help mregress-------------------------------------------------------------------------------

Title

mcd-- Minimum Covariance Determinant estimator of location and scatter

Syntaxmcdvarlist[if] [in] [,options]

optionsDescription ------------------------------------------------------------------------- Modele(#)maximal expected proportion of outliersproba(#)probability of selecting at least one clean sumple in the p-subset algorithmtrim(#)percentage of trimmingoutlierreturn robust Mahalanobis distances and flag outliersbestsamplflag oservations used for estimating the trimmed covariance matrixrawreturn the raw robust covariance matrixsetseed(#)set the seed

Description

mcdfinds the Minimum Covariance Determinant estimator of location and scatter. By default, the one step reweighted MCD robust covariance matrix is saved in matrix covRMCD and the one step reweighted MCD robust location vector is saved in matrix locationRMCD.

Options+-------+ ----+ Model +------------------------------------------------------------

e(#)sets the expected percentage of outliers existing in the dataset. Setting it high, slows down the algorithm. It is set by default to 0.2 but can take any value ranging from 0 to 0.5.

proba(#)sets the probability of having at least one non-corrupt sample among all those considered. It is set by default to 0.99 but can take any value ranging from 0 to 0.9999.

trim(#)sets the trimming. It is set by default to 0.5 but can take any value ranging from 0 to 0.5.

outliercreates a dummy identifying multivariate outliers and returns robust distances.

bestsampleflags the subsample used for estimating the MCD location vector and scatter matrix.

rawreturns the genuine MCD location vector (locationMCD) and covariance matrix (covMCD) rather than the one step reweighted (the default). The reweighted location vector and covariance matrix are computed using classical estimators on the dataset cleaned of identified outliers.

setseed(#)allows the user to set a seed. Setting the seed allows to replicate the results.

ExamplesSetup

. webuse autoEstimate robust Mahalanobis distances. mcd mpg headroom trunk weight length turn displacement gear_ratio,outlierDisplay the robust reweighted covariance matrix and location vector. matrix list covRMCD. matrix list locationRMCDSame as above bust using the raw data. mcd mpg headroom trunk weight length turn displacement gear_ratio,outlier rawDisplay the robust raw covariance matrix and location vector. matrix list covMCD. matrix list locationMCD

ReferencesRousseeuw, P.J. and Van Driessen, K. (1999). "A fast algorithm for the minimum covariance determinant estimator". Technometrics, 41, 212--223.

Also seeOnline:

[R] qreg,[R] regress;[R] rreg, mmregress, sregress, msregress, mregress