reduce_model: Reduce a linear model

In contefranz/statutils: statutils: a collection of statistical utilities regarding both modelling, data management and wrangling.

Description

The procedure takes a linear model of class lm (including the output from a stepwise procedure given by step) and reduces it by iteratively deleting predictors based on pvalues of the t-tests (see 'Details').

Usage

reduce_model(model, ...)

## S3 method for class 'lm'
reduce_model(model, data, alpha = 0.05, intercept = TRUE)

Arguments

model	An object of class "lm".
data	A data.frame or data.table with all components of model.
alpha	The significance level at which to cut off predictors. Default is 5%.
intercept	Specify whether the reduction procedure has to consider an intercept or not. Default is TRUE.

Details

The procedure makes use of the workhorse function of lm: lm.fit. This allows to explicitly specify the design matrix which enables to split main effects from interactions. In particular, conversely to what the function lm currently does, it allows the inclusion of just high order effects. For instance, it can fit just interaction terms given by x1*x2 or x1:x2 passed with the usual R formula notation.

Value

reduce_model returns an object of class "lm" with the same components (see lm for a complete list).

Author(s)

Francesco Grossetti [e-mail].

See Also

lm.fit lm step

Examples

data_factor = mtcars
data_factor$am = as.factor( data_factor$am )
levels( data_factor$am ) = c( "no", "yes" )

complete_model = lm( hp ~ qsec + cyl + mpg + disp + drat + wt + qsec + vs +
                     am + gear + carb + am:qsec, data = data_factor )
summary( complete_model )

reduced_model = reduce_model( complete_model, data = data_factors )
summary( reduced_model )

contefranz/statutils documentation built on May 13, 2019, 10:50 p.m.