profileReg: Fit Bayesian profile regression

In lvhoskovec/mmpack: Methods for multipollutant mixtures analysis

Description

Fits the Bayesian profile regression model using MCMC methods

Usage

profileReg(
  niter,
  nburn,
  X,
  Y,
  W,
  C = 20,
  scaleY = FALSE,
  DPgamma = TRUE,
  varsel = FALSE,
  priors,
  sup = TRUE
)

Arguments

niter	number of total iterations
nburn	number of burn-in iterations
X	n by p matrix of predictor data
Y	n by 1 vector of continuous response data
W	n by q matrix of covariate data
C	maximum number of clusters allowed
scaleY	logical; if TRUE response will be centered and scaled before model fit
DPgamma	logical; if TRUE (default) alpha has a gamma prior, else alpha has Unif(.03, 10) prior
varsel	logical; if TRUE binary cluster variable selection is implemented (see Chung and Dunson 2009), default is FALSE
priors	list of prior hyperparameters, see package documentation for details
sup	logical; if TRUE (default) fits supervised model, else fits unsupervised model

Value

list with components

• X: predictor data matrix

• W: covariate data matrix

• Y: response data vector

• C: maximum number of allowable clusters

• alpha: DP parameter estimates

• mu: array of cluster means estimates

• psi: cluster weights estimates

• Z: cluster indicators at each iteraction

• delta: regression coefficient estimates for theta (risk) and gamma (fixed effects)

• sig2inv: error precision estimates

• kap2inv: cluster intercept (risk) precision estimates

• phi2inv: fixed effect precision estimates

• rho: rho estimates

lvhoskovec/mmpack documentation built on Aug. 21, 2021, 4:05 p.m.