Empirical evaluation of the gamma-Pareto regression residuals for influence diagnostics

For the reliable results of regression analysis, influence diagnostic methods play an important role. The performance of various proposed residuals for the gamma-Pareto regression model (G-PRM) is presented in this study. These proposed residuals are partitioned into standardized and adjusted residual forms. For the detection of influential observations, difference of fits (DFFITS) was used in both standardized and adjusted residuals. To detect influential points, these residuals are compared through Monte Carlo Simulation also real data (Ardennes data) are analysed to show the benefits of the proposed methods. The likelihood residuals performed better for small dispersion than others. Additionally, all forms of adjusted residuals are the same, not much more efficient than the standardized form. All standardized residuals behave in the same way for wider dispersion, and they are superior to the likelihood residuals for the detection of influential points.