Repository Mirror for your Cloud Server and Webhosting

Type:

Package

Title:

Variance-Guided Regression Improving Upon OLS and ANOVA

Version:

0.1.5

Date:

2026-06-18

Author:

Sibei Liu [aut], Min Lu [aut, cre]

Maintainer:

Min Lu <luminwin@gmail.com>

License:

GPL-2 | GPL-3 [expanded from: GPL (≥ 2)]

Depends:

R (≥ 3.5.0)

Imports:

glmnet, lmtest, sandwich

Description:

Fits variance-guided linear regression models that provide an alternative to ordinary least squares (OLS) for general linear-model design matrices, including ANOVA-style encodings. The methods use an iteratively reweighted least squares estimator or an iteratively reweighted lasso estimator and implement the global linear mean-variance model from the associated 2026 Statistics in Medicine article <doi:10.1002/sim.70632>. Under the assumptions in that paper, the estimator matches the homoscedastic baseline in population predictive quasi-risk when variance is constant and improves on it when the variance depends on covariates. The grouping-based nonlinear prediction extension from Section 3 is available in the development version on GitHub.

Encoding:

UTF-8

LazyData:

true

URL:

https://github.com/luminwin/varGuid

BugReports:

https://github.com/luminwin/varGuid/issues

NeedsCompilation:

Packaged:

2026-06-18 12:00:00 UTC; minlu

Repository:

CRAN

Date/Publication:

2026-06-18 22:40:02 UTC

Data from the cobra2 Simulation

Description

Data with non-linear and interaction relationship with 500 observations and 15 variables.

Usage

data(cobra2d)

Author(s)

Sibei Liu and Min Lu

References

Liu, S. and Lu, M. (2026). Variance-Guided Regression for Heteroscedastic Data with a Grouping-Based Extension for Nonlinear Prediction. Statistics in Medicine. 45(13-14):e70632. doi:10.1002/sim.70632

Examples


# library(copula)
# cobra2 = function(n = 500, d = 15, sd = .1, corrv = 0) {
#   set.seed(1)
#   d <- max(10, d)
#   X <- matrix(runif(n * d, -1, 1), ncol = d)
#   paramlist <- lapply(1:d, function(j) {list(min=-1,max=1)})
#   myCop <- normalCopula(param=rep(corrv,dim(combn(d,2))[2]), dim = d, dispstr = "un")
#   myMvd <-  mvdc(copula=myCop, margins=rep("unif",d),paramMargins=paramlist)
#   X[, 1:d] <- rMvdc(n, myMvd)
#   dta <- data.frame(list(x = X, y = X[,1]*X[,2] + X[,3]^2 - X[,4]*X[,7] + X[,8]*X[,10] - X[,6]^2
#                          + rnorm(n, sd = sd)))
#   colnames(dta)[1:d] <- paste("x", 1:d, sep = "")
#   f <- "x1 * x2 + x3 ^ 2 - x4 * x7 + x8 * x10 - x6 ^ 2"
#   fs <- "I(x1 * x2) + I(x3 ^ 2) + I(-x4 * x7) + I(x8 * x10) - I(x6 ^ 2)"
#   list(f = f, fs = fs, dta = dta)
# }

data(cobra2d)

Fit Variance-Guided Linear Mean-Variance Models

Description

Fits the stage-1 variance-guided linear model for heteroscedastic data using iteratively reweighted least squares (IRLS) when lasso = FALSE or an iteratively reweighted lasso procedure when lasso = TRUE. For lasso = FALSE, the returned object also includes weighted least squares and heteroscedasticity-consistent inference summaries based on the final fit.

Usage

lmv(X, Y, M = 10, step = 1, tol = exp(-10), lasso = FALSE)

Arguments

X

Input matrix with observations in rows and predictors in columns.

Y

Numeric response vector.

M

Maximum number of iterations.

step

Scale parameter for the data weights.

tol

Tolerance parameter for convergence.

lasso

Logical; if TRUE, uses the iteratively reweighted lasso algorithm. Otherwise, IRLS is used.

Value

A list with the following components:

beta

Coefficient estimates from the final variance-guided fit.

obj.OLS

Unweighted baseline lm fit used to initialize IRLS when lasso = FALSE.

obj.lasso

Unweighted baseline glmnet fit used to initialize the iteratively reweighted lasso when lasso = TRUE.

obj.varGuid

Final fitted model from either lm or glmnet, depending on lasso.

res

Object returned by the variance-model update in the last iteration.

obj.varGuid.coef

For lasso = FALSE, a list of weighted least squares and heteroscedasticity-consistent coefficient summaries computed from the final fit.

X

The input design matrix X.

Author(s)

Sibei Liu and Min Lu

References

Examples


data(cobra2d, package = "varGuid")
dat <- cobra2d
set.seed(1)
tid <- sample(seq_len(nrow(dat)), 200)
train <- dat[-tid, ]
yid <- which(colnames(dat) == "y")

o <- lmv(X = train[, -yid], Y = train[, yid], lasso = FALSE)
summary(o$obj.varGuid)
summary(o$obj.OLS)
head(prd(o, train[, -yid], model = "baseline"))
head(prd(o, train[, -yid], model = "varGuid"))

o2 <- lmv(X = train[, -yid], Y = train[, yid], lasso = TRUE)
o2$beta
o2$obj.lasso$beta
head(prd(o2, train[, -yid], model = "baseline"))
head(prd(o2, train[, -yid], model = "varGuid"))

Predict from Fitted Stage-1 varGuid Models

Description

A lightweight prediction helper for objects returned by lmv(). For ordinary weighted least squares fits it dispatches to stats::predict(). For iteratively reweighted lasso fits it dispatches to glmnet::predict.glmnet() and returns a numeric vector.

Usage

prd(object, newdata, model = c("varGuid", "baseline"), ...)

Arguments

object

An object returned by lmv().

newdata

A matrix or data frame of predictors for prediction.

model

Which fitted model to use. "varGuid" uses the final variance-guided fit. "baseline" uses the initial OLS fit when lasso = FALSE or the initial lasso fit when lasso = TRUE.

...

Additional arguments passed to stats::predict() or glmnet::predict.glmnet().

Details

This CRAN release covers the global linear mean-variance model from Section 2 of Liu and Lu (2026). For the grouping-based nonlinear prediction extension from Section 3 of the paper, please use the development version available at devtools::install_github("luminwin/varGuid").

Value

A numeric vector of predictions.

Author(s)

Sibei Liu and Min Lu

References

Examples


data(cobra2d, package = "varGuid")
dat <- cobra2d
set.seed(1)
tid <- sample(seq_len(nrow(dat)), 200)
train <- dat[-tid, ]
yid <- which(colnames(dat) == "y")

o <- lmv(X = train[, -yid], Y = train[, yid], lasso = FALSE)
head(prd(o, train[, -yid], model = "baseline"))
head(prd(o, train[, -yid], model = "varGuid"))

o2 <- lmv(X = train[, -yid], Y = train[, yid], lasso = TRUE)
head(prd(o2, train[, -yid], model = "baseline"))
head(prd(o2, train[, -yid], model = "varGuid"))

Variance-Guided Regression Improving Upon OLS and ANOVA

Description

The varGuid package implements the global linear mean-variance model from Liu and Lu (2026) using iteratively reweighted least squares and iteratively reweighted lasso estimation for general linear-model design matrices, including ANOVA-style encodings. This CRAN release focuses on the Section 2 linear mean-variance model and its prediction utilities for fitted stage-1 models. For the grouping-based nonlinear prediction extension from Section 3 of the paper, please use the development version available at devtools::install_github("luminwin/varGuid").

Author(s)

Sibei Liu and Min Lu

References

Examples


data(cobra2d, package = "varGuid")
dat <- cobra2d
set.seed(1)
tid <- sample(seq_len(nrow(dat)), 200)
train <- dat[-tid, ]
yid <- which(colnames(dat) == "y")

o <- lmv(X = train[, -yid], Y = train[, yid], lasso = FALSE)
summary(o$obj.varGuid)
summary(o$obj.OLS)
head(prd(o, train[, -yid], model = "baseline"))
head(prd(o, train[, -yid], model = "varGuid"))

# Iteratively reweighted lasso:
o2 <- lmv(X = train[, -yid], Y = train[, yid], lasso = TRUE)
o2$beta
o2$obj.lasso$beta
head(prd(o2, train[, -yid], model = "baseline"))
head(prd(o2, train[, -yid], model = "varGuid"))

Package {varGuid}

Data from the cobra2 Simulation

Description

Usage

Author(s)

References

Examples

Fit Variance-Guided Linear Mean-Variance Models

Description

Usage

Arguments

Value

Author(s)

References

Examples

Predict from Fitted Stage-1 varGuid Models

Description

Usage

Arguments

Details

Value

Author(s)

References

Examples

Variance-Guided Regression Improving Upon OLS and ANOVA

Description

Author(s)

References

Examples