causali

Causal Inference, Unified — A DoWhy-Inspired Framework for R

causali provides a four-step causal inference workflow for R:

1. Model — Define your causal problem (outcome, treatment, confounders)
2. Estimate — Estimate the causal effect using your method of choice
3. Assumptions — Check critical assumptions (balance, parallel trends, etc.)
4. Refute — Stress-test your result with refutation tests

It wraps battle-tested R packages (MatchIt, WeightIt, did, fixest, ivreg, rdrobust, and others) under a single, clean API.

Installation

# From CRAN (coming soon)
# install.packages("causali")

# From GitHub
remotes::install_github("mzshaik/causali")

Quick Start

library(causali)

# Load sample data
data("lalonde", package = "causali")

# Step 1: Define the causal model
model <- causal_model(
  data       = lalonde,
  outcome    = re78,
  treatment  = treat,
  confounders = c(age, educ, race, married, nodegree, re74, re75),
  label      = "Effect of job training on 1978 earnings"
)

# Step 2: Estimate the effect
result <- estimate(model, method = "matching")

# Step 3: Check assumptions
assumptions(result)

# Step 4: Refute
refute(result)

# Summary + Report
summary(result)
report(result, file = "causal_report.html")
plot(result)

Features

Method	Description	Wraps
matching	Propensity score matching	MatchIt
weighting	Inverse probability weighting	WeightIt
did	Difference-in-Differences	did (Callaway & Sant'Anna)
iv	Instrumental variables	ivreg
rdd	Regression discontinuity	rdrobust
fixest	Fixed effects regression	fixest

Check	Description
assumptions()	Balance assessment, parallel trends, overlap check
refute()	Placebo test, random common cause, subset refutation, bootstrap

Why causali?

R has over 100 packages for causal inference but no unified API. causali gives you:

• One consistent interface — same causal_model() → estimate() → assumptions() → refute() regardless of method
• Automatic assumption checking — no more forgetting to verify balance or parallel trends
• Publication-ready reporting — gt tables and ggplot2 visualisations
• Pipeline-friendly — pipe from model through estimation to refutation

License

MIT