03 · What You Need to Know
How Recruitment Units and Analytical Units Relate
What Is the Unit of Recruitment?
The unit of recruitment is the entity approached, invited, or enrolled at a particular stage of the recruitment process.
Depending on the design, this may be an individual, household, classroom, school, hospital, organization, or community.
For example, researchers may first obtain institutional participation from universities and then invite eligible students within those universities to participate.
In this situation, recruitment occurs at more than one level. Universities are approached as participating institutions, while students are individually invited or enrolled.
Researchers should specify which recruitment stage they mean rather than assuming that a multistage recruitment process has only one recruitment unit.
Institutional agreement to participate does not necessarily establish that every individual within the institution has been recruited or has consented to participate.
What Is the Unit of Analysis?
The unit of analysis is the entity or level about which the study makes its principal analytical claims.
If the question concerns whether students' motivation is associated with academic performance, students may be the analytical units.
If the question concerns whether universities with different governance arrangements have different institutional research outputs, universities may be the analytical units.
The unit should be determined by the intended estimand rather than the administrative arrangements used to obtain access to participants.
Understanding what constitutes the unit of analysis is therefore essential before determining how recruitment should proceed.
How Do Recruitment, Sampling, Observation, and Analysis Differ?
These concepts describe related but distinct functions within a research design.
| Unit |
Principal function |
Illustrative educational study |
| Sampling unit |
Entity selected at a stage of the sampling procedure |
A school selected from a sampling frame |
| Recruitment unit |
Entity approached or enrolled at a recruitment stage |
A school invited to participate, followed by student invitations |
| Observation unit |
Entity or occurrence for which data are recorded |
An individual student's questionnaire response |
| Unit of analysis |
Entity or level addressed by the analytical claim |
A student whose engagement and achievement are analyzed |
| Unit of randomization |
Entity independently assigned to an experimental condition |
A classroom assigned to an instructional intervention |
These units may coincide, but there is no general requirement that they must.
Problems arise when researchers fail to recognize their differences and consequently misrepresent the population, independent observations, or treatment assignment structure.
When Is Recruiting Groups to Study Individuals Appropriate?
Group-based recruitment is often appropriate when individuals are naturally organized within institutions or when access is most efficiently obtained through those institutions.
Examples include recruiting schools before inviting students, hospitals before identifying patients, and companies before surveying employees.
This approach can reduce logistical costs and facilitate access to a defined population.
However, the participating groups may not represent the broader population of groups. Schools willing to participate in an educational technology study, for example, may have stronger digital infrastructure than schools that decline.
Researchers should distinguish operational convenience from population representativeness.
Recruiting individuals within selected groups can support valid individual-level research, but the selection of those groups influences who can enter the study.
How Is This Related to Multistage Sampling?
Multistage sampling selects units at successive levels.
For example, a researcher may select districts, then schools within districts, and finally students within schools.
Each stage has its own selection procedure and potentially different inclusion probabilities.
When the objective is population estimation, sampling weights may be necessary to reflect unequal selection probabilities and nonresponse adjustments.
The sampling design may also require variance estimation that accounts for stratification and clustering.
Importantly, recruiting participants through multiple stages does not automatically constitute probability sampling. Probability-based inference requires appropriate random selection procedures and known or estimable inclusion probabilities.
Conveniently selecting schools and then surveying students within them remains a different sampling design from randomly selecting schools from a complete frame.
Can the Recruitment Unit Be Larger Than the Analytical Unit?
Yes. This is common in institutional and household research.
For example, a hospital may be recruited to participate in a study of patient outcomes. The hospital provides access to records, but patients remain the principal analytical units.
Similarly, a household may be recruited as an entry point for collecting information about individual family members.
The analysis may concern individuals, while the recruitment process introduces household-level clustering and selection.
Researchers should specify whether every eligible individual within a recruited group is included or whether an additional selection stage determines individual participation.
These arrangements influence both the composition of the analytic sample and the statistical dependence among observations.
Can the Recruitment Unit Be Smaller Than the Analytical Unit?
Yes. Researchers may recruit individuals as informants while analyzing characteristics of larger groups.
For example, employees may be recruited individually to provide information about organizational climate. Their responses may subsequently be aggregated to organizational measures.
In this case, the recruitment process involves individuals, but the research question concerns organizations.
However, individual responses must adequately represent the intended group construct.
Researchers should examine response coverage, measurement reliability, and the conceptual justification for aggregation.
The principles of using individual-level data for group-level analysis are especially relevant.
How Can Institutional Recruitment Introduce Selection Bias?
Institutions may differ in their willingness or ability to participate.
For example, universities with established digital learning programs may be more willing to join a study of educational technology adoption.
If those universities also differ in student achievement, institutional resources, or teaching practices, the recruited sample may systematically differ from the intended population.
Selection can also occur within institutions. Administrators may invite only particular classes, teachers may encourage some students more strongly than others, or individuals with certain experiences may be more likely to respond.
Researchers should therefore evaluate selection at every relevant recruitment stage.
Correctly accounting for clustering does not eliminate selection bias. These are distinct methodological concerns.
What If Institutional Gatekeepers Control Recruitment?
Schools, hospitals, and organizations often require permission before researchers can approach members or access records.
Such institutional representatives are sometimes called gatekeepers.
Gatekeeper approval may be necessary for access, but it should not automatically be treated as a substitute for individual consent when individual consent is required.
Gatekeepers can also influence the recruitment process by determining which groups receive invitations, when invitations are distributed, or how participation is presented.
These decisions may affect recruitment equity and participant autonomy.
Researchers should document gatekeeper involvement, minimize unnecessary restrictions on eligible participants, and follow applicable ethics approval and consent requirements.
How Does Group-Based Recruitment Create Clustering?
Individuals recruited from the same group may share experiences, environments, and unmeasured characteristics.
For example, students from the same school may have similar educational resources, while patients from the same hospital may receive similar care protocols.
These shared influences can produce intracluster correlation.
If the analysis assumes independence without considering this structure, uncertainty estimates may be inappropriate.
Researchers should assess when clustering must be accounted for in research design.
Potential approaches include multilevel models, cluster-robust standard errors, generalized estimating equations, or design-based variance estimation, depending on the sampling design and estimand.
Does Recruiting More Individuals Compensate for Recruiting Few Groups?
Not necessarily. The value of additional individuals depends on the research question and the degree of dependence within groups.
Suppose 1,000 students are recruited from only two schools. The dataset may contain substantial individual-level information, but it provides limited independent evidence about differences between schools.
If the research question concerns school-level policies, the two schools cannot be treated as 1,000 independent institutional observations.
Similarly, if treatment is assigned at the school level, additional students do not create more independently assigned schools.
Researchers should distinguish participant counts from cluster counts in sample size planning.
How Does Recruitment Affect Generalizability?
The population represented by the study depends partly on which groups were recruited and which individuals participated within them.
For example, a study recruiting students from private urban universities may not adequately represent students attending rural public institutions.
Even if the student sample is large, institutional selection can limit external validity.
Researchers should distinguish the target population from the accessible population and the final analytic sample.
Where probability sampling is used, appropriate weights and design-based estimation may support population inference under the relevant assumptions.
Where groups are selected purposively or conveniently, researchers should justify the selection and limit generalizations accordingly.
What If Recruitment Occurs After Treatment Assignment?
This situation deserves particular attention in cluster-randomized trials.
Suppose schools are randomized to intervention and comparison conditions before individual students are identified or invited.
If recruiters know the assigned condition, recruitment decisions or participant willingness may differ between treatment groups.
Such differences can introduce selection bias, even when the original school-level randomization was valid.
Potential safeguards include identifying eligible participants before randomization, using standardized recruitment procedures, and masking recruiters to allocation where feasible.
Researchers should distinguish recruitment procedures from the units of randomization and measurement.
Can the Unit of Recruitment Change During a Study?
Recruitment procedures may evolve as a project expands to additional institutions or adopts new enrollment channels.
For example, a study may initially recruit students directly and later recruit through participating universities.
Such changes can alter who is reached, the probability of participation, and the dependence structure of the resulting sample.
Researchers should document changes and evaluate whether they affect comparability between recruitment periods.
When substantial changes occur, sensitivity analyses or separate reporting of recruitment cohorts may be appropriate.
Changes should not be concealed merely because the final outcome measures remain identical.
How Should Recruitment and Analysis Units Be Reported?
Methods sections should describe how groups and individuals were approached, selected, and enrolled.
Report the number of eligible and participating groups, the number of individuals recruited within groups, and the final analytic sample.
When possible, provide participation rates at each stage and explain relevant exclusions or nonresponse.
Describe the clustering structure and how it was addressed statistically.
For observational research, STROBE provides reporting guidance concerning participant selection, study size, statistical methods, and potential bias.
For cluster-randomized trials, the CONSORT extension provides additional recommendations for reporting recruitment and allocation at different levels.
