Syllabus

Accelerated Foundation in Applied Causal Inference

Fifteen modules in five parts take you from fundamental probability and modern data engineering, through structural causal models and Double Machine Learning, to quasi-experimental designs and a deployed decision-intelligence engine (FastAPI + Next.js).

Part	Module	Core topic & methods	Applied case study	Lessons
I. Quantitative Foundations Code over calculus	01	Modern Data Engineering: Tidy Data, Polars & DuckDB Tidy data, wide vs long · Un-pivoting and key cleaning · SQL joins over DataFrames · Join audits and missing periods · Lags, windows and publication delays · Lazy queries and Parquet	Macroeconomics: joining inflation and unemployment panels	1.1 Colab, the Modern Stack and Tidy Data 1.2 Relational Joins and SQL over DataFrames with DuckDB 1.3 Aggregation, Windows and the Analysis-Ready Panel
	02	Probability & Simulation: DGPs, LLN and Sampling Data generating processes · Distributions for treatments, counts and incomes · Law of Large Numbers · Sampling distributions and the CLT · Confidence-interval coverage · Conditional probability and a first confounder	Public health: simulating a patient cohort and its comorbidities	2.1 Data Generating Processes and Distributions 2.2 Law of Large Numbers, Sampling Distributions and Confidence Intervals 2.3 Conditional Probability: the Precursor to Confounding
	03	Linear Regression Mechanics, OVB & Inference OLS as minimising squared residuals · Omitted variable bias and its formula · Partialling out (Frisch–Waugh–Lovell) · Card's returns to schooling · Robust standard errors · Clustered data	Labour economics: the Mincer earnings equation and heteroskedasticity	3.1 OLS Mechanics: the Line of Best Fit 3.2 Omitted Variable Bias and Partialling Out 3.3 Statistical Inference and Robust Standard Errors
II. Causal Frameworks & Experiments Potential outcomes, DAGs, selection on observables	04	Potential Outcomes & Randomised Trials Y(1), Y(0) and the fundamental problem · ATE, ATT, ATU and selection bias · SUTVA · Balance checks and covariate adjustment · Permutation tests and clustered designs · Power and minimum detectable effects	Education: evaluating school-voucher lotteries; NSW and HIV-incentive experiments	4.1 Potential Outcomes: ATE, ATT, ATU and Selection Bias 4.2 Randomised Experiments: the Voucher Lottery 4.3 Inference for Experiments: Permutation Tests, Clusters and Power
	05	Structural Causal Models & DAGs Chains, forks and colliders · d-separation · The backdoor criterion and adjustment sets · Good and bad controls · The front-door criterion · Collider bias, M-bias, selected samples	Social policy: job-training DAGs and collider-bias simulations	5.1 DAGs and Structural Causal Models: Chains, Forks and Colliders 5.2 The Backdoor Criterion, Adjustment Sets and the Front Door 5.3 Collider Bias, M-Bias and Selected Samples
	06	Selection on Observables: Matching, IPW & Propensity Scores Unconfoundedness and overlap · Propensity scores and common support · Nearest-neighbour and Mahalanobis matching · Balance diagnostics · IPW for ATE and ATT · Trimming and effective sample size	Welfare economics: LaLonde's NSW vs CPS and targeted assistance programmes	6.1 Selection on Observables and the Propensity Score 6.2 Matching: Propensity Score and Mahalanobis Distance 6.3 Inverse Probability Weighting, Trimming and Diagnostics
III. Advanced Causal Machine Learning Orthogonalisation, heterogeneity and instruments	07	Doubly Robust Estimation & Double Machine Learning Regression adjustment, IPW and AIPW · Double robustness · Neyman orthogonality · Cross-fitting · DoubleML PLR and IRM · Choosing nuisance learners	Public health: high-dimensional controls (NHEFS smoking cessation)	7.1 Doubly Robust Estimation: AIPW 7.2 From Partialling Out to Double Machine Learning 7.3 DoubleML in Practice: PLR, IRM and a Real Health Application
	08	Heterogeneous Effects & Policy Learning CATE and GATEs · S-, T- and X-learners · Causal forests (EconML) · Validating heterogeneity · Doubly robust policy evaluation · Policy trees	Social services: optimal targeting of retraining stipends for the unemployed	8.1 Conditional Average Treatment Effects and Meta-Learners 8.2 Causal Forests with EconML 8.3 Policy Learning: Who Should Be Treated?
	09	Instrumental Variables & the LATE Relevance, independence, exclusion, monotonicity · Wald estimator and compliers · 2SLS with linearmodels · Weak instruments · Wu–Hausman and Sargan tests · Defending an instrument	Education economics: distance to college and the exclusion restriction	9.1 Instruments, the Wald Estimator and Compliers 9.2 Two-Stage Least Squares and Weak Instruments 9.3 The Exclusion Restriction, Endogeneity Tests and Over-Identification
IV. Quasi-Experiments for Public Policy Thresholds, trends and synthetic counterfactuals	10	Regression Discontinuity Designs (Sharp & Fuzzy) Continuity assumption · Local linear estimation and RD plots · rdrobust bandwidths and robust CIs · Placebo cutoffs and covariate checks · Fuzzy RDD · Density tests and regression kinks	Local government: firm revenue at distress-score thresholds; close elections; Uruguay's PANES	10.1 Sharp Regression Discontinuity: Grants at a Distress Threshold 10.2 Optimal Bandwidths and Robust Inference with rdrobust 10.3 Fuzzy RDD, Manipulation Tests and Regression Kinks
	11	Difference-in-Differences, Event Studies & Staggered Roll-outs 2×2 DiD and parallel trends · Two-way fixed effects (pyfixest) · Event studies and pre-trends · Staggered adoption and forbidden comparisons · Cohort-by-time effects · Two-stage DiD (did2s)	Labour policy: minimum-wage event studies; castle-doctrine laws	11.1 Difference-in-Differences: the 2×2 Design and Parallel Trends 11.2 Two-Way Fixed Effects and Event Studies with pyfixest 11.3 Staggered Adoption: When TWFE Fails, and What to Use Instead
	12	Synthetic Control Methods Convex weights by optimisation · Pre-treatment fit · In-space and in-time placebos · RMSPE ratios and permutation p-values · Predictor matching · Synthetic DiD and augmented SCM	Environmental policy: provincial cap-and-trade; Texas prison expansion	12.1 Synthetic Control by Convex Optimisation 12.2 Inference for Synthetic Controls: Placebos in Space and Time 12.3 Synthetic Control on Real Data: Texas Prison Expansion
V. Robustness & Production Deployment Spillovers, sensitivity and the decision engine	13	Network Interference & Spillover Effects SUTVA failures · Exposure mappings · Direct, spillover and total effects · Partial interference · Randomised saturation designs · Peer effects and the reflection problem	Agriculture: fertiliser adoption spillovers; weather-insurance networks	13.1 When SUTVA Fails: Spillovers and Exposure Mappings 13.2 Cluster Randomisation, Partial Interference and Saturation Designs 13.3 Social Networks and Insurance Take-up: Peer Effects in Real Data
	14	Partial Identification & Sensitivity Analysis Manski worst-case bounds · What assumptions buy · Lee bounds · Rosenbaum bounds (Γ) · Robustness values (Cinelli–Hazlett) · Coefficient stability (Oster's δ)	Programme evaluation: high attrition in job training; robustness of observational findings	14.1 Partial Identification: Manski Bounds and Lee Bounds 14.2 Rosenbaum Bounds: How Much Hidden Bias Would Overturn a Matched Study? 14.3 Omitted-Variable Sensitivity: Robustness Values and Oster's δ
	15	Capstone: Executive Decision Intelligence Deployment Model artifacts and versioning · FastAPI decision endpoints · Automated tests · Next.js + Recharts dashboard · GitHub Actions CI · The executive memo and rubric	Architecture: serving a causal model via FastAPI to a Next.js dashboard	15.1 From Notebook to Decision Engine: Serving a Causal Model with FastAPI 15.2 The Executive Dashboard: Next.js, Recharts and Vercel 15.3 CI/CD, the Executive Memo and the Capstone Submission

Core texts

[CA] Huber, M. (2023). Causal Analysis: Impact Evaluation and Causal Machine Learning with Applications in R. MIT Press.
[IE] Huber, M. (2025). Impact Evaluation in Firms and Organizations: With Applications in R and Python. MIT Press.
[CM] Fabozzi, F. J., Focardi, S. & Rudin, A. (2026). Causal Modeling for Finance and Business: Foundations, Frameworks, and Applications. MIT Press.

Reading list by module

01Instructor notebooks (this module)
02Instructor notebooks (this module)
03Instructor notebooks (this module); Huber, Causal Analysis, Ch. 3 (background on regression)
04Huber, Impact Evaluation in Firms and Organizations, Ch. 2 (basics) & Ch. 3 (A/B testing); Huber, Causal Analysis, Ch. 2 & 3.1
05Fabozzi, Focardi & Rudin, Causal Modeling for Finance and Business, Ch. 1, 4 & 5
06Huber, Causal Analysis, Ch. 4 (matching, propensity scores, IPW); Huber, Impact Evaluation in Firms and Organizations, Ch. 4
07Huber, Causal Analysis, Ch. 5.1–5.3; Huber, Impact Evaluation in Firms and Organizations, Ch. 5.1–5.3
08Huber, Causal Analysis, Ch. 5.4–5.5; Huber, Impact Evaluation in Firms and Organizations, Ch. 5.4
09Huber, Causal Analysis, Ch. 6; Huber, Impact Evaluation in Firms and Organizations, Ch. 6; Fabozzi, Focardi & Rudin, Ch. 7
10Huber, Causal Analysis, Ch. 9.1; Huber, Impact Evaluation in Firms and Organizations, Ch. 7
11Huber, Causal Analysis, Ch. 7; Huber, Impact Evaluation in Firms and Organizations, Ch. 8
12Huber, Causal Analysis, Ch. 8; Huber, Impact Evaluation in Firms and Organizations, Ch. 9; Fabozzi, Focardi & Rudin, Ch. 11
13Huber, Causal Analysis, Ch. 11
14Huber, Causal Analysis, Ch. 10
15Fabozzi, Focardi & Rudin, Ch. 8 (decision-making), Ch. 9 (business applications) & Ch. 10 (deployment)

How every lesson works

Concept and simulation: build the data generating process so the true effect is known.
Real benchmark: apply the estimator to a classic dataset (LaLonde NSW, Card, NHEFS, close elections, castle-doctrine laws, Texas prisons…).
Exercise: an unguided task, with a solution to check against.

Everything runs in Google Colab: each lesson has a notebook whose first cell installs the packages it needs. The capstone moves to GitHub Codespaces for the FastAPI backend and Next.js dashboard.

Capstone grading rubric

Data Engineering & Automation (20%). Raw data ingested and transformed with Polars/DuckDB into a tidy panel; reproducible Codespace or requirements; CI tests on every push.
Methodological & Structural Rigour (30%). A defended DAG; explicit assumption checks (overlap, parallel trends, continuity, first stage); sensitivity analysis or bounds.
Causal ML Execution (25%). Correct use of DoubleML/EconML for ATE, LATE or HTE; cross-fitting; compared and tuned nuisance learners.
Full-Stack Executive Deployment (25%). Model served by a tested FastAPI endpoint; a usable Next.js/Recharts dashboard; a README written as a two-page executive memo.

The capstone executive memo (two pages)

Policy context

The intervention, the population and the alternative it is compared with.

Structural assumptions

The DAG in plain words: selection on observables, an instrument or a design.

Impact estimates

Dollar or percentage-point effects, labelled ATE, ATT, LATE or HTE, with intervals.

Sensitivity & limits

Bounds, hidden-bias robustness and possible spillovers (SUTVA).

Deployment

How the model is served, monitored and retrained for ongoing executive use.