Causal Inference Estimators for Observational and Experimental Studies • CausalModel

CausalModel is an R port of the Python CausalModel package from Qu et al. (2021). It collects estimators for three settings I often teach together: observational studies, randomized experiments, and clustered interference.

Installation

# install.packages("devtools")
devtools::install_github("username/RCausalModel")

Observational Studies

library(CausalModel)

# Generate synthetic data with known ATE = 10
data <- generate_data(N = 5000, k = 2, tau = 10)
obs <- observational(data$Y, data$Z, data$X)

# Estimators
est_via_ols(obs)        # OLS with HC0 robust SE
est_via_ipw(obs)        # Inverse probability weighting
est_via_aipw(obs)       # Augmented IPW (doubly robust)
est_via_matching(obs)   # Nearest-neighbor matching
estimate(obs)           # Default (AIPW for binary Z)

# Continuous treatment -> Double ML
data_c <- generate_data_continuous(N = 5000, k = 2, tau = 10)
obs_c <- observational(data_c$Y, data_c$Z, data_c$X)
est_via_dml(obs_c)

Randomized Experiments

exp_obj <- experimental(Y, Z, X)

est_via_dm(exp_obj)                    # Difference in means
est_via_strata(exp_obj, strata = s)    # Stratified estimator
est_via_ancova(exp_obj)                # ANCOVA with interactions
test_via_fisher(exp_obj, n_draws = 1000)  # Fisher randomization test

Experimental Designs

design <- crd(treated_ratio = 0.5)        # Completely randomized
design <- bernoulli(treated_prob = 0.5)   # Bernoulli
Z <- draw(design, n = 100)               # Draw treatment vector

# Rerandomization for covariate balance
design <- crd(treated_ratio = 0.5, covariate = X, balance = TRUE, eps = 0.1)

Network Interference

# Generate clustered data with spillover effects
data <- generate_fixed_cluster(clusters = 5000, group_struct = c(2, 3, 4),
                                k = 2, tau = 1, gamma = c(0.5, 0.1, 0.2))

cl <- clustered(data$Y, data$Z, data$X,
                data$cluster_labels, data$group_labels,
                data$ingroup_labels)

# Estimate treatment effects under interference
result <- est_via_ipw(cl)    # IPW under partial interference
result <- est_via_aipw(cl)   # AIPW under partial interference

Custom Models

Supply your own outcome/propensity models:

# Built-in wrappers
est_via_aipw(obs, outcome_model = cm_ols(), propensity_model = cm_logistic())

# Random forest (requires randomForest package)
est_via_aipw(obs,
  outcome_model = cm_random_forest_regressor(),
  propensity_model = cm_random_forest_classifier())

Methods

Estimator	Function	Setting
OLS	`est_via_ols()`	Observational
IPW	`est_via_ipw()`	Observational / Interference
AIPW	`est_via_aipw()`	Observational / Interference
Matching	`est_via_matching()`	Observational
Double ML	`est_via_dml()`	Observational (continuous Z)
Difference in Means	`est_via_dm()`	Experimental
Stratified	`est_via_strata()`	Experimental
ANCOVA	`est_via_ancova()`	Experimental
Fisher Test	`test_via_fisher()`	Experimental

Vignettes

Full documentation: https://xiangao.github.io/RCausalModel/

Vignette	Description
Quickstart	Getting started: basic usage of all estimators
Simulations	Monte Carlo validation across all settings
Observational	Observational estimators with per-estimator simulations
Experimental	Design objects and experimental estimators
Interference	Clustered interference with normality and coverage simulations
Reference index	All documented functions on one page

References

Qu, Z., McNamara, R., Jia, R., & Imbens, G. (2021). CausalModel: A Python Package for Causal Inference.