Randomized Controlled Trial (RCT)

A randomized controlled trial (RCT), also spelled randomised controlled trial, is a research design in which participants are assigned by chance to one of two or more groups. One group, the treatment group, receives an intervention, such as a cash transfer, a loan offer, a training course or a new medicine; the other, the control group, does not, or receives a standard alternative. Because the assignment is random, the groups are alike on average in every respect, observed or not, so any later difference in outcomes can be credited to the intervention.

RCTs began in medicine and agriculture and became widespread in economics from the late 1990s, especially in development economics. Abhijit Banerjee, Esther Duflo and Michael Kremer shared the 2019 Nobel Memorial Prize in Economic Sciences for their experimental approach to reducing global poverty. RCTs are often called the "gold standard" of evidence, while natural experiments and other methods of econometrics are used where randomisation is impossible.

Suppose a school wants to know whether a new reading app helps children. If it simply gives the app to the keenest pupils, they would probably have done better anyway, so the test would prove nothing. Instead, the school puts every pupil's name in a hat and draws half of them to get the app.

Because the hat does not care who is keen, clever or well supported at home, both halves end up similar. If the app group reads better at the end of the year, the app is the most likely reason. That coin-toss fairness is the heart of an RCT.

1. Define the question and the outcome, such as household income, school attendance or loan repayment, and register the plan in advance.
2. Recruit a sample large enough to detect a realistic effect.
3. Randomly assign participants, or whole villages, schools or bank branches, to treatment and control groups.
4. Deliver the intervention and track outcomes for both groups over time.
5. Compare average outcomes and test whether the difference is larger than chance would explain.

The basic estimate is the average treatment effect:

• $\bar{Y}_{\text{treatment}}$ is the average outcome in the treatment group.
• $\bar{Y}_{\text{control}}$ is the average outcome in the control group.

For example, if households offered a savings account end the year with average savings of 260 dollars and the control group with 200 dollars, the estimated effect is $260 - 200 = 60$ dollars. Standard errors, often based on the normal distribution, show how precise that figure is.

Well-known economic RCTs include:

• Conditional cash transfers. Mexico's PROGRESA programme, launched in 1997, was rolled out to randomly chosen villages first, allowing researchers to measure its effects on school enrolment and health; the design influenced similar welfare schemes across Latin America and beyond.
• Deworming. Work by Edward Miguel and Michael Kremer in Kenya studied school-based deworming and its effects on attendance; its long-run findings remain debated.
• Microcredit. A set of RCTs in six countries, published together in 2015, found that access to microloans had modest average effects on household income, tempering earlier claims about microfinance and financial inclusion.
• Unconditional cash. Trials of direct cash transfers in Kenya and elsewhere have informed debates on universal basic income.
• Behavioural design. Governments test letters, defaults and reminders, drawing on behavioural economics, to raise tax compliance, pension saving or take-up of benefits.

The Abdul Latif Jameel Poverty Action Lab (J-PAL), founded at MIT in 2003, and Innovations for Poverty Action coordinate many such trials in developing countries. Firms also run large numbers of online A/B tests, which are RCTs for prices, products and web design.

RCTs have attracted serious criticism from economists including Angus Deaton, the 2015 Nobel laureate, and the philosopher Nancy Cartwright:

• External validity. A result from one region, time or delivery organisation may not hold elsewhere or when a programme is scaled up nationally.
• Narrow scope. Many big questions in macroeconomics, such as the effects of monetary policy, fiscal policy or trade, cannot be randomised.
• Average effects. An average can hide large gains for some and losses for others.
• Ethics. Withholding a likely beneficial programme from a control group, or experimenting on poor communities, raises ethical questions.
• General equilibrium effects. Giving cash to a few households in a village may not show what happens to local prices and wages if everyone gets it.

Supporters reply that RCTs have overturned popular beliefs, such as the large claimed benefits of microcredit, and that careful replication across countries addresses much of the external-validity concern. Most economists now see RCTs as one important source of evidence alongside natural experiments, observational data and economic models, rather than a replacement for them. Debates about how far economics should rely on experiments rather than formal theory form part of a wider discussion sometimes framed as physics envy.

• Nobel Prize: The Prize in Economic Sciences 2019
• Abdul Latif Jameel Poverty Action Lab (J-PAL)