Randomized Controlled Trials (RCTs)

Causality

Medium
RCTs estimate causal effects by randomly assigning units to treatment and control conditions. Their strength is that randomization helps break the link between treatment assignment and many confounding factors, making causal interpretation much cleaner.
Reasoning type
Experimental causal
Certainty level
High if well executed
Cognitive load
Medium
Formality
High

Core Idea

Definition

A randomized controlled trial infers causal effects by comparing outcomes across groups whose treatment assignment was determined by randomization.

In Plain English

If similar people or units are randomly split into groups, differences in outcome are more credibly attributed to the intervention.

Framework Structure

Components

Treatment Group
Control Group
Random Assignment
Outcome Comparison

Flow

Randomly assign units -> Apply treatment to one group -> Measure outcomes -> Compare treated and control groups

How to Apply

  • 1.Define the intervention and the outcome clearly
  • 2.Randomly assign units to treatment and control conditions
  • 3.Keep other conditions as comparable as possible
  • 4.Measure outcomes consistently across groups
  • 5.Interpret the difference in outcomes as evidence about the causal effect, while checking implementation quality

When to Use

  • Testing interventions where random assignment is feasible and ethical
  • Product experiments and A/B tests
  • Medical, educational, or policy evaluation
  • High-stakes causal questions where observational evidence is weak
  • Separating real effects from confounded correlations

When NOT to Use

  • When randomization is unethical or infeasible
  • When compliance, spillovers, or attrition undermine the design badly
  • When the trial environment becomes too artificial to generalize
  • When the question is exploratory and not yet ready for formal intervention testing

Example

Problem

A product team wants to know whether a new onboarding sequence increases activation.

Application

  • 1.Randomly assign new users to the old or new onboarding flow
  • 2.Ensure assignment is implemented reliably
  • 3.Measure activation and downstream quality metrics consistently
  • 4.Compare outcomes between groups to estimate the intervention's causal effect

Conclusion

Because assignment was randomized, the team can interpret differences more credibly as caused by the onboarding change.

Takeaway

RCTs are strongest when the design, execution, and interpretation are all handled with discipline.

Common Mistakes

  • Assuming randomization guarantees perfect execution
  • Ignoring attrition or noncompliance that reintroduce bias
  • Treating one successful trial as universally generalizable
  • Measuring only the easiest outcome while missing more important ones
  • Overlooking interference between treatment and control groups

How to Practice

assignment audit

Whenever reviewing an experiment, ask how treatment assignment actually happened and whether it was preserved.

outcome completeness check

Look for attrition, missingness, or compliance issues before trusting the comparison.

external validity note

After seeing the result, ask where else the finding should or should not be expected to hold.

Related Cognitive Biases

selection bias

Randomization is designed to reduce distortions caused by who receives the intervention.

confirmation bias

Even with strong design, people may still interpret ambiguous results in self-serving ways.

survivorship bias

Attrition or missing outcomes can quietly bias the comparison if ignored.

Related Frameworks

Related Skills

evaluating reliability
fact inference separation
probabilistic reasoning
comparing evidence

Variants & Extensions

A/B testing
Field experiments
Clinical trials
Cluster randomization

Typical Failure Modes

  • Noncompliance
  • Attrition bias
  • Weak generalization

Further Reading

  • Trustworthy Online Controlled Experiments by Ron Kohavi, Diane Tang, and Ya Xu
  • The Book of Why by Judea Pearl and Dana Mackenzie
  • Causal Inference in Statistics by Judea Pearl, Madelyn Glymour, and Nicholas P. Jewell