Title: Estimating ATE with Misclassified Outcomes and Mismeasured Covariates
Version: 1.0.0
Description: Addressing measurement error in covariates and misclassification in binary outcome variables within causal inference, the 'ATE.ERROR' package implements inverse probability weighted estimation methods proposed by Shu and Yi (2017, <doi:10.1177/0962280217743777>; 2019, <doi:10.1002/sim.8073>). These methods correct errors to accurately estimate average treatment effects (ATE). The package includes two main functions: ATE.ERROR.Y() for handling misclassification in the outcome variable and ATE.ERROR.XY() for correcting both outcome misclassification and covariate measurement error. It employs logistic regression for treatment assignment and uses bootstrap sampling to calculate standard errors and confidence intervals, with simulated datasets provided for practical demonstration.
License: GPL (≥ 3)
Encoding: UTF-8
RoxygenNote: 7.2.3
Depends: R (≥ 2.10)
LazyData: true
Suggests: knitr, rmarkdown
VignetteBuilder: knitr
NeedsCompilation: no
Packaged: 2024-09-05 02:37:25 UTC; rezanejad
Imports: ggplot2, MASS, mvtnorm, rlang, stats
Author: Aryan Rezanezhad [aut, cre], Grace Y. Yi [aut]
Maintainer: Aryan Rezanezhad <Aryan.rzn@gmail.com>
Repository: CRAN
Date/Publication: 2024-09-10 09:10:10 UTC

ATE.ERROR: Estimating ATE with Misclassified Outcomes and Mismeasured Covariates

Description

Addressing measurement error in covariates and misclassification in binary outcome variables within causal inference, the 'ATE.ERROR' package implements inverse probability weighted estimation methods proposed by Shu and Yi (2017, doi:10.1177/0962280217743777; 2019, doi:10.1002/sim.8073). These methods correct errors to accurately estimate average treatment effects (ATE). The package includes two main functions: ATE.ERROR.Y() for handling misclassification in the outcome variable and ATE.ERROR.XY() for correcting both outcome misclassification and covariate measurement error. It employs logistic regression for treatment assignment and uses bootstrap sampling to calculate standard errors and confidence intervals, with simulated datasets provided for practical demonstration.

Addressing measurement error in covariates and misclassification in binary outcome variables within causal inference, the 'ATE.ERROR' package implements inverse probability weighted estimation methods proposed by Shu and Yi (2017, doi:10.1177/0962280217743777; 2019, doi:10.1002/sim.8073). These methods correct errors to accurately estimate average treatment effects (ATE). The package includes two main functions: ATE.ERROR.Y() for handling misclassification in the outcome variable and ATE.ERROR.XY() for correcting both outcome misclassification and covariate measurement error. It employs logistic regression for treatment assignment and uses bootstrap sampling to calculate standard errors and confidence intervals, with simulated datasets provided for practical demonstration.

Author(s)

Maintainer: Aryan Rezanezhad Aryan.rzn@gmail.com

Authors:

ATE.ERROR.XY Function for Estimating Average Treatment Effect (ATE) with Measurement Error in X and Misclassification in Y

Description

The ATE.ERROR.XY function implements a method for estimating the Average Treatment Effect (ATE) that accounts for both measurement error in covariates and misclassification in the binary outcome variable Y.

Usage

ATE.ERROR.XY(
  Y_star,
  A,
  Z,
  X_star,
  p11,
  p10,
  sigma_epsilon,
  B = 100,
  Lambda = seq(0, 2, by = 0.5),
  extrapolation = "linear",
  bootstrap_number = 250
)

Arguments

Y_star

Numeric vector. The observed binary outcome variable, possibly misclassified.

A

Numeric vector. The treatment indicator (1 if treated, 0 if control).

Z

Numeric vector. A precisely measured covariate vector.

X_star

Numeric vector. A covariate vector subject to measurement error.

p11

Numeric. The probability of correctly classified Y given Y = 1.

p10

Numeric. The probability of misclassified Y given Y = 0.

sigma_epsilon

Numeric. The covariance matrix Sigma_epsilon for the measurement error model.

B

Integer. The number of simulated datasets.

Lambda

Numeric vector. A sequence of lambda values for simulated datasets.

extrapolation

Character. A regression model used for extrapolation ("linear", "quadratic", "nonlinear").

bootstrap_number

Numeric. The number of bootstrap samples (default is 250).

Details

The ATE.ERROR.XY function is designed to handle measurement error in covariates and misclassification in outcomes by using the augmented simulation-extrapolation approach.

Value

A list containing:

summary

A data frame with the following columns:

  • Naive_ATE: Naive estimate of the ATE.

  • Sigma_epsilon: The covariance matrix Sigma_epsilon for the measurement error model.

  • p10: The probability of misclassified Y given Y = 0.

  • p11: The probability of correctly classified Y given Y = 1.

  • Extrapolation: A regression model used for extrapolation ("linear", "quadratic", "nonlinear").

  • ATE: Mean ATE estimate from the bootstrap samples.

  • SE: Standard error of the ATE estimate.

  • CI: 95% confidence interval for the ATE estimate.

boxplot

A ggplot object representing the boxplot of the ATE estimates.

Examples


library(ATE.ERROR)
data(Simulated_data)
Y_star <- Simulated_data$Y_star
A <- Simulated_data$T
Z <- Simulated_data$Z
X_star <- Simulated_data$X_star
p11 <- 0.8
p10 <- 0.2
sigma_epsilon <- 0.1
B <- 100
Lambda <- seq(0, 2, by = 0.5)
bootstrap_number <- 10
result <- ATE.ERROR.XY(Y_star, A, Z, X_star, p11, p10, sigma_epsilon, B, Lambda, 
                       "linear", bootstrap_number)
print(result$summary)
print(result$boxplot)

ATE.ERROR.Y Function for Estimating Average Treatment Effect (ATE) with Misclassification in Y

Description

This function performs estimation of the Average Treatment Effect (ATE) using the ATE.ERROR.Y method, which accounts for misclassification in the binary outcome variable Y. The method calculates consistent estimates of the ATE in the presence of misclassified outcomes by leveraging logistic regression and bootstrap sampling.

Usage

ATE.ERROR.Y(Y_star, A, Z, X, p11, p10, bootstrap_number = 250)

Arguments

Y_star

Numeric vector. The observed binary outcome variable, which may be subject to misclassification.

A

Numeric vector. The binary treatment indicator (1 if treated, 0 if control).

Z

Numeric vector. A precisely measured covariate vector.

X

Numeric vector. A precisely measured covariate vector.

p11

Numeric. The probability of correctly classified Y given Y = 1.

p10

Numeric. The probability of misclassified Y given Y = 0.

bootstrap_number

Integer. The number of bootstrap samples (default is 250) used to obtain the associated variance estimate.

Details

The function first calculates consistent estimates of the ATE, correcting for misclassification in the outcome variable Y. The logistic model is used to estimate the propensity scores for the treatment assignment, which are then adjusted using the provided misclassification probabilities p11 and p10. Bootstrap sampling is performed to estimate the variance and construct confidence intervals for the ATE estimates.

Value

A list containing:

summary

A data frame with the following columns:

  • Naive_ATE: Naive estimate of the ATE, ignoring misclassification.

  • ATE: Mean ATE estimate from the bootstrap samples, accounting for misclassification.

  • SE: Standard error of the ATE estimate.

  • CI: 95% confidence interval for the ATE estimate.

boxplot

A ggplot object representing the boxplot of the ATE estimates.

Examples

library(ATE.ERROR)
data(Simulated_data)
Y_star <- Simulated_data$Y_star
A <- Simulated_data$T
Z <- Simulated_data$Z
X <- Simulated_data$X
p11 <- 0.8
p10 <- 0.2
bootstrap_number <- 250
result <- ATE.ERROR.Y(Y_star, A, Z, X, p11, p10, bootstrap_number)
print(result$summary)
print(result$boxplot)

Naive Estimation of Average Treatment Effect

Description

This function performs a naive estimation of the ATE. This approach gives us the so-called "naive estimate" by ignoring the difference between (X*, Y*) and (X, Y).

Usage

Naive_Estimation(Y_star, A, Z, X_star)

Arguments

Y_star

A numeric vector of outcomes with potential misclassification.

A

A numeric vector of treatment assignments.

Z

A numeric vector of covariate Z.

X_star

A numeric vector of covariate X with measurement error.

Value

A numeric value representing the estimated treatment effect.

Examples

library(ATE.ERROR)
data(Simulated_data)
Y_star <- Simulated_data$Y_star
A <- Simulated_data$T
Z <- Simulated_data$Z
X_star <- Simulated_data$X_star
Naive_ATE_XY <- Naive_Estimation(Y_star, A, Z, X_star)
print(Naive_ATE_XY)

Simulated Data

Description

A dataset containing simulated data generated by the generate_data function. This data includes misclassified outcome Y*, treatment assignment T, and covariates X and Z.

Usage

Simulated_data

Format

A data frame with 5000 rows and 6 variables:

X

a numeric vector generated from a standard normal distribution (mean = 0, standard deviation = 1)

X_star

a numeric vector where X_star is equal to X plus a random error. The random error is generated from a normal distribution with mean 0 and standard deviation 0.1

Y

a numeric vector generated from a Bernoulli distribution with a probability depending on T, Z, and X

Y_star

a numeric vector where Y_star is generated from a binomial distribution depending on Y with probabilities 0.8 if Y equals 1 and 0.2 if Y equals 0

T

a numeric vector generated from a binomial distribution with probability calculated using the logistic function of the sum of 0.2, Z, and X

Z

a numeric vector generated from a standard normal distribution (mean = 0, standard deviation = 1)

Source

Shu D, Yi GY (2019). Weighted causal inference methods with mismeasured covariates and misclassified outcomes. Statistics in Medicine. 38:1835-1854. doi:10.1002/sim.8073

True Estimation of Average Treatment Effect

Description

This function performs a true estimation of the Average Treatment Effect (ATE) using the generated values for X and Y. The consistent estimator is calculated as the difference between the expected value of the outcome for the treated group and the expected value of the outcome for the control group.

Usage

True_Estimation(Y, A, Z, X)

Arguments

Y

A numeric vector of outcomes.

A

A numeric vector of treatment assignments.

Z

A numeric vector of covariate Z.

X

A numeric vector of covariate X.

Details

The expected value for the treated group, E(Y_1), is calculated as the mean of the product of the treatment assignment and the outcome divided by the estimated propensity score.

The expected value for the control group, E(Y_0), is calculated as the mean of the product of the control assignment and the outcome divided by one minus the estimated propensity score.

The propensity score is estimated by applying a logistic regression model to the true values of the covariates and treatment assignments.

Value

A numeric value representing the estimated treatment effect.

Examples

library(ATE.ERROR)
data(Simulated_data)
Y <- Simulated_data$Y
A <- Simulated_data$T
Z <- Simulated_data$Z
X <- Simulated_data$X
True_ATE <- True_Estimation(Y, A, Z, X)
print(True_ATE)