robglm: Robust Generalized Linear Models
In abnormally-distributed/cvreg: Cross Validation and Robust Estimation Utilities

Robust Generalized Linear Models

robglm(
  formula,
  data,
  family = "binomial",
  method = c("mallows", "none", "schweppe", "welsch", "carroll"),
  k = 1.345,
  cw = 8
)

`formula`	A model formula
`data`	A data frame
`family`	One of "binomial", "poisson", "Gamma", "gaussian".
`method`	The method to be used to calculate weights for the design matrix. One of the options below. The default is "mallows". - "none" uses equal weights for all rows of the design matrix - "mallows" uses the diagonal of the hat matrix to calculate leverage weights as sqrt(1-h) - "carroll" uses the sqrt(mah.dist/ p) as a statistic to feed into the psi function ψ = u ((1-(u/cw)^2)^3 * I(u) <= cw)* to generate weights as ψ(u)/u. - "schweppe" multiplies the mallows weights by psi-weights derived from robustly standardized deviance residuals from an initial fit, i.e., sqrt(1-h)*psi((r-median(r))/mad(r)). Schweppe's weights prevent high-leverage observations that are not associated with large residuals (good leverage points) from being downweighted. This results in increased efficiency. - "welsch" is similar to Mallow's weights, but Welsch's weights are calculated as (1-h)/sqrt(h).
`k`	Tuning constant for Huber's psi. defaults to 1.345.
`cw`	Tuning constant for caroll weights. defaults to 8.