Can Participants Be Recruited After Clusters Are Randomized?

Can You Randomize Schools or Clinics Before Recruiting Their Participants?

Suppose you are evaluating an AI-supported teaching intervention across 30 schools. The schools agree to participate, and you randomly assign 15 to implement the intervention and 15 to continue conventional instruction. Only afterward do you begin identifying eligible students, obtaining consent, and enrolling participants for outcome assessment.

Is the trial still randomized? Yes, at the school level. But another question arises: could knowing which schools received the intervention influence which students ultimately enter the study?

Teachers in intervention schools might enthusiastically encourage participation, while those in comparison schools may recruit less actively. Students interested in artificial intelligence might also be more willing to enroll when they know their school will receive the new program.

The timing of recruitment matters because randomization protects the allocation of clusters, but it does not automatically protect the composition of participant samples assembled after that allocation becomes known.

Post-Randomization Recruitment Is Possible, but It Requires Safeguards

How Recruitment Timing Affects Cluster-Randomized Trials

Why cluster randomization often precedes individual recruitment

In individually randomized trials, researchers commonly establish eligibility and obtain consent before assigning participants to intervention conditions. Cluster-randomized trials may follow a different sequence because entire institutions, groups, or communities are allocated.

For example, a hospital may need to adopt a clinical protocol before researchers can identify patients who subsequently receive care. A school may implement an instructional policy at the beginning of an academic year, while eligible students enter classes throughout the year.

Some interventions are delivered continuously at the organizational level, making it impractical to recruit every future participant before allocation.

Post-randomization recruitment is therefore not inherently a protocol violation. Its acceptability depends on whether the recruitment process preserves a credible comparison between intervention and control conditions.

Distinguish cluster recruitment from participant recruitment

Cluster recruitment involves identifying and enrolling the schools, clinics, workplaces, or other groups that will be randomized. Participant recruitment concerns the individuals whose outcomes will be measured or who will participate in research procedures.

These activities may occur at different times.

The timing of cluster allocation should be clearly distinguished from the timing of participant identification, consent, intervention exposure, and outcome measurement.

How knowledge of allocation can introduce selection bias

Selection bias may arise when the probability of entering the analyzed participant sample depends on cluster allocation and characteristics related to the outcome.

Suppose teachers in intervention schools preferentially encourage high-performing students to participate, while comparison schools recruit all available students. The intervention arm may subsequently contain students with higher baseline achievement.

The schools were randomly allocated, but the participating students were selected through different processes after allocation.

This can undermine comparability at the participant level. Randomization of schools does not guarantee exchangeability among selectively recruited individuals.

The risk is particularly consequential when recruitment decisions are discretionary and recruiters are aware of treatment assignment.

Eligibility assessment should be independent of assigned treatment

Eligibility criteria should ordinarily be established before randomization and applied consistently across trial arms.

For example, a school-based trial might include all students enrolled in a specified grade at the beginning of the academic year, subject to clearly defined exclusions.

Researchers should avoid allowing intervention schools to enroll students according to motivation or anticipated benefit while comparison schools follow different criteria.

Objective administrative records can help establish an eligible population independently of recruiters' preferences. However, the records themselves must be sufficiently complete and comparable across clusters.

Eligibility definitions should also specify how students who transfer, enroll late, or leave during follow-up will be handled.

Identifying participants before randomization can be more important than obtaining every consent beforehand

Participant identification and informed consent are related but distinct procedures.

In some trials, researchers can establish a complete roster of eligible individuals before cluster randomization, even if consent for specific data collection activities occurs later.

For example, schools may provide an approved roster of students meeting prespecified eligibility criteria before allocation. Researchers can then approach those students or their guardians using standardized consent procedures.

This arrangement can reduce opportunities to selectively identify eligible participants after treatment assignment becomes known.

Nevertheless, differential consent after randomization can still affect the analyzed sample. Researchers should therefore examine both the eligibility process and the subsequent participation process.

Access to rosters and identifiable records must comply with applicable privacy, consent, and ethics requirements.

Masking recruiters may reduce allocation-related selection

Where practical, recruiters should be unaware of cluster allocation while determining eligibility and approaching participants.

Masking may be feasible when recruitment is conducted centrally using administrative records or when the intervention has not yet become visible.

However, masking can be difficult in educational or organizational interventions. Teachers and administrators may know whether their institution has received training, equipment, or new instructional procedures.

When masking is impossible, researchers should strengthen objective eligibility rules, centralized recruitment oversight, standardized invitations, and documentation of recruitment decisions.

Masking is a safeguard, not a substitute for transparent recruitment procedures.

Standardized recruitment procedures can reduce discretionary selection

Researchers should specify who will approach participants, when invitations will be distributed, what information will be provided, and how eligibility and consent decisions will be recorded.

Recruitment materials should be comparable across conditions, while accurately describing the intervention and any differences relevant to informed consent.

For example, all eligible students might receive invitations through the same centralized process rather than allowing individual teachers to select students they believe would benefit most.

Recruitment targets should not encourage selective enrollment of participants with favorable expected outcomes.

Where appropriate, independent monitoring can identify differences in recruitment rates or eligibility decisions across trial arms.

Consent requirements do not disappear because clusters are randomized

Cluster-level authorization and individual informed consent address different ethical questions.

A school administrator may authorize institutional participation or implementation of a school-wide program, but that authorization does not automatically replace consent required for individual research participation or identifiable data collection.

Some cluster-level interventions cannot realistically be accepted or refused individually. Nevertheless, consent for surveys, interviews, additional assessments, or identifiable data use may still be required, depending on the research and applicable ethical framework.

Researchers should distinguish consent to intervention exposure from consent to data collection and research participation.

Requirements vary across jurisdictions, institutions, intervention types, and ethics review decisions. They should be established before recruitment begins.

Post-randomization recruitment may be unavoidable in open-cohort trials

Some cluster-randomized trials enroll participants continuously. Patients may enter a clinic after the clinic has been randomized, or students may enroll in a school during an ongoing intervention.

In these open-cohort settings, researchers cannot identify every future participant before allocation.

The appropriate response is to define eligibility prospectively, specify when individuals enter the study population, and apply consistent recruitment and outcome ascertainment procedures.

Researchers should consider whether intervention assignment affects who enters or remains in the cluster. For example, a highly attractive program might influence enrollment decisions, potentially changing the population observed under each condition.

Such changes may be part of the intervention's broader effect or may complicate estimation of a particular participant-level causal effect. The interpretation depends on the estimand.

Recruitment bias and contamination are different problems

Contamination occurs when intervention exposure spreads across assigned conditions. Recruitment bias occurs when the process determining who enters the study differs in a way that compromises comparability.

Randomizing larger clusters may reduce some forms of intervention contamination, but it does not necessarily reduce recruitment bias.

Indeed, cluster-level allocation may make treatment assignment visible to recruiters before individual enrollment begins.

Researchers should therefore assess contamination and recruitment selection as separate design threats rather than assuming that solving one automatically addresses the other.

Intention-to-treat analysis does not automatically correct selective recruitment

Intention-to-treat analysis generally preserves participants or clusters in their assigned conditions regardless of subsequent adherence or intervention exposure.

However, if the participant sample itself was selected differently after randomization, analyzing recruited participants according to their assigned cluster does not necessarily remove selection bias.

The problem arises before the analysis: the individuals whose outcomes are observed may no longer represent comparable populations across treatment arms.

Statistical adjustment for measured baseline characteristics may help in some circumstances, but it cannot guarantee correction for unmeasured selection mechanisms.

Preventing allocation-related recruitment bias through design is generally preferable to attempting to repair it analytically.

Recruitment and participant flow must be reported transparently

The CONSORT extension for cluster-randomized trials emphasizes reporting participant and cluster flow, including recruitment and allocation information.

Researchers should distinguish the number of clusters randomized, individuals assessed for eligibility, individuals recruited or consenting, and participants included in each analysis.

Where feasible, these counts should be presented separately by trial arm, with reasons for exclusions or nonparticipation.

Important baseline characteristics should also be examined at both cluster and participant levels.

Transparent reporting helps readers evaluate whether knowledge of cluster allocation influenced recruitment.

Recruiting Students After Schools Have Been Randomized

Interpretation: Recruitment after randomization is not automatically unacceptable. The procedures reduce opportunities for discretionary selection and provide information needed to evaluate recruitment differences.

Nevertheless, if students in intervention schools are more likely to consent because they know the assigned program, differential participation may remain. The researchers must assess its implications for the intended estimand and acknowledge any unresolved selection concerns.

Common Misunderstandings About Post-Randomization Recruitment

Design Recruitment to Minimize the Influence of Cluster Allocation

Before randomization, determine whether eligible individuals can be identified, recruited, or consented in advance. If not, specify which recruitment activities must occur afterward and why.

Use objective eligibility criteria, consistent recruitment procedures, and independent recruitment personnel where feasible. Plan to document the complete participant flow and evaluate differences between trial arms.

The objective is not necessarily to complete every recruitment activity before randomization. It is to prevent knowledge of the assigned condition from systematically determining which participants contribute to the comparison.

Before Recruiting Participants After Cluster Allocation

Frequently Asked Questions About Recruitment After Cluster Randomization

Recruitment Timing Is Acceptable Only When Selection Risks Are Addressed

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.