2021, Vol. 6, Issue 4, Part B
Effect of non-normal error distribution on simple Linear/non-parametric regression modelsAuthor(s):
Opara Jude and George IsobeyeAbstract:
This study is on the effect of non-normal error distribution on simple linear regression versus its nonparametric equivalent. The error term for normality proved that it is not from a normal population using Ryan-Joiner, which violates the major assumption of simple linear regression. Hence, estimating its slope becomes immaterial and any inference drawn from the OLS won’t be reliable. Since, there is no need of employing the technique, due to its poor performance in the presence of error non-normality, then a feasible alternative technique which performs consistently and robust to non-normality residual is required. The simulation study conducted in this study suggested that the nonparametric Theil’s simple linear regression is an alternative to OLS when there is existence of non-normal error in a data set. The study recommended among others that further studies on simple linear regression should ensure that the underlying assumptions of OLS are fulfilled before estimation; otherwise its non-parametric equivalent should be employed, but if the researcher must continue with OLS after failure of assumption, then outliers should be checked and if detected, should be removed and re-examine the underlying assumptions.Pages: 131-136 | Views: 246 | Downloads: 16Download Full Article: Click Here
How to cite this article:
Opara Jude, George Isobeye. Effect of non-normal error distribution on simple Linear/non-parametric regression models. Int J Stat Appl Math 2021;6(4):131-136.