Implementação do modelo AR(1) X AR(1) na análise de experimentos de campo com dependência espacial para o pacote spanova

Abstract

In field experiments, statistical analysis is often challenged by spatial heterogeneity, in which dependence among experimental units represents a critical aspect that, if not properly modeled, may compromise the reliability of inferences. Traditional methods are limited when faced with the complexity of agricultural data that exhibit spatial dependence structures. In this context, this dissertation presents the development and computational integration of the AR(1) × AR(1) spatial covariance structure into the spANOVA package of the R software. The AR(1) × AR(1) model is employed due to its ability to efficiently model bidimensional residual autocorrelation in regular layouts, aiming at a more realistic representation of the data structure. The main objective is to expand the functionalities of spANOVA, enabling researchers to specify this model in mixedmodel analyses. The proposed methodology includes the computational development and validation of the AR(1) × AR(1) function, its integration into the package, and performance evaluation through simulations and applications to real datasets, using Restricted Maximum Likelihood (REML) for parameter estimation, with the aim of improving the analysis of agricultural data with spatial dependence.