Robust Mahalanobis Distance
Robust Mahalanobis Distance (MCD-based Multivariate Outlier Detection) · Also known as: MCD Mahalanobis distance, robust mahalanobis, minimum covariance determinant distance, Robust Mahalanobis Uzaklığı
Robust Mahalanobis Distance flags multivariate outliers by measuring how far each observation lies from the centre of the data using a robust covariance estimate. It builds on the robust-distance framework of Rousseeuw and Van Zomeren (1990) and the multivariate outlier-detection approach of Filzmoser, Garrett and Reimann (2005), replacing the classical mean and covariance with the Minimum Covariance Determinant (MCD) estimate so that the outliers themselves do not distort the distance.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use it when you have multivariate continuous data and want to detect outliers or leverage points without letting those outliers corrupt the reference centre and covariance. It assumes roughly ellipsoidal (elliptically symmetric) continuous data and works best with a reasonable sample size — at least about 50 observations. It is not reliable for very small samples (below about 20 observations, single-variable outlier methods are preferable), and the MCD breaks down when more than about a quarter of the data are outliers.
Strengths & limitations
- Resists the masking effect: the MCD centre and scatter are computed from the clean core, so clustered outliers cannot hide themselves.
- Detects multivariate outliers and leverage points that are invisible to one-variable-at-a-time screening.
- Provides a principled chi-square cut-off for flagging observations rather than an arbitrary threshold.
- Requires a moderate sample size (the method targets at least about 50 observations) and assumes approximately ellipsoidal continuous data.
- Becomes unreliable in very small samples — below roughly 20 observations univariate outlier detection is preferable.
- The MCD-based distance breaks down when the outlier fraction is too high (above about 25%).
Frequently asked
How is this different from the ordinary Mahalanobis distance?
The ordinary distance uses the sample mean and covariance, which are themselves pulled by outliers, so clustered outliers can mask one another. The robust version replaces them with the MCD centre and scatter estimated from the clean core of the data, so genuine outliers stand out instead of hiding.
What is the MCD estimator?
The Minimum Covariance Determinant estimator finds the most concentrated subset of the data — the half whose covariance matrix has the smallest determinant — and uses its centre and scatter as a high-breakdown estimate of location and shape that resists contamination.
How are observations flagged as outliers?
An observation is flagged when its squared robust distance exceeds a chi-square cut-off, commonly the 0.975 quantile of the chi-square distribution with degrees of freedom equal to the number of variables.
How many outliers can it tolerate?
The MCD-based distance is reliable while contamination stays modest. When more than about 25% of the observations are outliers the estimate breaks down, and an alternative such as an adjusted boxplot should be used.
Sources
- Rousseeuw, P. J. & Van Zomeren, B. C. (1990). Unmasking Multivariate Outliers and Leverage Points. Journal of the American Statistical Association, 85(411), 633-639. DOI: 10.1080/01621459.1990.10474920 ↗
- Filzmoser, P., Garrett, R. G. & Reimann, C. (2005). Multivariate Outlier Detection in Exploration Geochemistry. Computational Statistics & Data Analysis, 49(2), 561-587. DOI: 10.1016/j.cageo.2004.11.013 ↗
How to cite this page
ScholarGate. (2026, June 1). Robust Mahalanobis Distance (MCD-based Multivariate Outlier Detection). ScholarGate. https://scholargate.app/en/statistics/mahalanobis-robust
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Adjusted BoxplotStatistics↔ compare
- Least Trimmed SquaresStatistics↔ compare
- MAD EstimationStatistics↔ compare
- Robust ANOVAStatistics↔ compare
- Theil-Sen EstimatorStatistics↔ compare