STATISTICAL COMPARISON OF DIFFERENT METHODS OF ESTIMATION (THE GENERALIZED LEAST SQUARES, WEIGHTED RIDGE AND, WEIGHTED LEAST SQUARES) IN THE PRESENCE OF HETEROSCEDASTICITY AND NON-NORMAL ERRORS

Abstract

Common problems in multiple regression models are heteroscedasticity and non-normal errors, which produce undesirable effects on the least squares estimators. This study saw reasons to combine different methods of estimation (The Ordinary Least Squares, Weighted Least Squares and Weighted Ridge Regression) to deal with these problems. From a simulation study, the results of comparisons show that for the condition of heteroscedasticity, Weighted Least Squares (WLS) estimates are more efficient than the other estimators considered. This is because its  values in root mean square error (0.4788), residual standard error (2,7519) and residual mean absolute deviation (0.1167) has the best linear unbiased estimates. For the condition of heteroscedasticity and non-normal errors, Weighted Least Squares produces estimates that were more efficient and precise.