Abstract:
This research is aimed to compare the estimation methods for the dependent variables in multiple linear regression when the errors are from lognormal and gamma distribution with different levels of skewness 0.5,1,1.5 and 2 and their variance depend on the value of their first of independent and dependent variables under power relationship with the parameter 0,0.1,0.3 and 0.5. The estimation methods under comparisons are the Ordinary Least Square (OLS), Box-Cox transformation, and Weighted Least Square (WLS) methods. By comparing the Average Mean Square Errors (AMSE) and the Relative Efficiency (RE), we have found that the OLS method performs best when the variances of errors are from gamma distribution weakly depending on the values of their first of independent and dependent variables while the WLS method performs best when the variances of errors are strongly depending on the values of their first of independent and the Box-Cox transformation method performs best when the variances of errors are strongly depending on the values of their dependent variables . Furthermore when the variances of errors are from lognormal distribution the WLS method performs best when the variances of errors are weakly depending on the values of their first of independent and dependent variables while the Box-Cox transformation method performs best when the variances of errors are strongly depending on the values of their first of independent and dependent variables .