Entropy balancing for the construction of balanced samples in observational studies. Reweights a control group so that user-specified covariate moments match those of a treatment group, or reweights a survey sample to match known population characteristics. The weights can then be passed to regression or other downstream models to estimate treatment effects on the reweighted data.

The method is described in:

Hainmueller, J. (2012). "Entropy Balancing for Causal Effects: A Multivariate Reweighting Method to Produce Balanced Samples in Observational Studies." Political Analysis, 20(1), 25–46. doi:10.1093/pan/mpr025

Installation

# From CRAN (when published)
install.packages("ebal")
# Development version from GitHub
# install.packages("remotes")
remotes::install_github("j-hai/ebal")

Documentation

Project page — package overview and current release notes.
Explainer — a self-contained tutorial on entropy balancing for R and Stata users, including the Lalonde NSW benchmark.

Estimands at a glance

ebalance(..., estimand = ...) chooses what gets reweighted:

`estimand`	who gets reweighted	targets	answers
`"ATT"` (default)	controls	treated-group means	"what was the effect on those who actually got treatment?"
`"ATC"`	treated	control-group means	"what would the effect have been on the control population if it had been treated?"
`"ATE"`	both	overall sample means	"what is the average effect across the whole population?"

Reading weights: always use weights(fit), which returns a length-n vector aligned to your original Treatment/X and routes the per-side semantics correctly. fit$w is side-specific (controls under ATT; treated under ATC; control side only under ATE) and is kept for backward compatibility — prefer weights(fit) in new code. See ?ebalance for the full per-estimand slot table.

Quick start

library(ebal)
# Toy data
set.seed(1)
treatment <- c(rep(0, 50), rep(1, 30))
X <- rbind(replicate(3, rnorm(50, 0)), replicate(3, rnorm(30, 0.5)))
colnames(X) <- paste0("x", 1:3)
df <- data.frame(treat = treatment, X)
# Fit (formula interface)
fit <- ebalance(treat ~ x1 + x2 + x3, data = df)
# Or matrix interface
fit <- ebalance(Treatment = treatment, X = X)
print(fit)
summary(fit)        # balance table, before vs. after weighting
plot(fit)           # Love plot of standardized differences
plot(fit, type = "weights")  # weight histogram with ESS / max-weight ratio
diagnostics(fit)    # PASS / WARN / FAIL on ESS, balance, convergence, trimming
# Use weights in a downstream regression
df$w <- weights(fit)        # length nrow(df), correct shape per estimand
df$y <- treatment + rnorm(nrow(df))            # toy outcome for the example
mod  <- lm(y ~ treat, data = df, weights = w)

Trimming extreme weights

# Auto-minimize maximum weight ratio
trimmed <- ebalance.trim(fit)
# Or specify a target
trimmed <- ebalance.trim(fit, max.weight = 5)
trimmed$trim.feasible      # TRUE if target was met (or auto mode finished)
weights(trimmed)           # length-n vector for downstream use

If the requested max.weight is infeasible the function emits a warning and returns the most recent feasible fit with trim.feasible = FALSE.

What's new in 0.2.0

Formula interface: ebalance(treat ~ x1 + x2, data = df)
print(), summary(), plot(), weights() S3 methods
New trim.feasible field on ebalance.trim objects
Numerical hardening: no more NaN crashes on aggressive trim targets; graceful fallback when the inner solve becomes singular

See NEWS.md for the full change log.

License

GPL (>= 2). See LICENSE.note.