RigorMed
← Research resources

Education research

How to design an education study randomized by classroom

When whole classes receive a teaching activity, the class is the unit of assignment. Student scores can still be collected individually. The analysis and sample planning need to account for within-class correlation.

Education research

Gather class lists, allocation procedures, the activity schedule, outcome definitions, and baseline measures. Specify the study population and primary comparison before the activity begins.

Map schools, teachers, classes, and students. If a teacher appears in one arm only, explain how the design distinguishes teacher effects from the activity.

How to do it

  1. State what is assigned

    Specify whether schools, classes, or students enter allocation. Record stratification, who assigns groups, and when assignment becomes known. Describe the allocation procedure as it was actually performed.

  2. Define the primary outcome and time point

    Record the instrument, score range, assessment window, makeup policy, and scoring process. Keep baseline and follow-up fields separate so differences and attrition can be reviewed.

  3. Match analysis to the allocation level

    Agree on an analysis appropriate to the number of classes and study design. Describe how it handles within-class correlation and when adjustment variables were measured. Assess inferential uncertainty with few clusters.

  4. Plan sample size and student flow

    Record class and student counts. Explain assumptions about intraclass correlation, class size, and attrition. Report counts assigned, exposed, assessed, and analyzed.

Worked example

This synthetic study plans to recruit 12 classes with about 25 students each and randomize whole classes between two review activities.

Worked example
FieldPlanned entryReview point
Assignment6 classes versus 6 classesRecord stratification and randomization
OutcomeStudent delayed recall score on day 7Log absences, makeup sessions, and versions
AnalysisCompare assigned groups and account for class clusteringReport 12 classes and about 300 students separately
Missing scoresLog reasons by arm and classAn unobserved assessment is not a zero score

The numbers illustrate a protocol structure. A real study also requires appropriate approvals, data rights, and a design-specific analysis.

Common questions

Can I analyze 300 independent observations if there is one row per student?

A row is a measurement record. Classroom assignment calls for attention to clustering and the number of classes.

What if there is only one class per arm?

Other class or teacher differences may be confounded with the activity. The design offers limited evidence and is better suited to exploratory or pilot use.

Verified sources

The official sources below were checked on 2026-11-13. Access conditions and update dates are rechecked before publication.

  1. IES What Works Clearinghouse — Procedures and Standards

RigorMed workspace

Review my classroom study protocol

Bring an authorized teaching schedule, assignment unit, and outcome definitions. Draft a protocol, participant-flow table, and questions for analyst review.