robmed (version 1.2.1)

cov_Huber: Huber M-estimator of location and scatter

Description

Compute a Huber M-estimator of location and scatter, which is reasonably robust for a small number of variables.

Usage

cov_Huber(x, control = cov_control(...), ...)

Value

An object of class "cov_Huber" with the following components:

center

a numeric vector containing the location vector estimate.

cov

a numeric matrix containing the scatter matrix estimate.

prob

numeric; probability for the quantile of the \(\chi^{2}\) distribution used as cutoff point in the Huber weight function.

weights

a numeric vector containing the relative robustness weights for the observations.

tau

numeric; correction for Fisher consistency under multivariate normal distributions.

converged

a logical indicating whether the iterative reweighting algorithm converged.

iterations

an integer giving the number of iterations required to obtain the solution.

Arguments

x

a numeric matrix or data frame.

control

a list of tuning parameters as generated by cov_control().

...

additional arguments can be used to specify tuning parameters directly instead of via control.

Author

Andreas Alfons

Details

An iterative reweighting algorithm is used to compute the Huber M-estimator. The Huber weight function thereby corresponds to a convex optimization problem, resulting in a unique solution.

References

Huber, P.J. (1981) Robust Statistics. John Wiley & Sons.

Zu, J. and Yuan, K.-H. (2010) Local Influence and Robust Procedures for Mediation Analysis. Multivariate Behavioral Research, 45(1), 1--44. doi:10.1080/00273170903504695.

See Also

cov_control(), test_mediation(), fit_mediation()

Examples

data("BSG2014")

# define variables
x <- "ValueDiversity"
y <- "TeamCommitment"
m <- "TaskConflict"

# compute Huber M-estimator
cov_Huber(BSG2014[, c(x, y, m)])