5.3 Randomized Control Trials (RCTs)

Randomized control trials are a type of impact evaluation.  Characterized by a random assignment of units, RCTs document the impact of a program or intervention by comparing what happened for young people versus what would have happened without the program (also known of as the “counterfactual.”) RCTs seek to control for a number of external factors that might influence results 1.RCTs contribute to the field by:

  • Taking a scientific approach to quantifying impact
  • Answering specific and nuanced questions about issues relevant to a given program
  • Proving attribution in a more direct way
  • Producing results perceived to be less biased

RCTs are frequently considered the “gold standard” of impact evaluation for the reasons noted above. That does not mean, however, that every program or intervention should be evaluated via an RCT. The following section describes the how and why of RCTs in an effort to help programs think through when and where this would be an appropriate methodology.

RCTs represent the highest level of scientific rigor in impact evaluations. Given the unique requirements of rigorous evaluations, organizations need to consider several key points related to implementing an RCT. Also, since RCTs have to be planned prior to the intervention (in order to establish treatment and control groups), they should be considered within a larger program design planned during project design.

Checklist: Innovations for Poverty Action on Deciding Whether to Randomize

Innovations for Poverty Action (IPA), a nonprofit dedicated to discovering what works to help the world’s poor, evaluates programs around the world. They offer the following ideas to determine if randomization might be appropriate for a particular program. Consider if your program has the following:

  • An important, specific and testable question?
  • Timing—not too early or too late?
  • A representative program rather than gold-plated?
  • Time, expertise, and money to do it right?
  • Capacity and will to use results to inform programming and policy?