Development Economics Guide

A skill for conducting development economics research, covering impact evaluation methods, field experiment design, household survey analysis, key data sources, and the methodological toolkit used to study poverty, education, health, and institutions in developing countries.

Impact Evaluation Methods

The Identification Problem

Fundamental question: What is the causal effect of a program/policy?

Challenge: We observe outcomes for treated individuals, but we cannot
observe what would have happened to them without treatment
(the counterfactual).

Solutions (from strongest to weakest causal identification):
  1. Randomized Controlled Trials (RCTs / field experiments)
  2. Regression Discontinuity Design (RDD)
  3. Instrumental Variables (IV)
  4. Difference-in-Differences (DiD)
  5. Matching / Propensity Score Methods
  6. Cross-sectional regression with controls (weakest)

Randomized Controlled Trials in Development

def design_field_experiment(intervention: str,
                             unit: str,
                             clusters: int,
                             expected_effect: float) -> dict:
    """
    Design a cluster-randomized field experiment.

    Args:
        intervention: Description of the program/policy
        unit: Unit of randomization (individual, household, village, school)
        clusters: Number of clusters available
        expected_effect: Expected effect size (standard deviations)
    """
    return {
        "intervention": intervention,
        "randomization_unit": unit,
        "design_considerations": {
            "cluster_vs_individual": (
                "Cluster randomization when intervention operates at group level "
                "or to avoid spillovers between treated and control within clusters."
            ),
            "stratification": (
                "Stratify randomization by baseline covariates (e.g., region, "
                "baseline outcome) to improve balance and statistical power."
            ),
            "sample_size": {
                "clusters": clusters,
                "note": (
                    "With cluster randomization, power depends more on number "
                    "of clusters than individuals per cluster. Aim for 20+ "
                    "clusters per arm. Account for ICC (intracluster correlation)."
                )
            },
            "expected_effect": expected_effect,
            "pre_registration": "Register at AEA RCT Registry (socialscienceregistry.org)"
        },
        "threats": [
            "Attrition (differential dropout between arms)",
            "Non-compliance (some treated do not take up, some controls do)",
            "Spillovers (treatment affects control units)",
            "Hawthorne effects (behavior changes from being observed)",
            "Ethical concerns (withholding a beneficial intervention)"
        ]
    }