Notes on Causal Inferencing Lecture

Introduction

• Topic: Causal Inferencing
• Context: Special episode of Code Emporium
• No face video due to focus on technical details
• Community: Join Discord server for community interaction

Randomized Controlled Tests (RCTs)

• Known as A/B tests in industry
• Use RCTs to test effects, e.g., impact of emails on purchase conversion in eCommerce

Steps in RCTs

1. Select users on uniform criteria
2. Split users into two groups (control and treatment)
3. Send emails to treatment group only
4. Monitor purchase conversion
5. Analyze results to infer impact of emails

Importance

• Randomization: Ensures comparability between groups by controlling for variables
• Causality: Helps in proving causation, e.g., email increases conversion

Limitations of RCTs

• Impracticality: Some tests are impossible to set up (e.g., billboard ads in random cities)
• Time constraints: Experiments might take too long

Challenges in Causal Inferencing

1. Confounders

• Example: Medical trial with age as a confounding variable
• Definition: Variables that affect both treatment and outcome
• Solution: Randomization or controlling confounders in analysis

2. Selection Bias

• Occurs when treatment group isn't representative of the population

3. Counterfactuals

• Definition: Hypothetical scenario of what would happen without treatment
• Methods: Machine learning, matching

Assumptions for Causality

1. Causal Markov Condition

• Uses causal graphs to represent causation

2. SUTVA (Stable Unit Treatment Value Assumption)

• Ensures no interaction between treatment and control groups

3. Ignorability

• Assumes no missing confounders affecting treatment and outcome

Measuring Treatment Effects

Average Treatment Effect (ATE)

• Calculation: Difference in success rates between treatment and control
• Adjust for confounders using counterfactuals

Conditional Average Treatment Effect (CATE)

• Evaluates treatment effect conditioned on variables (e.g., age)
• Reveals treatment heterogeneity

Conclusion

• Summary: Introduced RCTs, challenges in causal inferencing, necessary assumptions, and methods to measure treatment effects
• Future Content: More in-depth exploration of causal inferencing planned
• Community Engagement: Encourage joining Discord and subscribing for updates