Nedensel Çıkarım

Causal Reasoning

1. Prediction versus CausationWhy a good predictor can be a bad guide to intervention: confounding, Simpson's paradox, and the question 'what if we act?'.1–2 sa
2. Causal Graphs and the Backdoor CriterionEncoding causal assumptions as a DAG and deciding which variables to adjust for.1–2 sa
3. Causal Discovery and Structure LearningRecovering a causal graph from data: independence tests, Markov equivalence, and what data cannot orient.1–2 sa

Effect Estimation

1. Potential Outcomes and Treatment EffectsEach unit has an outcome under treatment and under control; we only see one. ATE, ATT, and identification assumptions.1–2 sa
2. Randomized ExperimentsRandomisation breaks confounding by design; the gold standard and its practical limits.1–2 sa
3. Propensity Scores and Inverse Probability WeightingAdjusting observational data by modelling the probability of treatment.1–2 sa
4. Quasi-Experimental Designs: Instruments and Difference-in-DifferencesNatural experiments when randomisation is impossible: instrumental variables, difference-in-differences, regression discontinuity.1–2 sa
5. Heterogeneous Effects and Machine Learning for Causal EstimationUsing flexible ML models to estimate effects that vary by individual, with orthogonalisation to avoid bias.1–2 sa