Robust Fitting of Linear Models using Huber Psi Function

Description

Using iteratively reweighted least squares (IRLS), the function calculates the optimal weights to perform m-estimator or bounded influence regression. Returns robust beta estimates and prints robust ANOVA table

Usage

1
robustRegH(formula,data,tune=1.345,m=TRUE,max.it=1000,tol=1e-5,anova.table=FALSE)

Arguments

formula

Model

data

A data frame containing the variables in the model.

tune

Tuning Constant. Default value of 1.345 is 95% asymptotically efficient against outliers

m

If TRUE, calculates m estimates of beta. If FALSE, calculates bounded influence estimates of beta

max.it

Maximum number of iterations to achieve convergence in IRLS algorithm

tol

Tolerance level in determining convergence

anova.table

If TRUE, prints robust ANOVA table

Details

M-estimates of beta should be used when evaluating least squares estimates of beta and diagnostics show outliers. Least squares estimates of beta are used as starting points to achieve convergence.

Bounded influence estimates of beta should be used when evaluating least squares estimates of beta and diagnostics show large values of the "Hat Matrix" diagonals and outliers.

Note

Original package written in 2006

Author(s)

Ian M. Johnson ian@alpha-analysis.com

References

P. J. Huber (1981) Robust Statistics. Wiley.

Birch (1983) Robust F-Test

See Also

robustRegBS()

Examples

1
2
3
4
5
data(stackloss)
robustRegH(stack.loss~Air.Flow+Water.Temp,data=stackloss)

#If X matrix contained large values of H matrix (high influence points)
robustRegH(stack.loss~Air.Flow+Water.Temp,data=stackloss,m=FALSE)