ATE: Estimate the Average Treatment Effect
In asadharis/ATE: Inference for Average Treatment Effects using Covariate Balancing

Description Usage Arguments Value Author(s) References See Also Examples

View source: R/main_function.R

The main function for estimating the average treatment effect or the average treatment effect on the treated. This function creates an ATE object which can be used as inputs for generic S3 plot or summary functions. This function uses a covariate balancing method which creates weights for each subject, without a need to specific a propensity score or an outcome regression model. The main function implements the BFGS algorithm with backtracking. For details of the estimator see Chan et al. (2015).

ATE (Y, treat, X, theta = 1, ATT = FALSE,
    verbose = FALSE, max.iter = 100, tol = 1e-10,
    initial.values = NULL,
    backtrack.alpha = 0.3,
    backtrack.beta = 0.5)

`Y`	The response vector of length n. This has to be a numeric vector.
`treat`	The vector of treatment assignments of length n. Must be coded as {0, 1, ...} and must be binary vector when `ATT = TRUE`.
`X`	A N x p-matrix of covariates `X` to be matched. This matrix does not need to include an intercept.
`theta`	A real scalar parameter for the Cressie-Read family of objective functions. The default is θ = 0 (exponential tilting). Other popular examples are θ = -1 (empirical likelihood) and θ = 1 (quadratic loss). See `CRFamily` for details.
`ATT`	A logical value indicating whether to estimate the treatment effect on the treated.
`verbose`	A logical value indicating whether to print the progress of the function. Default value is `FALSE`.
`max.iter`	The maximum number of iterations for the BFGS algorithm.
`tol`	The absolute tolerance used to determine a stopping criteria for the BFGS algorithm.
`initial.values`	A numeric vector or matrix of possible initial values for the BFGS algorithm. Must be a J x K matrix where J is the number of treatment arms. For `ATT = TRUE` this must be a K-vector.
`backtrack.alpha`	A scalar parameter for the backtracking line search algorithm with α in (0,0.5).
`backtrack.beta`	A scalar parameter for for the backtracking line search algorithm with β in (0,1).

The function reruns an object of type 'ATE', a list with the following elements

`estimate`	The vector of point estimates for the average treatment effect. For a binary treatment it also contains the average difference of treatment effects.
`vcov`	The estimated variance covariance matrix for the estimates of the treatment effects for each treatment group.
`lambda`	The resulting solution of the main optimization problems, hat{λ}, as described in Chan et al.(2015). In the case of a simple, binary treatment study, the object has `lambda.treat` and `lambda.placebo` and when `ATT = TRUE`, we only have `lambda.placebo`. For a multiple treatment study design we have `lam.mat`, a matrix with each row for each hat{λ}, corresponding to each treatment arm.
`weights`	The weights obtained by the balancing covariate method for each treatment group. In the case of `ATT = TRUE`, we only have weights for the untreated. For binary treatment the list will have `weights.treat` and `weights.placebo`. For multiple treatment effect the list contains a J\times N matrix `weights.mat`.
`gp`	A string specifying the type of study design. For binary treatment effect with `ATT = FALSE` is denoted by group "simple". With `ATT = TRUE` we have "ATT" and finally "MT" is for multiple treatment arms.
`converge`	A logical value indicating convergence of BFGS algorithm.
`X, Y, treat`	The data used for estimation.
`J`	A scalar indicating the number of treatment arms.
`K`	A scalar indicating the one plus the dimension of the range space of X. K = 1 + p.
`call`	The matched call.

Asad Haris, Gary Chan.

Chan, K. C. G. and Yam, S. C. P. and Zhang, Z. (2015). "Globally Efficient Nonparametric Inference of Average Treatment Effects by Empirical Balancing Calibration Weighting", under revision.

Haris, Asad and Chan, K. C. G. (2015). "ATE: An R package for Nonparametric Inference of Average Treatment Effects", under revision

summary.ATE, plot.ATE

library(ATE)
# Binary treatment.
set.seed(25)
n <- 200
Z <- matrix(rnorm(4 * n), ncol = 4, nrow = n)
prop <- 1 / (1 + exp(Z[, 1] - 0.5 * Z[, 2] + 0.25 * Z[, 3] + 0.1 * Z[, 4]))
treat <- rbinom(n, 1, prop)
Y <- 200 + 10 * treat + (1.5 * treat - 0.5) *
     (27.4 * Z[, 1] + 13.7 * Z[, 2] + 13.7 * Z[, 3] + 13.7 * Z[, 4]) + rnorm(n)
X <- cbind(exp(Z[, 1]) / 2, Z[, 2] / (1 + exp(Z[, 1])),
           (Z[, 1] * Z[, 3] / 25 + 0.6) ^ 3, (Z[, 2] + Z[, 4] + 20) ^ 2)

# Estimation of average treatment effects (ATE).
fit1 <- ATE(Y, treat, X)
summary(fit1)
# plot(fit1)

# Estimation of average treatment effects on treated (ATT).
fit2 <- ATE(Y, treat, X, ATT = TRUE)
summary(fit2)
# plot(fit2)


# Four treatment groups.
set.seed(25)
n <- 200
Z <- matrix(rnorm(4 * n), ncol = 4, nrow = n)
prop1 <- 1 / (1 + exp(1 + Z[, 1] - 0.5 * Z[, 2] + 0.25 * Z[, 3] + 0.1 * Z[, 4]))
prop2 <- 1 / (1 + exp(Z[, 1] - 0.5 * Z[, 2] + 0.25 * Z[, 3] + 0.1 * Z[, 4]))

U <- runif(n)
treat <- numeric(n)
treat[U > (1 - prop2)] = 2
treat[U < (1 - prop2) & U > (prop2 - prop1)] = 1

Y <- 210 + 10 * treat + (27.4 * Z[, 1] + 13.7 * Z[, 2] + 13.7 * Z[, 3] +
                        13.7 * Z[, 4]) + rnorm(n)
X <- cbind(exp(Z[, 1]) / 2, Z[, 2] / (1 + exp(Z[, 1])),
           (Z[, 1] * Z[, 3] / 25 + 0.6) ^ 3, (Z[, 2] + Z[, 4] + 20) ^ 2)

fit3 <- ATE(Y, treat, X)
summary(fit3)
# plot(fit3)