Author(s): Esemokumo Perewarebo Akpos and Victor-Edema Uyodhu A

Abstract: The study is on effect of outliers in simple linear regression data. Data used for the study are birth weight and mother’s age with 41 observations. A standard residual approach for detecting outliers is employed in detecting outliers in the regression data. The birth weight and mother’s age data showed that the error term is not from a normal distribution using Anderson-Darling statistic, with a coefficient of determination of 52.4%, but after detecting an outlier and removing it; the data set was re-analyzed and the error term is proven to be from a normal distribution with a coefficient of determination of 58.3%. Hence, the goodness of fit measures showed that the data set by removing the influential point (MSE = 0.074), (AIC = 0.306), (SIC = 0.390), (HQC = 0.336) proved highly efficient than the one without removing the influential point (MSE = 0.103), (AIC = 0.612), (SIC = 0.696), (HQC = 0.643) since their MSE, AIC, SIC and HQC are significantly smaller as indicated. The study concludes that the bivariate linear regression can only be employed in the data used for this study when outliers are adequately removed from the data since the normality assumption is not achieved when the influential point is present.

Pages: 68-74 | Views: 452 | Downloads: 25

Download Full Article: Click Here