Skip to contents

Update and re-fit probit model

Source: R/model_fitting.R
update.RprobitB_fit.Rd

This function estimates a nested probit model based on a given RprobitB_fit object.

Usage

# S3 method for RprobitB_fit
update(
  object,
  form,
  re,
  alternatives,
  id,
  idc,
  standardize,
  impute,
  scale,
  R,
  B,
  Q,
  print_progress,
  prior,
  latent_classes,
  seed,
  ...
)

Arguments

object

An object of class RprobitB_fit.

form

A formula object that is used to specify the model equation. The structure is choice ~ A | B | C, where

  • choice is the name of the dependent variable (the choices),

  • A are names of alternative and choice situation specific covariates with a coefficient that is constant across alternatives,

  • B are names of choice situation specific covariates with alternative specific coefficients,

  • and C are names of alternative and choice situation specific covariates with alternative specific coefficients.

Multiple covariates (of one type) are separated by a + sign. By default, alternative specific constants (ASCs) are added to the model. They can be removed by adding +0 in the second spot.

In the ordered probit model (ordered = TRUE), the formula object has the simple structure choice ~ A. ASCs are not estimated.

re

A character (vector) of covariates of form with random effects. If re = NULL (the default), there are no random effects. To have random effects for the ASCs, include "ASC" in re.

alternatives

A character vector with the names of the choice alternatives. If not specified, the choice set is defined by the observed choices. If ordered = TRUE, alternatives is assumed to be specified with the alternatives ordered from worst to best.

id

A character, the name of the column in choice_data that contains unique identifier for each decision maker. The default is "id".

idc

A character, the name of the column in choice_data that contains unique identifier for each choice situation of each decision maker. Per default, these identifier are generated by the order of appearance.

standardize

A character vector of names of covariates that get standardized. Covariates of type 1 or 3 have to be addressed by <covariate>_<alternative>. If standardize = "all", all covariates get standardized.

impute

A character that specifies how to handle missing covariate entries in choice_data, one of:

  • "complete_cases", removes all rows containing missing covariate entries (the default),

  • "zero", replaces missing covariate entries by zero (only for numeric columns),

  • "mean", imputes missing covariate entries by the mean (only for numeric columns).

scale

A character which determines the utility scale. It is of the form <parameter> := <value>, where <parameter> is either the name of a fixed effect or Sigma_<j>,<j> for the <j>th diagonal element of Sigma, and <value> is the value of the fixed parameter.

R

The number of iterations of the Gibbs sampler.

B

The length of the burn-in period, i.e. a non-negative number of samples to be discarded.

Q

The thinning factor for the Gibbs samples, i.e. only every Qth sample is kept.

print_progress

A boolean, determining whether to print the Gibbs sampler progress and the estimated remaining computation time.

prior

A named list of parameters for the prior distributions. See the documentation of check_prior for details about which parameters can be specified.

latent_classes

Either NULL (for no latent classes) or a list of parameters specifying the number of latent classes and their updating scheme:

  • C: The fixed number (greater or equal 1) of latent classes, which is set to 1 per default. If either weight_update = TRUE or dp_update = TRUE (i.e. if classes are updated), C equals the initial number of latent classes.

  • weight_update: A boolean, set to TRUE to weight-based update the latent classes. See ... for details.

  • dp_update: A boolean, set to TRUE to update the latent classes based on a Dirichlet process. See ... for details.

  • Cmax: The maximum number of latent classes.

  • buffer: The number of iterations to wait before a next weight-based update of the latent classes.

  • epsmin: The threshold weight (between 0 and 1) for removing a latent class in the weight-based updating scheme.

  • epsmax: The threshold weight (between 0 and 1) for splitting a latent class in the weight-based updating scheme.

  • distmin: The (non-negative) threshold in class mean difference for joining two latent classes in the weight-based updating scheme.

seed

Set a seed for the Gibbs sampling.

...

Ignored.

Value

An object of class RprobitB_fit.

Details

All parameters (except for object) are optional and if not specified retrieved from the specification for object.