Package 'selcorr'

Post-Selection Inference for Generalized Linear Models

Description

selcorr returns (unconditional) post-selection confidence intervals and p-values for the coefficients of (generalized) linear models.

Usage

selcorr(
  object,
  fixed.vars = NULL,
  further.vars = NULL,
  boot.repl = 0,
  k = 2,
  conf.level = 0.95,
  quiet = FALSE
)

Arguments

object	an object representing a model of an appropriate class. This is used as the initial model in a (bidirectional) stepwise model selection.
fixed.vars	the names of all independent variables that must be included in the selected model. The default is none.
further.vars	the names of all independent variables that can be included in the selected model, but are not part of object. The default is none.
boot.repl	a number or list of bootstrap replicates. The default is no bootstrapping. See Details and Examples for clarification.
k	the multiple of the number of degrees of freedom used as penalty in the model selection. The default k = 2 corresponds to the AIC.
conf.level	the level of the confidence intervals.
quiet	if TRUE, then selcorr does not generate an output.

Details

When boot.repl = 0, an approximate asymptotic distribution of the test statistic is used to calculate p-values and calibrate the profile-likelihood confidence intervals. This approach is faster, but p-values and confidence intervals can be more precisely calibrated by parametrically bootstrapping the test statistic (with boot.repl the number of replicates). Parallel computing can be used to speed up the bootstrapping: see Examples.

Value

the selected model is returned, without correction for model-selection, but with up to two additional components. There is an output component corresponding to the post-selection inference, which is also printed unless quiet = TRUE. When boot.repl is not 0, there is also a boot.repl component corresponding to the bootstrap replicates.

Examples

## linear regression:
selcorr(lm(Fertility ~ ., swiss))

## logistic regression:
swiss.lr = within(swiss, Fertility <- (Fertility > 70))
selcorr(glm(Fertility ~ ., binomial, swiss.lr))

## parallel bootstrapping:
## Not run: 
library(future.apply)
plan(multisession)
boot.repl = future_replicate(8, selcorr(lm(Fertility ~ ., swiss), boot.repl = 1000,
                                        quiet = TRUE)$boot.repl, simplify = FALSE)
plan(sequential)
selcorr(lm(Fertility ~ ., swiss), boot.repl = do.call("rbind", boot.repl))
## End(Not run)

---