competition experiments of citrus cultivations one generally uses many treatments, which requires the use of big blocks and plots with few plants. One has debated that in these conditions there can occur the correlation between neighboring plots, violating, thus, the presupposition of errors independent from the variance analysis. The present work has had as objective to evaluate different model parametrizations, considering or not the spatial dependence between plots, in two competition experiments of Pera orange tree clones (Citrus sinensis L. Osbeck). One has utilized the separable auto-regressive structure of first order (AR1 x AR1) as a model of spatial dependence between the errors. The results found indicate that the spatial modeling of the errors by utilizing separable auto-regressive models of first order for selection experiments of Pera orange tree clones normally bring small gains in terms of quality of adjustment. The analysis not considering the block factor plus the separable auto-regressive spatial adjustment of first order has presented better quality of adjustment between the models evaluated.
Citrus sinensis; productivity; canopy volume; spatial autocorrelation; autoregressive models; plant breeding
