Skip to content
LeadAfrik
Advanced · Self-paced2026 Edition

Impact Evaluation & Randomised Trials

How to know whether a policy actually worked. The counterfactual problem, the randomised controlled trial and the credibility revolution, quasi-experimental methods when you can't randomise, and the leap from a clean estimate to a policy decision.

8

Modules

~6h 50m

Reading time

Advanced

Level

Self-paced

Format

§

Syllabus

  1. 01

    The fundamental problem of causal inference

    The counterfactual, the potential-outcomes framework, and selection bias — why a before-after comparison usually lies.

    ~50 minModule 01
    →
  2. 02

    The randomised controlled trial

    Why randomisation identifies a causal effect, the credibility revolution, and the 2019 Nobel for Banerjee, Duflo, and Kremer.

    ~50 minModule 02
    →
  3. 03

    Designing an RCT

    Power calculations and minimum detectable effects, the unit of randomisation, stratification, and the pre-analysis plan.

    ~55 minModule 03
    →
  4. 04

    Threats to validity

    Attrition, spillovers, Hawthorne and John Henry effects, and non-compliance — with the local average treatment effect it leaves you.

    ~50 minModule 04
    →
  5. 05

    When you can't randomise — difference-in-differences

    The parallel-trends assumption, two-way fixed effects, event-study plots, and the staggered-adoption pitfalls.

    ~55 minModule 05
    →
  6. 06

    Regression discontinuity

    The running variable and the cut-off, sharp vs fuzzy designs, and why the estimate is local to the threshold.

    ~50 minModule 06
    →
  7. 07

    Matching and instrumental variables

    Propensity-score matching and selection on observables, and the instrument that buys you causation when selection is on unobservables.

    ~50 minModule 07
    →
  8. 08

    From estimate to policy

    External validity, the scale-up problem, cost-effectiveness comparison, and using the J-PAL evidence base responsibly.

    ~50 minModule 08
    →

How to use this course

Start with module 01 if the material is new; skip ahead if you have prior exposure. Each module is self-contained but the arc is sequential — the projects in the final module assume the toolkit from modules 1-11. Every module ends with key takeaways and a curated further-reading list with primary sources.