dfbetas: Assess influence of observations on parameter estimates for...

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

Description

DFBETAs are a measure of stability in the parameter estimates for a model. They aim to assess whether there are an observations with strong leverage over the parameter estimate. The DFBETAs for the ith observation are the difference between the full-data parameter estimates and the parameter estimates without the ith row of observations:

\hat{β}_j - \tilde{β}_{(-i, j)}

. DFBETAs for an observaton-variable pair greater than +/- 2/sqrt(n) are considered to indicate that the ith observation causes potential instability in the estimate for the jth variable.

Usage

dfBetas(
  formula,
  data,
  family = c("gaussian", "poisson", "binomial", "quasibinomial", "quasipoisson",
    "inverse.gaussian", "Gamma"),
  plot = TRUE,
  which = "plotall",
  nrows = NULL,
  ncols = NULL
)

Arguments

formula	a model formula
data	a data frame
family	a character string identifying the GLM family to use. The default is "gaussian".
plot	whether to plot or not. defaults to TRUE. if FALSE a data frame of each dfbeta is returned.
which	which variable to plot. defaults to "plotall", which creates a facetplot with ggplot2. otherwise, a specific variable's name can be entered, such as "Sepal.Length", and a single plot will be created with base graphics.
nrows	alter the number of rows for the facetplot if you wish.
ncols	alter the number of columns for the facetplot if you wish.

Value

a plot or a data frame.

References

Belsley, Kuh, and Welsch (1980) Regression Diagnostics. pages 13-14.

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