cegraph

Getting Started

This guide walks you through your first causal analysis with cegraph. You'll create a causal graph, add variables and relationships, and estimate an average causal effect — all in under 5 minutes.

1. Create a Graph

Start by importing CausalEffectGraph and creating an empty graph: 

from cegraph import CausalEffectGraph

ceg = CausalEffectGraph()

2. Add Variables and Edges

Add directed edges to define causal relationships. Each edge represents a direct causal effect from one node to another: 

ceg.add_edge("age", "treatment")
ceg.add_edge("age", "outcome")
ceg.add_edge("treatment", "outcome")

This creates a simple confounding structure where age confounds the relationship between treatment and outcome.

3. Inspect the Graph

Query the graph to explore its structure: 

print(ceg.has_edge("treatment", "outcome"))  # True
print(ceg.parents("outcome"))                # ['age', 'treatment']
print(ceg.children("age"))                   # ['treatment', 'outcome']

4. Compute a Causal Effect

Generate some synthetic data and estimate the average causal effect: 

import numpy as np

# Simulate data: age -> treatment, age -> outcome, treatment -> outcome
n = 1000
age = np.random.randn(n)
treatment = 0.5 * age + np.random.randn(n) * 0.5
outcome = 0.8 * treatment + 0.3 * age + np.random.randn(n) * 0.5

data = np.column_stack([age, treatment, outcome])

# Estimate ACE with backdoor adjustment
ace = ceg.average_causal_effect(
    data=data,
    treatment="treatment",
    outcome="outcome",
    covariates=["age"],
)
print(f"Average Causal Effect: {ace:.4f}")

The estimated ACE should be close to the true effect of 0.8.

5. Find Adjustment Sets

Let the graph automatically find valid adjustment sets: 

adj = ceg.backdoor_adjustment_set(
    target="outcome",
    cause="treatment",
)
print(f"Backdoor adjustment set: {adj}")

Next Steps

Edit this page on GitHub