Outliers and Influence Points NCSS metrics and descriptions
Diagnostics for Outliers and High Influence Points Outliers from the model Residuals – yi – yihat measures distance from the data to the model Standardized residual - residual divided by its standard deviation, assuring variance of the observed residuals is constant 2or 3 a priori Rstudent – standardized residual with sj (root MSE calculated without observation j, also denoted MSEj) rather than s (root MSE) in the denominator. Outliers in the data (high leverage) Hat diagonal – see text pages 206 – 208 4/N High influence points Cook’s D – attempts to measure the influence of each observation on all N fitted values, i.e. all estimated parameters .5 or 1 or 4/(N-2)
Diagnostics for Outliers and High Influence Points Dffits – attempts to measure the influence of an observation on its individual prediction 1 CovRatio – flags influential observations on the generalized variance of the regression coefficients 1 – 3p/N DFBETAS - measures the influence of an observation on the estimated BETA coefficient 2/root(N) or 1 or 2