A program delivered to whole schools
Imagine a fictional study of a school-wide scheduling change. Assigning individual students to different school schedules might be impractical. Researchers instead assign ten schools to the new schedule and ten to the comparison condition.
The trial includes many students, but the assignment occurred at twenty schools. Students at one school share staff, environment, and implementation. Their outcomes may therefore resemble one another more than outcomes from students in different schools.
Why the total headcount is incomplete
Adding many students to a small number of schools does not provide the same information as adding more independently assigned schools. How much clustering matters depends on within-school similarity, group sizes, and the design.
An analysis that ignores this structure can overstate precision. Cochrane's guidance on trial variants explains why cluster designs require appropriate methods.
Read the participant flow twice
- At the cluster level: how many groups were randomized, lost, and analyzed?
- At the individual level: how many people were eligible, recruited, measured, and analyzed within those groups?
Recruiting individuals after the cluster assignment becomes known can introduce additional concerns if recruitment differs by group. The timing belongs in the methods.
What the result applies to
A successful school-level intervention may depend on leadership, staffing, or other shared conditions. That does not reduce its value; it identifies the unit at which the program operates.
When reading a headline about “1,000 participants,” look for the number of clusters as well. The design's strength is not captured by the largest number in the abstract. Understanding who was randomized helps explain both the practical reason for the trial and the uncertainty around its result.
