User Guide

This guide will help you get started with Causal Inference and Causalis and understand its core concepts.

• Introduction to Causal Inference
  • Why causality?
  • The core idea
  • Notation
  • What we aim to estimate (estimands)
  • When are these estimands credible?
• Installation
  • Prerequisites
  • Install from pip
  • Install from GitHub
  • Optional: conda environment after cloning the repo
• Why DML IRM is the Best Modern Choice for Causal Inference
  • 1. Combines Flexibility with Rigor
  • 2. Doubly Robust Protection
  • 3. Neyman Orthogonality: Immunity to Regularization Bias
  • 4. Cross-Fitting Eliminates Overfitting Bias
  • 5. High-Dimensional Capability
  • 6. Unified Framework for Multiple Estimands
  • 7. Built-In Diagnostic Framework
  • 8. Proven Theoretical Foundation
  • 9. Practical Performance
  • 10. Transparent Confounding Adjustment
  • When DML IRM Shines
  • Math of DML IRM
  • Scores
• CausalData at a glance
  • Quick start
  • API essentials
  • Validation (on construction)
  • Common snippets
  • Tips
• Inference
  • A Very Short ATE Example
  • Advanced Usage
• Refutation: quick checks and sensitivity
  • 1) Placebo and subset checks
  • 2) Overlap (positivity) diagnostics
  • 3) Score/orthogonality diagnostics (Neyman)
  • 4) Unconfoundedness/balance checks
  • 5) Sensitivity analysis to hidden confounding
  • 6) SUTVA (design prompts)
  • Minimal workflow (ATE)
• Math notation for EDA, IRM, and refutation diagnostics (CausalKit)
  • 1. Variables and parameters
  • 2. Nuisance functions (IRM)
  • 3. Scores, EIFs, and estimators
  • 4. Estimation (cross-fitting)
  • 5. Positivity (overlap) & trimming
  • 6. Refutation diagnostics
  • 7. Implementation guardrails