03 · What You Need to Know
How to Conduct Rigorous Research When Eligible Participants Are Inherently Rare
What makes a research population genuinely rare?
A rare population consists of relatively few individuals or units satisfying a scientifically defined set of eligibility characteristics within a specified geographic, institutional, or temporal scope.
Examples may include people with uncommon medical conditions, professionals holding unusual combinations of qualifications, students participating in highly specialized programs, or communities with narrowly defined characteristics.
However, a population should not be described as inherently rare merely because the researcher has access to few participants.
For example, identifying only twelve teachers at one institution does not establish that teachers with the relevant characteristics are rare nationally.
Researchers should distinguish the size of the target population from the size of the accessible population and the achieved sample.
This distinction determines whether the appropriate response is to improve recruitment or reconsider the research design.
How is a rare population different from a hard-to-reach population?
A rare population is limited primarily by the number of eligible individuals who exist within the defined scope.
A hard-to-reach population may be considerably larger but difficult to identify, contact, or recruit because of geographic dispersion, privacy concerns, institutional barriers, stigma, or other access limitations.
The two conditions can occur together, but they require different feasibility assessments.
| Population situation |
Primary constraint |
Potential response |
| Genuinely rare population |
Few eligible individuals exist within the target population. |
Consider complete enumeration, alternative designs, or a revised estimand. |
| Hard-to-reach population |
Eligible individuals exist but are difficult to identify or contact. |
Improve recruitment frames, partnerships, and appropriate access strategies. |
| Rare and hard-to-reach population |
Both population size and accessibility constrain recruitment. |
Combine careful population definition with ethically appropriate recruitment and design adaptation. |
| Artificially narrow population |
Eligibility criteria exclude individuals who may be scientifically relevant. |
Reassess whether restrictions are necessary for the research question. |
Researchers should not broaden eligibility simply to increase the sample if doing so changes the intended population or undermines construct validity.
What should you do when the calculated sample size exceeds the eligible population?
First, examine what the sample size calculation was designed to achieve.
Was it intended to detect a particular effect, estimate a population proportion within a specified margin of error, or support a complex statistical model?
Different objectives produce different sample size requirements.
For example, a calculation for detecting a small correlation with high power may require substantially more participants than a descriptive study intended to characterize the available population.
Next, determine whether the assumptions are appropriate, including expected variability, effect size, sampling design, and the relevant population.
If the calculation is appropriate but the requirement remains impossible, the original inferential objective may be infeasible.
Researchers should then evaluate alternative designs or more realistic objectives rather than arbitrarily reducing the required sample and declaring the problem resolved.
Should you recruit the entire rare population?
Complete enumeration may be appropriate when the eligible population is small, identifiable, and accessible.
Recruiting everyone can maximize coverage and reduce the loss of information associated with sampling only a subset.
For example, if a national professional group contains 42 eligible specialists, inviting all 42 may be more sensible than randomly selecting 25.
However, complete recruitment does not guarantee complete participation.
Nonresponse may be particularly consequential when a few missing individuals represent a substantial proportion of the population.
Researchers should distinguish the eligible population, those invited, those participating, and those providing usable outcome information.
Complete recruitment also does not automatically establish adequate power for a causal comparison or justify generalization beyond the defined population.
See whether to recruit the entire eligible population for the relevant design considerations.
Can multisite recruitment overcome the sample size problem?
Multisite recruitment may help when eligible individuals are distributed across institutions or geographic areas.
For example, a rare educational population may contain only a few students at each university but a larger number across a national consortium.
However, additional sites must contribute participants who satisfy sufficiently consistent eligibility criteria.
Differences in institutional practices, participant characteristics, measurement instruments, and recruitment procedures can affect comparability.
Researchers should evaluate whether combining participants across sites supports the same target population and estimand.
Site clustering and heterogeneity may also affect statistical precision.
The decision should follow the principles of multisite recruitment for rare cases, while recognizing that recruiting a rare population is not necessarily identical to recruiting cases defined by a rare outcome.
Can registries identify additional eligible participants?
Registries may help locate individuals who satisfy uncommon eligibility criteria across institutions or extended periods.
For example, a condition-specific registry may contain records of patients who would be difficult to identify through ordinary clinic recruitment.
However, registry inclusion may depend on diagnosis, access to services, voluntary participation, or institutional reporting.
These processes can affect representativeness and selection bias.
Researchers should assess registry coverage, data quality, eligibility definitions, and access permissions before relying on the records.
A registry may improve feasibility without necessarily providing a valid comparison group or sufficient confounding information.
These issues are discussed further in using research registries for rare-outcome studies.
Should historical data be included?
Historical observations may increase the number of eligible participants or events available for analysis.
For example, a specialized program enrolling ten eligible students annually may have accumulated records across several previous cohorts.
However, earlier cohorts may differ in admission requirements, instructional practices, assessment instruments, or other relevant characteristics.
Combining historical and contemporary participants may therefore change the population represented by the analysis.
Researchers should examine eligibility, measurement equivalence, calendar-period effects, and missing data.
Historical information should be included because it supports the intended research question, not simply because it increases the reported sample size.
See when historical data can increase research sample size.
Can repeated measurements compensate for having few participants?
Repeated measurements may improve information about within-person change, trajectories, and responses to interventions.
For example, measuring the performance of fifteen participants weekly may reveal patterns that a single assessment cannot capture.
However, repeated observations are generally correlated within participants.
Fifteen individuals measured ten times do not constitute 150 independent individuals.
The usefulness of repeated measurement depends on the estimand and the expected temporal behavior of the outcome.
Repeated measures may improve precision for certain within-person comparisons while providing limited additional information about between-person variation or population heterogeneity.
The design should follow the principles of repeated measurement in small populations.
When are single-case experimental designs appropriate?
Single-case experimental designs may be valuable when the objective concerns an intervention effect on an outcome that can be measured repeatedly.
For example, researchers may evaluate whether a structured instructional support intervention improves independent task completion among a few students with specialized learning needs.
Repeated baseline and intervention observations can provide opportunities to examine functional relationships.
However, the design requires planned experimental manipulation and credible contrasts.
Merely studying one participant or measuring an outcome before and after intervention is insufficient.
Researchers should evaluate whether the intervention can be introduced, delayed, alternated, or withdrawn ethically and meaningfully.
The relevant decision is addressed in when a single-case experimental design is appropriate.
Can crossover designs improve efficiency?
Crossover designs allow participants to experience multiple treatment conditions in a planned sequence.
When effects are sufficiently reversible and outcomes can be measured repeatedly, participants may serve as their own controls.
This can reduce variation associated with stable individual characteristics.
However, crossover designs may be unsuitable for interventions that produce lasting learning, irreversible outcomes, or substantial carryover effects.
For example, students may retain knowledge acquired during the first instructional condition, affecting performance under the second.
The potential precision advantage should therefore be evaluated against treatment reversibility and period effects.
Researchers should consult the requirements for crossover designs in small populations before choosing this approach.
Can quantitative research remain valid?
Yes. A rare population does not automatically require qualitative research.
Quantitative studies can describe population characteristics, estimate outcome frequencies, investigate associations, or evaluate intervention effects under appropriate designs.
However, limited independent observations may produce imprecise estimates and constrain model complexity.
For example, fitting a regression model with numerous predictors among twenty participants may yield unstable results.
Researchers should prioritize scientifically essential variables and avoid analyses requiring more information than the data contain.
Appropriate small-sample methods may improve inferential calibration, but they do not automatically solve confounding or insufficient precision.
The broader issue is explained in quantitative research with small populations.
When might qualitative research be more appropriate?
Qualitative methods may be especially useful when the objective concerns experiences, meanings, institutional processes, or mechanisms that cannot be adequately represented through numerical measurements alone.
For example, interviews with members of a rare professional group may illuminate how specialized roles are negotiated within organizations.
However, qualitative research should not be selected merely because a quantitative power calculation produces an unattainable target.
Qualitative designs require their own justification concerning case selection, data adequacy, analytic procedures, reflexivity, and interpretation.
A small number of participants may provide rich evidence for a focused qualitative question, but the findings should not be presented as statistically representative without an appropriate basis.
The research objective, rather than the numerical constraint alone, should determine the methodological approach.
Can mixed methods strengthen rare-population research?
Mixed-methods research may be appropriate when quantitative patterns and qualitative explanations are both needed.
For example, researchers might describe academic outcomes for all members of a specialized student population and interview selected participants to understand the experiences underlying those patterns.
However, qualitative evidence does not statistically compensate for low quantitative power.
Likewise, a descriptive survey does not automatically validate qualitative interpretations.
The design should explain how the two components address complementary aspects of the research question and how their findings will be integrated.
Mixed methods are most useful when integration produces an answer that neither component could adequately provide alone.
Can Bayesian methods help when recruitment is fixed?
Bayesian methods may incorporate credible prior information and express uncertainty through posterior distributions.
For example, a small study may use relevant historical evidence to inform estimation of an outcome probability.
Hierarchical models may also support partial pooling across related small groups under appropriate assumptions.
However, posterior conclusions may depend strongly on prior assumptions when current data are sparse.
Researchers should justify the prior, assess historical comparability, and report sensitivity to plausible alternatives.
Bayesian methods do not create new observations or eliminate design bias.
See Bayesian designs for small sample sizes for further methodological considerations.
What can exact tests and specialized estimators contribute?
Exact tests, randomization-based procedures, and selected penalized estimators may be useful when conventional large-sample approximations are unreliable.
For example, Fisher's exact test may be appropriate for certain sparse contingency tables under its sampling assumptions.
Penalized logistic regression may help address estimation difficulties caused by separation.
However, these methods do not automatically increase the amount of information available.
An exact test can remain underpowered, and a penalized model can produce an estimate that is highly uncertain.
Researchers should consider exact tests and small-sample methods during research design planning, not merely after encountering analytical problems.
Should the study prioritize estimation over significance testing?
Often, particularly when the eligible population is too small to support decisive hypothesis tests.
Effect estimates and uncertainty intervals can describe what the data suggest and which substantively important possibilities remain unresolved.
For example, a treatment estimate may favor an intervention while its confidence interval includes meaningful benefit, no effect, and possible harm.
That result should not be described as proof of effectiveness or ineffectiveness.
Researchers should establish meaningful effect thresholds before examining outcomes where feasible.
However, estimation does not automatically make a small study conclusive. Wide intervals may demonstrate that the evidence is insufficient for a particular decision.
The distinction is examined in estimation versus significance testing in small-population studies.
What if the entire population is still too small for useful precision?
Researchers should acknowledge that the original inferential objective may not be achievable.
For example, a study may be unable to distinguish a small treatment benefit from no effect, even when every eligible participant is recruited.
The appropriate response may be to narrow the question, prioritize description, investigate individual trajectories, or contribute evidence to a larger coordinated research program.
However, researchers should avoid changing outcomes or hypotheses after examining the data merely to obtain a favorable conclusion.
Revisions should be scientifically justified and documented transparently.
Some questions genuinely cannot be answered with adequate precision using the available population and current methods.
Recognizing that limitation is preferable to presenting unsupported certainty.
How should researchers address selection bias?
Rare populations may be identified through specialist services, institutional networks, voluntary organizations, or registries.
These recruitment channels may systematically exclude some eligible individuals.
For example, a rare disease study recruiting only from specialist hospitals may overrepresent individuals with severe conditions or greater access to care.
Similarly, an educational study recruiting through professional associations may miss individuals who are not members.
Researchers should describe the recruitment frame and assess how inclusion mechanisms relate to the outcomes under investigation.
A high participation rate within a selective recruitment frame does not establish representativeness of the wider rare population.
Why is confidentiality particularly important?
Members of rare populations may be identifiable through combinations of characteristics even when names are removed.
For example, reporting a participant's exact professional role, institution, age, and specialized qualification may make identification possible when very few individuals share those characteristics.
Researchers should assess deductive disclosure risks during data collection, analysis, and publication.
Small-cell suppression, aggregation, careful quotation practices, and appropriate data-access restrictions may be necessary.
However, privacy protections should be balanced with the need to report scientifically meaningful information.
Researchers should obtain relevant ethical approvals and comply with applicable data-governance requirements.
Can rare-population findings generalize?
Generalizability depends on the target population, recruitment process, design, and supporting assumptions.
A complete census may provide strong descriptive evidence about the defined finite population during the study period.
However, it does not automatically establish that the same patterns apply to future cohorts or other settings.
Experimental replication across participants and contexts may strengthen broader causal claims.
Qualitative research may support transferability through detailed contextual evidence and transparent interpretation.
Researchers should distinguish direct conclusions about observed participants from extrapolations requiring additional assumptions.
The small size of a population does not eliminate its scientific importance, but it does shape the scope of defensible inference.
How should the final design be justified?
The research protocol should explain why the population is limited, how eligibility was defined, and which recruitment options were considered.
It should identify the primary estimand and the information required to answer the question.
Researchers should describe why the selected design is more appropriate than feasible alternatives.
Where the population cannot support conventional power or precision targets, the limitation should be stated before data collection.
The protocol should also specify how uncertainty, missingness, selection, and generalizability will be addressed.
A defensible study does not need to claim that its sample size is ideal. It needs to demonstrate that its objectives and conclusions are proportionate to the evidence available.
