controlSel {nonprobsvy}

Control parameters for selection model

Description

controlSel constructs a list with all necessary control parameters for selection model.

Usage

controlSel(
  method = "glm.fit",
  epsilon = 1e-04,
  maxit = 500,
  trace = FALSE,
  optimizer = c("maxLik", "optim"),
  maxLik_method = "NR",
  optim_method = "BFGS",
  dependence = FALSE,
  key = NULL,
  est_method_sel = c("mle", "gee"),
  h = c(1, 2),
  penalty = c("SCAD", "lasso", "MCP"),
  a_SCAD = 3.7,
  a_MCP = 3,
  lambda = -1,
  lambda_min = 0.001,
  nlambda = 50,
  nfolds = 10,
  print_level = 0,
  start_type = c("glm", "naive", "zero")
)

Arguments

method

estimation method.

epsilon

Tolerance for fitting algorithms by default 1e-6.

maxit

Maximum number of iterations.

trace

logical value. If TRUE trace steps of the fitting algorithms. Default is FALSE

optimizer

• optimization function for maximum likelihood estimation.

maxLik_method

maximisation method that will be passed to maxLik::maxLik() function. Default is NR.

optim_method

maximisation method that will be passed to stats::optim() function. Default is BFGS.

dependence

logical value - TRUE if samples are dependent.

key

binary key variable

est_method_sel

Method of estimation for propensity score model.

h

Smooth function for the generalized estimating equations methods taking the following values

• if 1 then \(\mathbf{h}\left(\mathbf{x}, \boldsymbol{\theta}\right) = \frac{\pi(\mathbf{x}, \boldsymbol{\theta})}{\mathbf{x}}\)

• if 2 then \( \mathbf{h}\left(\mathbf{x}, \boldsymbol{\theta}\right) = \mathbf{x}\)

penalty

The penanlization function used during variables selection.

a_SCAD

The tuning parameter of the SCAD penalty for selection model. Default is 3.7.

a_MCP

The tuning parameter of the MCP penalty for selection model. Default is 3.

lambda

A user-specified \(\lambda\) value during variable selection model fitting.

lambda_min

The smallest value for lambda, as a fraction of lambda.max. Default is .001.

nlambda

The number of lambda values. Default is 50.

nfolds

The number of folds for cross validation. Default is 10.

print_level

this argument determines the level of printing which is done during the optimization (for propensity score model) process.

start_type

• Type of method for start points for model fitting taking the following values

  • if glm then start taken from the glm function called on samples.

  • if naive then start consists of a vector which has the value of an estimated parameter for one-dimensional data (on intercept) and 0 for the rest.

  • if zero then start is a vector of zeros.

Value

List with selected parameters.

Author(s)

Łukasz Chrostowski, Maciej Beręsewicz

See Also

nonprob() – for fitting procedure with non-probability samples.