step_AIC: Stepwise model construction and inspection

In MASSExtra: Some 'MASS' Enhancements

Stepwise model construction and inspection

Description

Front-ends to stepAIC and dropterm with changed defaults. step_BIC implements a stepwise selection with BIC as the criterion and step_GIC uses an experimental criterion with a penalty midway between AIC and BIC: the "Goldilocks" criterion.

Usage

step_AIC(object, ..., trace = 0, k = 2)

step_BIC(object, ..., trace = 0, k = max(2, log(nobs(object))))

step_GIC(object, ..., trace = 0, k = (2 + log(nobs(object)))/2)

drop_term(
  object,
  ...,
  test = default_test(object),
  k,
  sorted = TRUE,
  decreasing = TRUE,
  delta = TRUE
)

add_term(
  object,
  ...,
  test = default_test(object),
  k,
  sorted = TRUE,
  decreasing = TRUE,
  delta = TRUE
)

Arguments

object	as for stepAIC
...	additional arguments passed on to main function in MASS
trace, k	as for stepAIC
sorted, test	as for dropterm and addterm
decreasing	in drop_term should the rows be displayed in decreasing order, that is best to worst terms, from that of dropterm?
delta	Should the criterion be displayed (FALSE) or the change in the in the criterion relative to the present model (TRUE)?

Value

A fitted model object after stepwise refinement, or a data frame with extra class membership for single term functions.

Examples

fm <- glm.nb(Days ~ .^3, quine)
drop_term(fm_aic <- step_AIC(fm))
drop_term(fm_bic <- step_BIC(fm))

MASSExtra documentation built on Feb. 16, 2023, 10:55 p.m.