Classification of common bean families under different scenarios of spatial dependence and experimental precision

Abstract

The objective of this work was to evaluate the efficiency of spatial statistical analysis, compared to the usual analyzes in randomized blocks and in lattice, for the classification of common bean (Phaseolus vulgaris) families under different scenarios of spatial dependence and experimental precision. Twelve scenarios were considered, formed by four classes of spatial dependence between errors (null, low, medium, and high) and three classes of experimental precision (medium, high, and very high). For the three classes of experimental precision, the following values of selective accuracy were defined: 0.60, 0.80, and 0.95, respectively. The four classes of spatial dependence between errors assumed, respectively, the following range values for the exponential geostatistical model: 0, 10, 20, and 40 m. In experiments with very high experimental precision or with the absence of spatial dependence, the efficiency of the spatial analysis for the classification of common bean families is similar to that of the usual analyses. In experiments with high and, especially, with medium experimental precision, the spatial analysis is more efficient than the usual analyses for the classification of common bean families, when the errors show spatial dependence.