Skip to contents

ateCAL: reweight experimental results to match external sample

ateCAL.Rd

Generalization or transportation estimator with method of moments weighting. Imputes counterfactual outcome for each observation i under each treament a as $$Y^a = \omega \frac{A = a}{\pi^a (X)} (Y - \mu^a(X)) + \mu^a(X) $$ Where \(\omega\) is solved for using entropy balancing for balance with target moments.

Usage

ateCAL(
  y,
  a,
  X,
  targetMoments,
  treatProb = NULL,
  nuismod = c("rlm", "rf"),
  estimator = c("ACW", "CW"),
  glmnet_lamchoice = "lambda.min",
  glmnet_alpha = 1,
  separateMus = TRUE,
  noi = FALSE
)

Arguments

y

outcome vector [length n]

a

treatment vector [length n]

X

covariate matrix [n X k]

targetMoments

k-vector of target moments (means in overall sample for generalization, means in target sample for transportation

treatProb

treatment probability, non-null results in no pscore fit

nuismod

one of c("rlm", "rf") : choose how to fit nuisance functions (cross-fit) for outcome models

estimator

one of c("CW", "ACW"). Calibration weights (CW), which fit a set of balancing weights that reweights the sample to match target sample moments. ACW augments this with an outcome model (default)

glmnet_lamchoice

choice of lambda (shrinkage parameter) for regularized linear regressions. Only relevant when nuismod == "rlm"

glmnet_alpha

in [0, 1], choice of alpha in glmnet. 1 (default) corresponds with L1 regularization (LASSO) and 0 corresponds with L2 regularization (ridge), while intermediate values correspond with a mix of the two (elastic net)

separateMus

boolean for whether to fit separate outcome models for each treatment group or a single pooled model. The former is recommended and is the default, but a pooled model may be fit when data is scarce / computation is burdensome.

noi

boolean for printing marginal means and causal contrasts table (it gets returned anyway). On by default.

Value

list containing treatment effects table , nuisance function estimates, and influence function values

Examples

df = selDGP(n = 10000)
id = with(df, s == 1)
X = df$X[id == 1, ]; y = df$y[id == 1]; a = df$a[id == 1]
Xtar = colMeans(df$X[!id, ])
ateCAL(y, a, X, targetMoments = Xtar, estimator = "ACW") %>% summary()
#>      parameter       est         se     ci.ll    ci.ul pval
#> 1      E{Y(0)} 1.3290194 0.11525310 1.1031233 1.554915    0
#> 2      E{Y(1)} 1.9960402 0.08746187 1.8246150 2.167466    0
#> 3 E{Y(1)-Y(0)} 0.6670209 0.13026080 0.4117097 0.922332    0