grpRobGLM: Grouped Generalized Linear Models

In abnormally-distributed/cvreg: Cross Validation and Robust Estimation Utilities

Description

This is similar to the grpOLS function, but extended to the case of generalized linear models. This supports the Gaussian, Binomial, Poisson, and Gamma likelihoods. Note that the covariates in this method must be numeric, and not grouped dummy variable representing a factor. The algorithm implemented here differs from the grpOLS function, so the results for the Gaussian likelihood may slightly differ when compared to the grpOLS results.

Usage

grpRobGLM(
  X,
  Y,
  idx,
  family = gaussian(),
  ranks = NULL,
  tol = 1e-08,
  maxiter = 500
)

Arguments

X	a model matrix (must be numeric, not categorical)
Y	the outcome variable
idx	group id labels
family	one of "gaussian", "Gamma", "binomial", "poisson", "quasibinomial", "quasipoisson", "inverse.gaussian", or "negative.binomial". The family may also provided as an unquoted evaluation of a family function, ie, 'binomial(link="probit")'.
ranks	an indicator for each group whether the covariates of said group are active.
tol	convergence tolerance for IRWLS. Deaults to 1e-8.
maxiter	the maximum number of iterations. defaults to 500.

Value

an sdr object

References

Liu, Y., Chiaromonte, F. and Li, B. (2017) Structured Ordinary Least Squares: A Sufficient Dimension Reduction approach for regressions with partitioned predictors and heterogeneous units. Biom, 73: 529-539. doi:10.1111/biom.12579

abnormally-distributed/cvreg documentation built on May 3, 2020, 3:45 p.m.