varComprob.control: Tuning Parameters for 'varComprob()' and Auxiliaries

In robustvarComp: Robust Estimation of Variance Component Models

Tuning Parameters for varComprob() and Auxiliaries

Description

Tuning Parameters for varComprob() which performs S, composite S and Tau estimators for variance component models.

Usage

varComprob.control(init = NULL, lower = 0, upper = Inf, epsilon = 0.001,
  tuning.chi = NULL, bb = 0.5, tuning.psi = NULL,
  arp.chi = 0.1, arp.psi = NULL, max.it = 100,
  rel.tol.beta = 1e-06, rel.tol.gamma = 1e-05, rel.tol.scale = 1e-05,
  trace.lev = 0,
  method = c("compositeTau", "compositeS", "compositeMM",
    "Tau", "S", "MM"),
  psi = c("optimal", "bisquare", "rocke"),
  beta.univ = FALSE, gamma.univ = FALSE,
  fixed.init = c("lmrob.S", "lmRob"),
  cov.init = c("TSGS", "2SGS", "covOGK"),
  cov = TRUE, ...)

Arguments

init	A list with initial values. The only components used are beta, gamma, eta0, scales (in case of composite methods), scale (in case of classic methods) and Sigma. If one of the components is empty a suitable automatic initial values is calculated.
lower	A numeric vector with length equals to gamma. This parameter is passed to the optim function.
upper	A numeric vector with length equals to gamma. This parameter is passed to the optim function.
epsilon	A positive numeric scalar. This value is set to the non negative elements of gamma (checked by lower) when the automatic initial values are negative.
tuning.chi	tuning constant vector for the rho_1 function. If NULL, as by default, this is set, depending on "psi" (for now only "bisquare" and "optimal"), to a suitable value.
bb	expected value under the normal model of the rho function with tuning constant equal to tuning.chi.
tuning.psi	tuning constant vector for the rho_2 function. If NULL, as by default, this is set, depending on "psi" (for now only "bisquare" and "optimal"), to a suitable value.
arp.chi	tuning constant vector for the rho_1 function in case psi is set to "rocke".
arp.psi	tuning constant vector for the rho_2 function in case psi is set to "rocke".
max.it	integer specifying the maximum number of IRWLS iterations.
rel.tol.beta	(for the RWLS iterations algorithm of the fixed parameters): relative convergence tolerance for the parameter vector.
rel.tol.gamma	(for the optim function used in the estimation procedure of the random variance parameters): relative convergence tolerance for the parameter vector.
rel.tol.scale	relative convergence tolerance for the scale vector.
trace.lev	integer indicating if the progress of the algorithm should be traced (increasingly); default 'trace.lev = 0' does no tracing.
method	string specifying the estimator-chain. For now available procedures are 'compositeS', 'compositeTau' and 'S'. Default is set to 'compositeTau'.
psi	string specifying the type psi-function used. Available choices for the composite methods are "bisquare" and "optimal". For classic methods "rocke", "bisquare" and "optimal". Defaut is set to "optimal".
beta.univ	logical. If TRUE a robust simple regression is performed for each explanatory variable in order to get starting values for the fixed effect parameters.
gamma.univ	logical. If TRUE a simple regression is performed for each explanatory variable in order to get starting values for the random variance parameters.
fixed.init	string with function name to be used to calculate initial value of the fixed effect parameters. Possible values are "lmrob.S" and "lmRob".
cov.init	function or string with function name to be used to calculate initial covariance matrix estimate if necessary. Possible string value is "TSGS", "2SGS" and "covOGK". Default is set to "TSGS".
cov	logical. If TRUE the estimated variance-covariance matrix for the fixed and random parameters is reported.
...	further arguments.

Value

returns a named 'list' with over twenty components, corresponding to the arguments.

Author(s)

Claudio Agostinelli and Victor J. Yohai

Examples

   ## Show the default settings:
     str(varComprob.control())

robustvarComp documentation built on Dec. 28, 2022, 2:36 a.m.