CAUSAL INFERENCE FOR DATA ANALYSTS

Your data can tell you what happened. It can predict what may happen next. But can it tell you what will actually change if you take action?

A customer who uses a feature may be more likely to stay-but did the feature cause the retention?

A marketing campaign may reach customers who spend more-but did the campaign generate those purchases?

A new policy may appear alongside better performance-but would the improvement have happened anyway?

These are not ordinary analytics questions. They are causal questions.

Causal Inference for Data Analysts is a practical guide to moving beyond correlation, prediction, and conventional reporting to understand whether an intervention actually changes an outcome.

Written for analysts working with real business and technology data, this book focuses on the reasoning that must come before sophisticated modeling: defining the decision, specifying the treatment, identifying the right comparison, exposing assumptions, selecting a credible design, and determining whether the available data can support the causal claim at all.

Inside, you'll learn how to:

• Distinguish descriptive, predictive, and causal questions before choosing a model
• Think in counterfactuals and define the comparison that a decision actually requires
• Work with ATE, ATT, ATU, CATE, and other decision-relevant estimands
• Use directed acyclic graphs to reason about confounders, mediators, colliders, and adjustment sets
• Understand why controlling for more variables can sometimes increase rather than reduce bias
• Determine whether a causal effect is identifiable before selecting an estimator
• Design and interpret randomized experiments and A/B tests without confusing assignment with exposure
• Diagnose sample-ratio mismatch, attrition, interference, noncompliance, peeking, and multiple testing
• Use regression adjustment, matching, propensity scores, weighting, standardization, and doubly robust methods
• Apply instrumental variables, regression discontinuity, difference-in-differences, and synthetic control
• Estimate heterogeneous effects and understand where causal machine learning can-and cannot-help
• Work with causal forests, meta-learners, double machine learning, uplift modeling, and policy targeting
• Analyze panel data, repeated interventions, time-varying effects, and dynamic decisions
• Test the credibility of conclusions using balance checks, overlap diagnostics, placebo tests, negative controls, and sensitivity analysis
• Separate statistical uncertainty from uncertainty in the causal assumptions themselves
• Build reproducible professional workflows that connect data preparation, identification, estimation, diagnostics, and decision-making
• Communicate causal evidence without presenting assumptions as certainty

One of the book's most important principles is that a sophisticated model cannot rescue a weak causal design. The manuscript repeatedly emphasizes that analysts must define the causal question and identification strategy before choosing the estimator.

The book also goes beyond traditional causal methods by examining heterogeneous treatment effects and causal machine learning, including meta-learners, causal forests, doubly robust learners, and double machine learning-while warning against mistaking flexible prediction for causal identification.

Whether you're a data analyst, product analyst, business analyst, data scientist, experimentation professional, growth analyst, marketing analyst, operations analyst, economist, or analytics leader, this book will help you ask a more valuable question:

Not simply "What is associated with the outcome?" but "What would actually change the outcome if we intervened?"

Move beyond reporting pattern