bfa_mixed: Initialize and fit a mixed-scale Gaussian factor model, with...

In bfa: Bayesian Factor Analysis

Description

This function performs a specified number of MCMC iterations and returns an object containing summary statistics from the MCMC samples as well as the actual samples if keep.scores or keep.loadings are TRUE. Default behavior is to save only the loadings.

Usage

  bfa_mixed(x, data = NULL, num.factor = 1, restrict = NA,
  normal.dist = NA, center.data = TRUE, scale.data = TRUE,
  nsim = 0, nburn = 0, thin = 1, print.status = 500,
  keep.scores = FALSE, keep.loadings = TRUE, loading.prior
  = c("gdp", "pointmass", "normal"), factor.scales = FALSE,
  px = TRUE, coda = "loadings", coda.scale = TRUE, imh.iter
  = 500, imh.burn = 500, ...)

Arguments

x	A formula, matrix or bfa object.
data	The data if x is a formula
num.factor	Number of factors
restrict	A matrix or list giving restrictions on factor loadings. A matrix should be the same size as the loadings matrix. Acceptable values are 0 (identically 0), 1 (unrestricted), or 2 (strictly positive). List elements should be character vectors of the form c('variable',1, ">0") where 'variable' is the manifest variable, 1 is the factor, and ">0" is the restriction. Acceptable restrictions are ">0" or "0".
normal.dist	A character vector specifying which variables should be treated as observed Gaussian. Defaults to all numeric variables if x is a formula.
center.data	Center data
scale.data	Scale data
nsim	Number of iterations past burn-in
nburn	Number of initial (burn-in) iterations to discard
thin	Keep every thin'th MCMC sample (i.e. save nsim/thin samples)
print.status	How often to print status messages to console
keep.scores	Save samples of factor scores
keep.loadings	Save samples of factor loadings
loading.prior	Specify point mass ("pointmass", default) or normal priors ("normal")
factor.scales	Include a separate scale parameter for each factor
px	Use parameter expansion for ordinal variables (recommended)
coda	Create mcmc objects to allow use of functions from the coda package: "all" for loadings and scores, "loadings" or "scores" for one or the other, or "none" for neither
coda.scale	Put the loadings on the correlation scale when creating mcmc objects
imh.iter	Iterations used to build IMH proposal
imh.burn	Burn-in before collecting samples used to build IMH proposal (total burn-in is nburn+imh.iter+imh.burn)
...	Prior parameters and other (experimental) arguments (see details)

Details

Additional parameters:

• loadings.var: Factor loading prior variance

• tau.a, tau.b: Gamma hyperparameters (scale=1/b) for factor precisions (if factor.scales=T)

• rho.a, rho.b: Beta hyperparameters for point mass prior

• sigma2.a, sigma2.b: Gamma hyperparameters for error precisions (for numeric variables)

• gdp.alpha, gdp.beta: GDP prior parameters

Value

A bfa object with posterior samples.

bfa documentation built on May 2, 2019, 4:55 p.m.