03 · What You Need to Know
The Number of Sites Matters Only in Relation to What the Study Is Trying to Establish
What Counts as a Research Site?
A research site is a setting in which participants, interventions, observations, records, or other study activities are located. Depending on the discipline, sites may be universities, schools, hospitals, clinics, communities, companies, laboratories, regions, online environments, or other organizational or geographical settings.
A single-site study conducts the relevant research within one such setting. A multisite study includes two or more distinct settings. In clinical research, the term multicenter is commonly used when multiple hospitals, clinics, or research centers participate.
Yet counting sites is only the beginning. Two departments within the same university might function as distinct analytical contexts for one research question but not another. Conversely, several physical locations belonging to one standardized organization may operate so similarly that treating them as substantively distinct sites adds little analytical variation.
Define a site according to the structure relevant to the study rather than merely counting addresses.
A Single-Site Study Is Not Automatically a Weak Study
Single-site research can be entirely appropriate when the question concerns a particular setting, when the phenomenon is locally bounded, when the site provides sufficient participants and variation, or when intensive investigation is more important than breadth across settings.
Suppose a university introduces a distinctive institutional policy and the research question concerns how that specific implementation unfolded. Adding unrelated universities may change rather than improve the question.
Similarly, early-stage feasibility work may reasonably begin at one site before researchers invest in a larger multisite project. A well-executed single-site study with clear boundaries and appropriately qualified conclusions can be more informative than a poorly coordinated multisite study.
Single Site Does Not Mean Single Case
A research site is where the study occurs. A case is the bounded unit around which a case study inquiry is organized. Those concepts can coincide, but they do not have to.
For example, researchers might survey 1,500 students at one university to examine relationships among learning behaviors. The university is the study site, but it is not necessarily being investigated as a case. Conversely, researchers might conduct an in-depth investigation of that university's implementation of a new curriculum, in which the institution or implementation becomes the bounded case.
Before treating one setting as methodological justification for case study, determine whether the setting is actually the bounded case or merely where data collection happens.
What Can Additional Sites Actually Add?
Multisite research can serve several different purposes, and those purposes should be distinguished.
First, additional sites can increase access to participants. A rare population or outcome may be difficult to study within one institution, while several settings can provide enough eligible participants within a reasonable period.
Second, sites can provide contextual heterogeneity. Researchers may intentionally include urban and rural schools, public and private institutions, large and small hospitals, or settings serving different populations because those differences matter to the research question.
Third, multisite designs can examine whether an association, intervention effect, implementation process, or other finding appears consistently across settings.
Fourth, sites themselves may become an analytical level. Researchers can investigate whether contextual characteristics explain differences in outcomes.
These are distinct rationales. “We wanted a bigger sample” is not the same methodological objective as “we wanted to test whether the effect varies across institutional contexts.”
More Sites Do Not Automatically Mean Greater Generalizability
It is tempting to assume that one site is local and ten sites are generalizable. The inference is not that simple.
If ten universities are selected because collaborators were conveniently available, they may still represent a narrow subset of universities. Increasing the number of convenience sites does not transform them into a probability sample of institutions.
Generalizability depends on the target population, site-selection process, participant sampling, study eligibility criteria, contextual variation, measurement, and the relationship between the studied settings and the settings to which researchers wish to apply the findings.
Watch Out
Do not equate the number of sites with representativeness. A multisite study can broaden the range of observed contexts without statistically representing every setting in the target population.
Site Selection Should Follow the Research Purpose
If the objective is primarily recruitment, researchers may select sites that provide access to the required population while maintaining necessary study conditions.
If the objective is to understand contextual variation, site selection should deliberately capture meaningful contrasts. For example, researchers studying technology implementation might select institutions with different levels of infrastructure, governance, or prior digital experience.
If researchers want to estimate population-level effects across a defined universe of sites, the sampling strategy becomes more demanding. The selection of settings should support the intended inference rather than relying exclusively on accessibility.
This is analogous to participant sampling: more units are not automatically more representative if the mechanism by which they entered the study remains narrow.
Multisite Research Introduces Site-Level Variation
Once several sites participate, researchers must consider whether observations within the same site are more similar to one another than observations from different sites.
Students within one university share policies, infrastructure, instructors, curricula, and institutional culture. Patients within one hospital share clinical systems and practices. Employees within one company share organizational structures. These common conditions can create clustering.
When clustering is relevant, treating every individual observation as completely independent may produce inappropriate standard errors and misleading precision. Depending on the design and number of sites, researchers may need cluster-aware statistical approaches such as multilevel models, generalized estimating equations, cluster-robust methods, or design-specific analyses.
The appropriate method depends on the research design, sampling structure, number and size of clusters, outcome, and inferential objective. Merely adding “site” as another variable does not solve every form of clustering.
A Larger Total Sample Does Not Necessarily Mean Enough Information About Site Differences
Suppose a study includes 5,000 students from only three universities. The individual-level sample is large, but the site-level sample is still three.
If the research question concerns student-level associations, that dataset may provide substantial information. If the question concerns why universities differ or whether an effect varies systematically according to institutional characteristics, three sites provide much less information at the institutional level.
This distinction between the number of individuals and the number of higher-level units is essential in clustered and multilevel research. Thousands of participants cannot magically turn three institutions into thirty.
Standardization Becomes More Difficult Across Sites
A single research team at one site can often maintain relatively close control over recruitment, measurement, intervention delivery, data entry, and follow-up. Multisite research distributes those activities across different teams and contexts.
Sites may interpret eligibility criteria differently. Equipment may vary. Staff may receive different training. Recruitment channels may reach different populations. An intervention may be implemented with different levels of fidelity. Data systems may encode the same concept differently.
Multisite protocols therefore often require detailed operating procedures, common definitions, training, data dictionaries, quality-control systems, communication procedures, and monitoring appropriate to the study.
Standardization should not mean pretending that contexts are identical. Researchers need to distinguish unwanted procedural inconsistency from meaningful contextual variation that the study should preserve and measure.
Multisite Intervention Research Must Consider Implementation Differences
If an intervention is delivered at several sites, the treatment may not function identically everywhere. Differences in staffing, infrastructure, leadership, participant characteristics, organizational readiness, or fidelity may influence outcomes.
Those differences can initially look like methodological noise. Sometimes they are substantively important. If an intervention succeeds in one setting and fails in another, understanding why may be more useful than reporting only an average effect across all sites.
Researchers should therefore decide in advance whether site variation is something to minimize, adjust for, stratify by, model, or investigate as part of the research question.
More Sites Create More Governance and Coordination
Multisite research may involve separate institutional approvals, data-sharing arrangements, contracts, recruitment processes, privacy requirements, local investigators, staff training, data-transfer procedures, and monitoring.
These requirements vary substantially by country, institution, discipline, and type of research. Researchers should verify the applicable ethics and governance requirements directly with the responsible institutions rather than assuming that approval at one site automatically covers another.
The methodological benefit of another site should therefore be weighed against the operational consequences of adding it.
Site Number and Measurement Timing Can Interact
Adding sites is not the only way to make a design more informative. Sometimes additional observations within the same sites provide more relevant evidence than expanding geographically.
For example, if the research question concerns how an institutional policy changes outcomes over time, repeated measurements before and after implementation may be more informative than adding several institutions measured only once. Conversely, if the question concerns whether the pattern differs across institutional contexts, additional sites may matter more than additional measurements within one site.
Thinking explicitly about what the timing of data collection contributes can prevent researchers from assuming that expansion across settings is the only route to a stronger design.
