Modelagem de regressão resistente a outliers verticais e horizontais das capturas de camarões no estado do Rio de Janeiro

Abstract

This work proposes the application of simple linear regression modeling techniques to three shrimp species in the State of Rio de Janeiro (seven-beard, pink, and whitebeard). The research is developed using the Least Absolute Deviations (LAD) method in fitting linear regression models, which aims to eliminate the influence of discrepant data outside the vertical and horizontal axes present in the studied dataset. The study follows the approach outlined in Dodge's (1997) article, as it is a model resistant to the presence of outliers. In this model, the identification and suppression of extreme points are performed on standardized residuals using the R software for statistical analysis. The results show that the exclusion of extreme points has a relevant effect on the accuracy of regression coefficient estimates, reducing residual variability and improving the robustness of the models. However, it is worth noting that the removal of outliers should be justified by a reasonably clear reason. Nevertheless, correlations were established between the species, which showed strong positive relationships and moderately negative relationships to varying degrees depending on each case. This study contributes to the understanding of techniques resistant to the presence of outliers in the database of shrimp species catches and provides significant results regarding the technique from Dodge’s (1997) article.