Flagship course · MEL and data
Causal Inference for Development: Designs, Estimators & Judgement
Judge which design a causal claim can bear, from the potential-outcomes model to synthetic control and causal machine learning.
Start module 1What you will be able to do
- State what has to be true for a comparison to recover a causal effect before you run the regression.
- Run each design in R and Stata on household surveys, programme rosters and district panels.
- Read a results table, question its design and decide what claim it can carry when you commission evidence.
About this course
From Potential Outcomes to Practitioner Practice
A design-based course in causal inference for people who design, read, and commission impact evidence. It starts with the potential-outcomes model, works through the estimator toolkit from randomisation to synthetic control and causal machine learning, and ends with the practitioner's real job, which is judging which design a claim can bear. Built on Imbens and Rubin, Angrist and Pischke, and Cunningham, with cases from Indian programmes.
- R + Stata Throughout
- South Asia Focus
- Identification First
Why a whole course on causal inference?
Development practice runs on causal claims. A programme raised enrolment. A transfer cut stunting. A reform moved wages. Most of these claims are made from comparisons that cannot bear them. This course is about the gap between the comparison you can run and the one you actually need, and the designs that close it.
Where this sits. It goes one level past Econometrics 101, which introduces the methods. It carries the conceptual spine of the MEL flagship (counterfactual, attribution, contribution) into the estimators that operationalise it. If you have run a comparison and wondered whether it was honest, start here.
Identification before estimation
Every method is taught as an answer to one question: what has to be true for this comparison to recover a causal effect? The assumption comes first, the regression second.
Code you can run
R and Stata for each design, on data shapes you will actually meet: household surveys, programme rosters, district panels. Paired with the DevEconomics Toolkit's DiD, RDD, and synthetic control apps.
The buyer's seat
The final part is for the practitioner who commissions evidence rather than produces it: reading a results table, interrogating a design, and choosing what a claim can be made to bear.
"The most credible and influential research designs use random assignment… We hope to find natural or quasi-experiments that mimic a randomized trial by changing the variable of interest while other factors are kept balanced." : Joshua Angrist & Jörn-Steffen Pischke, Mostly Harmless Econometrics (2009), p. 11
Assess yourself: causal inference
Six auto-graded questions on the core ideas of the course. Pick an answer and check it; each explains the reasoning. Nothing is stored and there's no sign-in.
Papers & Resources
A short, opinionated reading list. The textbooks anchor the theory; the rest are the field's reference points for each design.
Core textbooks
Imbens & Rubin, Causal Inference for Statistics, Social, and Biomedical Sciences (2015). Angrist & Pischke, Mostly Harmless Econometrics (2009). Cunningham, Causal Inference: The Mixtape (2021, free online). Morgan & Winship, Counterfactuals and Causal Inference (2nd ed., 2014).
Sibling courses
Pairs with The Evidence Question (MEL) for the conceptual spine and Econometrics 101 for the entry-level treatment. The DevEconomics Toolkit ships interactive DiD, RDD, and synthetic control apps used in the worked examples.
The credibility debate
Banerjee & Duflo, Poor Economics (2011), against Deaton, "Instruments, Randomization, and Learning about Development" (JEL 2010) and Pritchett & Sandefur on external validity (2015). Read all three before Module 12.
Practice the material and continue learning
Every flagship course is part of a wider open-source learning network. The cards below cross-link this course with hands-on labs, the course lexicon, the 101 Series, book summaries, reference handouts, live dojos, and Pro Studio tools.
The companion tool to this course. Frame a causal question, choose a design (RCT, difference-in-differences, regression discontinuity, matching, synthetic control), estimate power and stress-test the threats to validity. Not sure an RCT fits? Run the RCT Readiness Diagnostic first.
Walk from an evaluation question to a defensible sampling strategy (simple random, stratified, cluster or multistage), sized for power and minimum detectable effect, and see how design choices shape what you can detect.
Searchable definitions for the identification vocabulary: ATE, ATT and LATE, DiD, RDD, IV, propensity scores, parallel trends, and the SUTVA and ignorability assumptions.
Free foundational primers that pair well with this flagship:
Interactive Reading Companions across the ImpactMojo library, from Mostly Harmless Econometrics to Poor Economics, that deepen this flagship.
84 print-optimised handouts across 10 tracks: methods, ethics, frameworks, lexicons, quick-reference cards.
56 bi-weekly dojo sessions in the South Asian dev-practitioner cohort: case clinics, paper discussions, live Q&A.
11 Pro Studio tools plus 1:1 coaching, cohort access, and certificates. Coaching is offered on a sliding scale.