Causal Modeling — causes, effects and correlation: practical intensive

Course overview

Description generated based on course syllabus and open data.

This course covers causal modeling (Causal Modeling) using Python and Double Machine Learning to estimate causal effects beyond correlation. Format: 2 days, 6 hours total, with practical cases and real-world walkthroughs.

Who it suits and who it does not: Causal Modeling, causality and correlation

• Suits: Junior Data Scientist — to systematize causality basics and start with DML.
• Suits: Senior Data Scientist, Data Analyst, BI Engineer — for effect estimation, ATE/CATE, uplift, and sound A/B analysis.
• Suits: Team Lead, Technical PM, Product Manager — for interpreting causal effects in products and services.
• Not suited: those expecting correlation-only methods without causal inference.
• Not suited: no basic Python and data handling skills.
• Not suited: if only ready-made templates are desired without analytical work.

Problem → outcome in DML-based causal modeling

• Problem: Correlation misleads in product and marketing → Outcome: DML for causality with confounder handling.
• Problem: Biased effect estimates due to mixed factors → Outcome: separate outcome and treatment models with cross-fitting and regularization.
• Problem: Hard-to-interpret A/B tests → Outcome: causal metrics ATE, CATE, uplift and robustness checks.
• Problem: Difficulty forecasting impact of changes → Outcome: counterfactual scenarios via causal graphs and DML.

Comparison with alternatives: correlation vs causation

• Correlation analysis: fast but cannot separate causes from effects and is confounder-sensitive.
• A/B experiments without causal framing: valid locally, hard to generalize across segments.
• Classical econometrics without DML: relies on strong assumptions and simple models.
• Causal modeling with DML: combines ML and identification, reduces bias and exposes segment-level effects (CATE).

Post-course outcomes: Causal Modeling in Python with DML

• Understanding of correlation vs causation and the role of confounders.
• Practice with ATE, CATE, uplift estimation and sensitivity analysis.
• Construction of DML pipelines for product analytics and A/B testing.
• Interpretation of causal effects for decision support and scenario forecasting.

Structure and syllabus: 2 days, 6 hours (Causal Modeling, DML, Python)

• Day 1: causation vs correlation, causal graphs, identification, DML concept, Python setup.
• Day 2: ATE, CATE, uplift modeling, robustness checks, A/B case studies, impact on key metrics.

Lecturer and experience

Lecturer: Rostyslav Buchko, Principal Data Scientist at TextNow, former Principal Data Scientist at Amazon retail; Silicon Valley experience and publications in medicine, innovation and Data Science.

Tools and data for causal models

• Language and libraries: Python, pandas, scikit-learn, econml, DoWhy.
• Metrics and checks: ATE, CATE, uplift, cross-fitting validation, sensitivity analysis.
• Use cases: product features, marketing campaigns, A/B experiments.