dagwise 0.1.0

A full-stack Python causal inference engine — DAGs, identification, estimation, and refutation

🔷 dagwise

A full-stack DAG-based causal inference engine for Python

Answer "does X cause Y?" — not just "are X and Y correlated?"

---

What is dagwise?

dagwise is a Python library for causal inference — the science of determining whether X truly causes Y, not just correlates with it.

Unlike machine learning which predicts outcomes, dagwise answers "what if" questions:

• Would this drug have worked if given to a different patient?
• Did this marketing campaign actually drive sales?
• What's the true effect of this policy?

✨ Features

Feature	dagwise
DAG (no networkx dependency)	✅
Full ID Algorithm (Shpitser-Pearl)	✅
Double ML	✅
Staggered DiD (Callaway-Sant'Anna)	✅
Refutation Tests	✅
Sensitivity Analysis	✅
Clean Pythonic API	✅

📦 Installation

pip install dagwise

# With plotting support
pip install dagwise[plot]

# With DAG visualization
pip install dagwise[viz]

# Everything
pip install dagwise[all]

⚡ Quickstart

import pandas as pd
from dagwise import CausalModel
from dagwise.core import CausalDAG

# 1. Define your causal structure
dag = CausalDAG()
dag.add_edge("Age", "Smoking")
dag.add_edge("Age", "LungCancer")
dag.add_edge("Smoking", "LungCancer")

# 2. Create causal model
model = CausalModel(dag=dag, treatment="Smoking", outcome="LungCancer")

# 3. Identify the causal effect
result = model.identify()
print(result.estimand)
# P(LungCancer|do(Smoking)) = Σ_Age P(LungCancer|Smoking,Age)·P(Age)

# 4. Estimate the effect
data = pd.read_csv("smoking_data.csv")
ate = model.estimate(data=data, method="double_ml", covariates=["Age"])
print(ate.summary())

# 5. Validate with refutation tests
from dagwise.inference import RefutationTest
refutation = RefutationTest.placebo_treatment(estimator=ate, data=data,
                                               treatment="Smoking", outcome="LungCancer")
print(f"Refutation passed: {refutation.passed}")

🧠 Supported Methods

Identification

• Backdoor Criterion & Adjustment Sets
• Frontdoor Criterion
• ID Algorithm (Shpitser & Pearl, 2006) — full general identification
• Instrumental Variables

Estimation

• Inverse Probability Weighting (IPW)
• Outcome Regression (G-computation)
• Doubly Robust / AIPW
• Double ML (Chernozhukov et al., 2018)
• Propensity Score Matching
• Instrumental Variables (2SLS)
• Difference-in-Differences (2x2 + Staggered/Callaway-Sant'Anna)
• Regression Discontinuity Design

Inference & Validation

• Bootstrap Confidence Intervals
• Sensitivity Analysis (E-values, Rosenbaum Bounds)
• Refutation Tests (Placebo, Random Common Cause, Data Subset)

📖 Documentation

• Quickstart Guide
• API Reference
• DAG & Identification Concepts
• Estimators Guide
• Examples

🤝 Contributing

Contributions are welcome! Please read CONTRIBUTING.md first.

git clone https://github.com/keyreyla/dagwise.git
cd dagwise
pip install -e ".[dev]"
pytest

📄 License

MIT — see LICENSE

📚 References

• Shpitser & Pearl (2006) — ID Algorithm
• Chernozhukov et al. (2018) — Double ML
• Callaway & Sant'Anna (2021) — Staggered DiD
• Pearl (2009) — Causality: Models, Reasoning, and Inference

Citation

@software{dagwise2025,
  title  = {dagwise: A full-stack DAG-based causal inference engine},
  author = {Your Name},
  year   = {2025},
  url    = {https://github.com/keyreyla/dagwise}
}