| Abstract: | A combination of R-mode factor residual, factor score, and univariate screening methods is more successful than univariate screening (based on cumulative frequency curves) alone in identifying outliers and sharpening definition of background populations. The factor models for the screened data sets are more easy to interpret than those for the original data sets, and are used to check the background nature of the screened populations. The resulting data sets can be used either for statistical evaluation of single-element data, or as training sets for multivariate classification. The screening method is illustrated with stream-sediment data from South Greenland. |
