Questions: Randomized Experiments in Development Economics

Question 1

A deworming RCT in Kenya found a 25% increase in school attendance. A policymaker wants to immediately scale the program nationally in Guatemala. What is the most important methodological concern?

• A. The study had selection bias because villages self-selected into the program
• B. External validity — results from one context may not generalize to different health burdens, institutions, and incentive structures elsewhere
• C. The control group was contaminated by exposure to the treatment villages
• D. The RCT sample size was too small to achieve statistical significance

Answer: (B)

RCTs solve selection bias through randomization, so option A is incorrect. The key concern here is external validity: an RCT establishes a causal effect in a specific context, but that effect may not transfer to another country with different disease ecology, school systems, parental incentives, and implementation capacity. This is one of the principal limitations of RCTs in development economics — causal identification in one place is not the same as a universal finding.

Question 2

Why does random assignment address selection bias in a way that controlling for observed confounders in observational studies cannot?

• A. Randomization guarantees that the treatment and control groups have identical individual characteristics
• B. Randomization makes treatment and control groups statistically equivalent in expectation across all confounders — observed and unobserved
• C. Random assignment eliminates the need for statistical analysis because the groups are identical
• D. Observational studies cannot control for any confounders, only RCTs can

Answer: (B)

Option A overstates the claim — randomization does not make individual groups identical, it makes them equivalent in expectation across the full distribution of characteristics. The critical advantage over observational methods is that randomization balances all confounders — including ones the researcher never measured or even thought of. Observational studies can control for observed confounders using matching or regression, but unmeasured confounders remain a threat. RCTs remove that threat by design.

Question 3

RCTs in development economics eliminate ethical concerns about withholding beneficial programs, because the control group receives the program once the trial ends.

• T. True
• F. False

Answer: False

This is false. Ethical concerns are a genuine limitation of RCTs. During the trial period, the control group does not receive the program — and if the program turns out to be beneficial, those individuals were denied a potentially valuable intervention. Researchers and IRBs must weigh the social value of rigorous evidence against the harm of withholding treatment. Ethical issues are not resolved simply by later rollout; they bear on the design and justification of the trial itself.

Question 4

An RCT-tested development program that succeeded in a carefully managed research trial will work equally well when implemented at national scale by government agencies.

• T. True
• F. False

Answer: False

This is the scalability problem, a major limitation of RCTs. Small pilots benefit from researcher oversight, motivated field staff, tight implementation protocols, and Hawthorne effects that a national rollout cannot replicate. Governments implement at scale with existing bureaucratic capacity, variable local conditions, and without the research team's intensive monitoring. The causal estimate from the trial may be correct for that context, but the magnitude of impact often shrinks substantially when programs scale.

Question 5

What problem does random assignment solve that makes RCTs the gold standard for causal inference in development economics?

Model answer: Random assignment solves the fundamental problem of selection bias — the fact that, in observational data, who receives a program is not random. Villages that receive aid may be wealthier, better-governed, or more motivated, so comparing treated and untreated villages confounds the program's effect with pre-existing differences. By randomly assigning treatment, RCTs ensure that treatment and control groups are statistically equivalent in expectation on all characteristics, so any observed difference in outcomes can be attributed causally to the program.

The deeper point is that random assignment balances not just the confounders researchers can measure (which regression or matching can also address) but also unobserved confounders — motivation, governance quality, historical context — that can never be fully measured. This is what makes the causal claim credible in a way that observational analysis cannot match, and why Banerjee and Duflo's RCT approach transformed the evidentiary standards of development economics.