A Comparison of Two Procedures, the Mahalanobis Distance and the Andrews-Pregibon Statistic, for Identifying Multivariate Outliers.

Autor/in	Jarrell, Michele Glankler
Titel	A Comparison of Two Procedures, the Mahalanobis Distance and the Andrews-Pregibon Statistic, for Identifying Multivariate Outliers.
Quelle	(1992), (27 Seiten) PDF als Volltext kostenfreie Datei Verfügbarkeit
Sprache	englisch
Dokumenttyp	gedruckt; online; Monographie
Schlagwörter	Comparative Analysis; Computer Simulation; Error of Measurement; Mathematical Models; Multivariate Analysis; Research Problems + Suchen Sie Ihr Suchwort? Computergrafik; Computersimulation; Messfehler; Mathematical model; Mathematisches Modell; Multivariate Analyse; Forschungskritik
Abstract	This repeated measures factorial design study compared the results of two procedures for identifying multivariate outliers under varying conditions, the Mahalanobis distance and the Andrews-Pregibon statistic. Results were analyzed for the total number of outliers identified and number of false outliers identified. Simulated data were limited to three dimensions. Using known population parameters, 1,100 samples of size 150 were computer-generated. Outliers were induced by replacing randomly selected data points in each sample with plus or minus the value of three or six standard deviations (SDs) from the mean. The samples had outliers induced into one dimension, then into two dimensions, and finally, into three dimensions. Each sample was manipulated in terms of the number of dimensions and the extremeness of the outliers. Six outlier samples were produced from each original sample. Analyses of variance were run using the number of false outliers and then the number of total outliers as the dependent variable. Tukey's honestly significant difference was calculated for each of the main effects. Each procedure accounted for less than one percent of the total variability in the model. The degree of outlier extremeness accounted for most of the variability in each case, with the three SD group accounting for more outliers than the six SD group. Other than calculation ease, there is no reason to choose one procedure over the other. Included are 8 tables and 39 references. (RLC)
Erfasst von	ERIC (Education Resources Information Center), Washington, DC