Performing Causal Tasks

Once we have modeled causal relationships, we are ready to move to the next step: performing causal tasks. DoWhy supports the following tasks.

1. Effect estimation: If we change X, how much will it cause Y to change?

2. Attribution: Why did an event happen? How to explain an outcome? Which of my variables caused the anomaly?

3. Counterfactual estimation: What if X had been changed to a different value than its observed value? What would have been the values of other variables?

4. Prediction: Given an input with new values of some input features, what will be the value of Y?

This chapter describes how to perform each of these tasks.

• Estimating Causal Effects
  • Identifying causal effect
  • Estimating average causal effect using backdoor
  • Estimating average causal effect with natural experiments
  • Estimating conditional average causal effect
  • Estimating average causal effect using GCM
  • Estimating direct and indirect effects
• Explaining Observed Effects and Root-Cause Analysis
  • Anomaly Attribution
  • Attributing Distributional Changes
  • Quantifying Intrinsic Causal Influence
  • Unit Change Attribution
  • Feature Relevance
• Asking and Answering What-If Questions
  • Simulating the Impact of Interventions
  • Computing Counterfactuals
• Predicting outcome for out-of-distribution inputs