Example Notebooks

Contents:

• Getting started with DoWhy: A simple example
  • Interface 1 (recommended): Input causal graph
  • Interface 2: Specify common causes and instruments
  • Refuting the estimate
    • Adding a random common cause variable
    • Adding an unobserved common cause variable
    • Replacing treatment with a random (placebo) variable
    • Removing a random subset of the data
• Confounding Example: Finding causal effects from observed data
  • Let’s create a mystery dataset for which we need to determine whether there is a causal effect.
  • Using DoWhy to resolve the mystery: Does Treatment cause Outcome?
    • STEP 1: Model the problem as a causal graph
    • STEP 2: Identify causal effect using properties of the formal causal graph
    • STEP 3: Estimate the causal effect
    • Checking if the estimate is correct
    • Step 4: Refuting the estimate
    • Adding a random common cause variable
    • Replacing treatment with a random (placebo) variable
    • Removing a random subset of the data
• DoWhy: Different estimation methods for causal inference
  • Identifying the causal estimand
  • Method 1: Regression
  • Method 2: Stratification
  • Method 3: Matching
  • Method 4: Weighting
  • Method 5: Instrumental Variable
  • Method 6: Regression Discontinuity
• Conditional Average Treatment Effects (CATE) with DoWhy and EconML
  • Linear Model
  • EconML methods
    • CATE Object and Confidence Intervals
    • Can provide a new inputs as target units and estimate CATE on them.
    • Can also retrieve the raw EconML estimator object for any further operations
  • Works with any EconML method
    • Continuous treatment, Continuous outcome
    • Binary treatment, Binary outcome
    • Instrumental Variable Method
  • Refuting the estimate
    • Random
    • Adding an unobserved common cause variable
    • Replacing treatment with a random (placebo) variable
    • Removing a random subset of the data
• Demo for the DoWhy causal API
  • Comparing the estimate to Linear Regression
• Do-sampler Introduction
  • Pearlian Interventions
  • Statefulness
  • Integration
  • Specifying Interventions
  • Demo
• Different ways to load an input graph
  • I. Generating dummy data
  • II. Loading GML or DOT graphs
    • GML format
    • DOT format
• Simple example on using Instrumental Variables method for estimation
• DoWhy example on ihdp (Infant Health and Development Program) dataset
  • Loading Data
  • 1.Model
  • 2.Identify
  • 3. Estimate (using different methods)
    • 3.1 Using Linear Regression
    • 3.2 Using Propensity Score Matching
    • 3.3 Using Propensity Score Stratification
    • 3.4 Using Propensity Score Weighting
  • 4. Refute
    • 4.3 Data Subset Refuter
• DoWhy example on the Lalonde dataset
  • 1. Load the data
  • Run DoWhy analysis: model, identify, estimate
  • Sanity check: compare to manual IPW estimate
• Lalonde Pandas API Example
  • Getting the Data
  • The causal Namespace
  • The do Operation
  • Treatment Effect Estimation
  • Specifying Interventions