Comparison of Statistical Methodologies for Philanthropic Studies

Philanthropic data related to charitable donations is not homogeneously distributed. Outliers, are often present, corresponding to small and large donation amounts. Modeling these specific features necessitates the application of tailored statistical methodologies designed to incorporate extremes in the distribution. Standard and robust regressions employing M-estimation will be compared and discussed alongside forecasting procedures using a simulated dataset derived from a real-world application. The real-world application involves tax data from the Canton of Geneva from 2009-2011. A randomly drawn subsample from the donor dataset comprises 100000 observations with three variables. This subsample has been estimated and predicted for a specific illustrative scope. When the purpose is estimation, robust regression techniques with M-estimation are the most effective for handling this type of data. Conversely, when the goal is prediction, ARIMA models tend to incorporate the peculiar characteristics of the data better, leading to more accurate forecasts in the future.