Year of Graduation
Quality of Linear Regression Basing on Assumption About Distribution of Standardized Residuals: Analysis of Skewness and Kurtosis
Applied Methods of Social Analysis of Markets
This study investigates distributions of standardized residuals in OLS – regressions. The issue is that researchers usually do not know how far they are allowed to violate normality of residuals without loosing in quality of their models. Therefore, they either ignore the problem emerged or switch to more complicated models, loosing one of the most useful tool to predict independent variable. Key objective of the study is to propose boarders and practical recommendations of main steps in solving the problem. As a result, paper illustrates that models with different level of prediction have different stability towards kurtosis and skewness of residual distribution. Basing on this, study suggests recommendations on how to work with different types of abnormal distributions of residuals in linear regression.