Outlier detection methods in multiple linear
regression are reviewed. Eight statistics for outlier detection
have been investigated and compared. It is found from Monte
Carlo simulation that Mahalanobis distance (MDi ) identifiers
the presence of outliers more often than the others for small,
medium and large sample sizes with different percentages
outliers in the regressors and in both the regressors and the
dependent variable. The next best statistics for the detection
are Hat matrix (hii ) ,Cook’s square distance (CDi ) and DEFFITi
distance . As for the dependent variable outlier, Cook’s square
distance (CDi ) and PRESS residual (r(i) ) perform better than
the others.