Simulation of coffee quality using ABIC criteria in discriminating regions with different spatial dependency structures

Abstract

The paper addresses the application of discriminant analysis in proportion data with spatial dependence, focusing on the discrimination of specialty coffees in simulated microregions. Using Monte Carlo simulations, the study explores scenarios with varying levels of spatial dependence and link functions, comparing the effects of semivariogram models (spherical, exponential, and Gaussian) on the accuracy and precision of the analyses. The results indicate that quadratic discriminant analysis tends to be more effective for smaller samples (n=25), while strong spatial dependence favors the exponential model, improving specificity and reducing the false positive rate as n increases. When employing the complementary log-log link function, similar patterns are observed, with the Gaussian model performing best in larger samples. The study highlights the importance of selecting the appropriate model based on factors such as spatial dependence, sample size, and link function to ensure accurate results. The balance between specificity and sensitivity is crucial in the final decision, emphasizing the need for a careful approach to the analysis of proportion data with spatial dependence.