Help for package AnchorRegression

Type:

Package

Title:

Perform AnchorRegression

Version:

0.1.3

Author:

Simon Zimmermann

Maintainer:

Simon Zimmermann <zimmersi@hu-berlin.de>

Description:

Performs AnchorRegression proposed by Rothenhäusler et al. 2020. The code is adapted from the original paper repository. (https://github.com/rothenhaeusler/anchor-regression) The code was developed independently from the authors of the paper.

License:

MIT + file LICENSE

URL:

https://github.com/simzim96/AnchorRegression

BugReports:

https://github.com/simzim96/AnchorRegression/issues

Depends:

R (≥ 2.0.0)

Imports:

glmnet (≥ 1.4), selectiveInference (≥ 1.0.0), mgcv (≥ 1.0)

Encoding:

UTF-8

RoxygenNote:

7.1.1

Suggests:

knitr, rmarkdown, testthat

NeedsCompilation:

Packaged:

2021-01-05 17:37:17 UTC; simon

Repository:

CRAN

Date/Publication:

2021-01-06 14:40:02 UTC

anchor_prediction

Description

Perform a prediction for an Anchor Regression model as described in Rothenhäusler et al.2020

Usage

anchor_prediction(anchor_model, x, anchor, gamma, target_variable)

Arguments

anchor_model

is the Anchor Regression model object

x

is a dataframe containing the matrix x containing the independent variables

anchor

is a dataframe containing the matrix anchor containing the anchor variable

gamma

is the regularization parameter for the Anchor Regression

target_variable

is the target variable name contained in the x dataframe

Value

A list of predictions.

Examples

x <- as.data.frame(matrix(data = rnorm(100),nrow = 100,ncol = 10))
anchor <- as.data.frame(matrix(data = rnorm(200),nrow = 100,ncol = 2))
colnames(anchor) <- c('X1','X2')
gamma <- 2
target_variable <- 'V2'
anchor_model <- anchor_regression(x, anchor, gamma, target_variable)
anchor_prediction(anchor_model$model, x, anchor, gamma, target_variable)

anchor_prediction

Description

Perform a prediction for an Anchor Regression model as described in Rothenhäusler et al.2020

Usage

anchor_prediction_gam(
  anchor_model,
  x,
  anchor,
  gamma,
  target_variable,
  bin_factor
)

Arguments

anchor_model

is the Anchor Regression model object

x

is a dataframe containing the matrix x containing the independent variables

anchor

is a dataframe containing the matrix anchor containing the anchor variable

gamma

is the regularization parameter for the Anchor Regression

target_variable

is the target variable name contained in the x dataframe

bin_factor

binary variable that can be transformed to a factor to partial out effects

Value

A list of predictions.

Examples

x <- as.data.frame(matrix(data = rnorm(10000),nrow = 1000,ncol = 10))
x$bin <- sample(nrow(x),x = c(1,0),prob = c(0.5,0.5),replace = TRUE)
anchor <- as.data.frame(matrix(data = rnorm(2000),nrow = 1000,ncol = 2))
colnames(anchor) <- c('X1','X2')
gamma <- 2
target_variable <- 'V2'

anchor_model <- anchor_regression_gam(x, anchor, gamma, target_variable,"bin")
anchor_prediction_gam(anchor_model$model, x, anchor, gamma, target_variable,"bin")

anchor_regression

Description

Perform an Anchor Regression as described in Rothenhäusler et al.2020

Usage

anchor_regression(x, anchor, gamma, target_variable, lambda = "CV")

Arguments

x

is a dataframe containing the matrix x containing the independent variables

anchor

is a dataframe containing the matrix anchor containing the anchor variable

gamma

is the regularization parameter for the Anchor Regression

target_variable

is the target variable name contained in the x dataframe

lambda

indicates the lambda that is used in the Anchor Regression. 'CV' is used if it should be estimated by cross validation on the full subset.

Value

A list with coefficient values and a list with the respective names overview_print. Additionally the transformed data as x and y plus the fixed lambda coefficient.

Examples

x <- as.data.frame(matrix(data = rnorm(1000),nrow = 100,ncol = 10))
anchor <- as.data.frame(matrix(data = rnorm(200),nrow = 100,ncol = 2))
colnames(anchor) <- c('X1','X2')
gamma <- 2
target_variable <- 'V2'
anchor_regression(x, anchor, gamma, target_variable)

anchor_regression_gam

Description

Perform an Generalized Additive Anchor Regression

Usage

anchor_regression_gam(x, anchor, gamma, target_variable, bin_factor = NULL)

Arguments

x

is a dataframe containing the matrix x containing the independent variables

anchor

is a dataframe containing the matrix anchor containing the anchor variable

gamma

is the regularization parameter for the Anchor Regression

target_variable

is the target variable name contained in the x dataframe

bin_factor

binary variable that can be transformed to a factor to partial out effects

Value

A list with coefficient values and a list with the respective names overview_print. Additionally the transformed data as x and y plus the fixed lambda coefficient.

Examples

x <- as.data.frame(matrix(data = rnorm(10000),nrow = 1000,ncol = 10))
x$bin <- sample(nrow(x),x = c(1,0),prob = c(0.5,0.5),replace = TRUE)
anchor <- as.data.frame(matrix(data = rnorm(2000),nrow = 1000,ncol = 2))
colnames(anchor) <- c('X1','X2')
gamma <- 2
target_variable <- 'V2'
anchor_regression_gam(x, anchor, gamma, target_variable,bin_factor =  "bin")

anchor_stability

Description

Perform an Anchor Stability Analysis as described in Rothenhäusler et al.2020

Usage

anchor_stability(
  x,
  anchor,
  target_variable,
  lambda = 0,
  alpha = 0.05,
  p_procedure = "naive"
)

Arguments

x

is a dataframe containing the matrix x containing the independent variables

anchor

is a dataframe containing the matrix anchor containing the anchor variable

target_variable

is the target variable name contained in the x dataframe

lambda

indicates the lambda that is used in the Anchor Regression. 'CV' is used if it should be estimated by cross validation on the full subset.

alpha

significance level for test decision on coefficient significance

p_procedure

procedure to estimate stability. Option 1: naive - stable if effect is non-zero in all cases; Option 2: post-lasso - post selection inference using SelectiveInference package

Value

A dataframe containing the stability values for each coefficient

Examples

x <- as.data.frame(matrix(data = rnorm(1000),nrow = 100,ncol = 10))
anchor <- as.data.frame(matrix(data = rnorm(200),nrow = 100,ncol = 2))
colnames(anchor) <- c('X1','X2')
gamma <- 2
target_variable <- 'V2'
anchor_stability(x, anchor, target_variable, lambda, alpha=0.05, p_procedure = "naive")

weighted_anchor_regression

Description

Perform a prediction for a Weighted Anchor Regression model

Usage

weighted_anchor_prediction(names, coeff, x, anchor, gamma, target_variable)

Arguments

names

list of variable names corresponding to the coefficients in coeff

coeff

list of coefficients corresponding to the coefficients in names

x

is a dataframe containing the matrix x containing the independent variables

anchor

is a dataframe containing the matrix anchor containing the anchor variable

gamma

is the regularization parameter for the Anchor Regression

target_variable

is the target variable name contained in the x dataframe

Value

A list of predictions.

Examples

   # number of observed environments
   environments <- 10

   # populate list with generated data of x and anchor
   data_x_list <- c()
   data_anchor_list <- c()
   for(e in 1:environments){
     x <- as.data.frame(matrix(data = rnorm(100),nrow = 100,ncol = 10))
     anchor <- as.data.frame(matrix(data = rnorm(200),nrow = 100,ncol = 2))
     colnames(anchor) <- c('X1','X2')
     data_x_list[[e]] <- x
     data_anchor_list[[e]]  <- anchor
   }

   # estimate model
   gamma <- 2
   target_variable <- 'V2'
   weighted_anchor_model <- weighted_anchor_regression(data_x_list,
                                                       data_anchor_list,
                                                       gamma,
                                                       target_variable,
                                                       anchor_model_pre=NULL,
                                                       test_split=0.4,
                                                       lambda=0)
   weighted_anchor_prediction(weighted_anchor_model$names,
                              weighted_anchor_model$coeff,
                              x,
                              anchor,
                              gamma,
                              target_variable)

weighted_anchor_regression

Description

Estimates weighted Anchor Regression coefficients

Usage

weighted_anchor_regression(
  data_x_list,
  data_anchor_list,
  gamma,
  target_variable,
  anchor_model_pre = NULL,
  test_split = 0.4,
  lambda = 0
)

Arguments

data_x_list

list containing coefficient dataframes for different environments

data_anchor_list

list containing anchor dataframes for different environments

gamma

is the regularization parameter for the Anchor Regression

target_variable

is the target variable name contained in the x dataframe

anchor_model_pre

is the pre estimated model for the Anchor Regression. In case of NULL a new model is estimated.

test_split

is desired test/train split for the estimation

lambda

penalization coefficient for Anchor Shrinkage. Initially set to 0.

Value

A list estimated coefficients with names, weights and the raw coefficient matrix

Examples

   environments <- 10 # number of observed environments

   # populate list with generated data of x and anchor
   data_x_list <- c()
   data_anchor_list <- c()
   for(e in 1:environments){
     x <- as.data.frame(matrix(data = rnorm(100),nrow = 100,ncol = 10))
     anchor <- as.data.frame(matrix(data = rnorm(200),nrow = 100,ncol = 2))
     colnames(anchor) <- c('X1','X2')
     data_x_list[[e]] <- x
     data_anchor_list[[e]]  <- anchor
   }

   # estimate model
   gamma <- 2
   target_variable <- 'V2'
   weighted_anchor_regression(data_x_list,
                              data_anchor_list,
                              gamma,
                              target_variable,
                              anchor_model_pre=NULL,
                              test_split=0.4,
                              lambda=0)