In this article, we introduce a new Stata command, smultiv, that implements the S-estimator of multivariate location and scatter. Using simulated data, we show that smultiv outperforms mcd, an alternative robust estimator. Finally, we use smultiv to perform robust principal component analysis and least-squares regression on a real dataset.
CampbellN. A.1984. Mixture models and atypical values. Mathematical Geology16: 465–477.
2.
CampbellN. A., LopuhaäH. P., and RousseeuwP. J.1998. On the calculation of a robust S-estimator of a covariance matrix. Statistics in Medicine17: 2685–2695.
3.
RousseeuwP. J., and LeroyA. M.1987. Robust Regression and Outlier Detection.New York: Wiley.
4.
RousseeuwP. J., and van DriessenK.1999. A fast algorithm for the minimum covariance determinant estimator. Technometrics41: 212–223.
5.
VerardiV., and CrouxC.2009. Robust regression in Stata. Stata Journal9: 439–453.
6.
VerardiV., and CrouxC.2010. Software update: st0173_1: Robust regression in Stata. Stata Journal10: 313.
7.
VerardiV., and DehonC.2010. Multivariate outlier detection in Stata. Stata Journal10: 259–266.
8.
WilksS. S.1932. Certain generalizations in the analysis of variance. Biometrika24: 471–494.
9.
WilsonE. B., and HilfertyM. M.1931. The distribution of chi-square. Proceedings of the National Academy of Sciences of the United States of America17: 684–688.
