ER function

Robust EM-algorithm ER

The ER function is an implementation of the ER-algorithm of Little and Smith (1987).

ER(
  data,
  weights,
  alpha = 0.01,
  psi.par = c(2, 1.25),
  em.steps = 100,
  steps.output = FALSE,
  Estep.output = FALSE,
  tolerance = 1e-06
)

Arguments

• data: a data frame or matrix with the data.
• weights: sampling weights.
• alpha: probability for the quantile of the cut-off.
• psi.par: further parameters passed to the psi-function.
• em.steps: number of iteration steps of the EM-algorithm.
• steps.output: if TRUE, verbose output.
• Estep.output: if TRUE, estimators are output at each iteration.
• tolerance: convergence criterion (relative change).

Returns

• sample.size: Number of observations
• number.of.variables: Number of variables
• significance.level: alpha
• computation.time: Elapsed computation time
• good.data: Indices of the data in the final good subset
• outliers: Indices of the outliers
• center: Final estimate of the center
• scatter: Final estimate of the covariance matrix
• dist: Final Mahalanobis distances
• rob.weights: Robustness weights in the final EM step

Details

The M-step of the EM-algorithm uses a one-step M-estimator.

Examples

data(bushfirem, bushfire.weights)
det.res <- ER(bushfirem, weights = bushfire.weights, alpha = 0.05,
steps.output = TRUE, em.steps = 100, tol = 2e-6)
PlotMD(det.res$dist, ncol(bushfirem))

References

Little, R. and P. Smith (1987). Editing and imputation for quantitative survey data. Journal of the American Statistical Association, 82, 58-68.

See Also

BEM

Author(s)

Beat Hulliger