Stepwise R-Squared backward regression
Source: R/ols-stepaic-backward-regression.R
ols_step_backward_r2.RdBuild regression model from a set of candidate predictor variables by removing predictors based on r-squared, in a stepwise manner until there is no variable left to remove any more.
Usage
ols_step_backward_r2(model, ...)
# Default S3 method
ols_step_backward_r2(
model,
include = NULL,
exclude = NULL,
progress = FALSE,
details = FALSE,
steps = NULL,
...
)
# S3 method for class 'ols_step_backward_r2'
plot(x, print_plot = TRUE, details = TRUE, digits = 3, ...)Arguments
- model
An object of class
lm; the model should include all candidate predictor variables.- ...
Other arguments.
- include
Character or numeric vector; variables to be included in selection process.
- exclude
Character or numeric vector; variables to be excluded from selection process.
- progress
Logical; if
TRUE, will display variable selection progress.- details
Logical; if
TRUE, will print the regression result at each step.- steps
Number of steps after which the stepwise procedures should stop.
- x
An object of class
ols_step_backward_*.- print_plot
logical; if
TRUE, prints the plot else returns a plot object.- digits
Number of decimal places to display.
Value
List containing the following components:
- model
final model; an object of class
lm- metrics
selection metrics
- others
list; info used for plotting and printing
Details
Refits accept explicit contrast matrices or the base contrast names
contr.treatment, contr.sum, contr.helmert, contr.poly and contr.SAS
(resolved from the stats namespace); models fitted with custom named contrast
functions must be refitted with an explicit matrix, for example
contrasts = list(g = my_matrix). Refits carry the fitted model's frozen total
offset as an argument with all formula offset() terms removed for
standardized-beta and selection computations. Returned stepwise models are
fitted from the original call, preserving formula and argument offsets for
prediction on new data. Their coefficients, fitted values and total offset
must be identical to the frozen-offset selection fit; otherwise selection
stops with an error because prediction with new data is not supported for
that model.
For non-hierarchical stepwise selection, candidate rankings and criterion improvement tests use 10 significant digits: values that agree to 10 significant digits are ties, equal criterion values mean no improvement, and a remaining tie goes to the first candidate in formula order after any F tie is broken by p value (smaller for entry, larger for removal). P-value entry and removal cutoffs are compared without rounding.
References
Venables, W. N. and Ripley, B. D. (2002) Modern Applied Statistics with S. Fourth edition. Springer.
See also
Other backward selection procedures:
ols_step_backward_adj_r2(),
ols_step_backward_aic(),
ols_step_backward_p(),
ols_step_backward_sbc(),
ols_step_backward_sbic()
Examples
# stepwise backward regression
model <- lm(y ~ ., data = surgical)
ols_step_backward_r2(model)
#> [1] "No variables have been removed from the model."
# final model and selection metrics
k <- ols_step_backward_aic(model)
k$metrics
#> step variable r2 adj_r2 aic sbc sbic
#> 1 1 alc_mod 0.7817703 0.7485615 734.4068 752.3077 584.2757
#> 2 2 gender 0.7814169 0.7535127 732.4942 748.4061 581.9383
#> 3 3 age 0.7809054 0.7580831 730.6204 744.5433 579.6377
k$model
#>
#> Call:
#> (function (formula, data, subset, weights, na.action, method = "qr",
#> model = TRUE, x = FALSE, y = FALSE, qr = TRUE, singular.ok = TRUE,
#> contrasts = NULL, offset, ...)
#> {
#> ret.x <- x
#> ret.y <- y
#> cl <- match.call()
#> mf <- match.call(expand.dots = FALSE)
#> m <- match(c("formula", "data", "subset", "weights", "na.action",
#> "offset"), names(mf), 0L)
#> mf <- mf[c(1L, m)]
#> mf$drop.unused.levels <- TRUE
#> mf[[1L]] <- quote(stats::model.frame)
#> mf <- eval(mf, parent.frame())
#> if (method == "model.frame")
#> return(mf)
#> else if (method != "qr")
#> warning(gettextf("method = '%s' is not supported. Using 'qr'",
#> method), domain = NA)
#> mt <- attr(mf, "terms")
#> y <- model.response(mf, "numeric")
#> w <- as.vector(model.weights(mf))
#> if (!is.null(w) && !is.numeric(w))
#> stop("'weights' must be a numeric vector")
#> offset <- model.offset(mf)
#> mlm <- is.matrix(y)
#> ny <- if (mlm)
#> nrow(y)
#> else length(y)
#> if (!is.null(offset)) {
#> if (!mlm)
#> offset <- as.vector(offset)
#> if (NROW(offset) != ny)
#> stop(gettextf("number of offsets is %d, should equal %d (number of observations)",
#> NROW(offset), ny), domain = NA)
#> }
#> if (is.empty.model(mt)) {
#> x <- NULL
#> z <- list(coefficients = if (mlm) matrix(NA_real_, 0,
#> ncol(y)) else numeric(), residuals = y, fitted.values = 0 *
#> y, weights = w, rank = 0L, df.residual = if (!is.null(w)) sum(w !=
#> 0) else ny)
#> if (!is.null(offset)) {
#> z$fitted.values <- offset
#> z$residuals <- y - offset
#> }
#> }
#> else {
#> x <- model.matrix(mt, mf, contrasts)
#> z <- if (is.null(w))
#> lm.fit(x, y, offset = offset, singular.ok = singular.ok,
#> ...)
#> else lm.wfit(x, y, w, offset = offset, singular.ok = singular.ok,
#> ...)
#> }
#> class(z) <- c(if (mlm) "mlm", "lm")
#> z$na.action <- attr(mf, "na.action")
#> z$offset <- offset
#> z$contrasts <- attr(x, "contrasts")
#> z$xlevels <- .getXlevels(mt, mf)
#> z$call <- cl
#> z$terms <- mt
#> if (model)
#> z$model <- mf
#> if (ret.x)
#> z$x <- x
#> if (ret.y)
#> z$y <- y
#> if (!qr)
#> z$qr <- NULL
#> z
#> })(formula = y ~ bcs + pindex + enzyme_test + liver_test + alc_heavy,
#> data = surgical)
#>
#> Coefficients:
#> (Intercept) bcs pindex enzyme_test liver_test alc_heavy
#> -1178.330 59.864 8.924 9.748 58.064 317.848
#>
# include or exclude variable
# force variables to be included in the selection process
ols_step_backward_r2(model, include = c("alc_mod", "gender"))
#> [1] "No variables have been removed from the model."
# use index of variable instead of name
ols_step_backward_r2(model, include = c(7, 6))
#> [1] "No variables have been removed from the model."
# force variable to be excluded from selection process
ols_step_backward_r2(model, exclude = c("alc_heavy", "bcs"))
#> [1] "No variables have been removed from the model."
# use index of variable instead of name
ols_step_backward_r2(model, exclude = c(8, 1))
#> [1] "No variables have been removed from the model."
# steps
ols_step_backward_r2(model, steps = 2)
#> [1] "No variables have been removed from the model."