When Is It Appropriate to Deliberately Oversample an Underrepresented Group?

Why Would Researchers Intentionally Recruit More People From One Group?

Suppose a population contains a relatively small subgroup that is central to the research question. If researchers sample strictly in proportion to population size, only a few members of that group may enter the study. The sample might resemble the population numerically while providing too little information to estimate outcomes for the subgroup with useful precision or compare it meaningfully with others.

Researchers can sometimes address this problem by deliberately sampling members of the smaller or underrepresented group at a higher rate. This is commonly called oversampling.

The word can sound methodologically suspicious. After all, if one group is deliberately recruited disproportionately, does that not distort the sample? The answer depends on what researchers are trying to estimate and how the sampling design is incorporated into the analysis.

Oversample When Proportional Sampling Would Leave Too Little Information for an Important Group

Oversampling Trades Proportionality for More Information About a Group

Oversampling is easiest to understand as a sample-allocation decision. Researchers intentionally select members of one population subgroup at a higher rate than their population share would otherwise produce.

The objective is usually not to make the group appear more common than it really is. It is to obtain enough observations to answer questions about that group.

Proportional Sampling Can Produce Too Few Participants From a Small Population

Suppose a subgroup constitutes 5% of a population and researchers draw a sample of 400 using a design that yields approximately proportional group representation. Only about 20 sampled participants would be expected from that subgroup on average.

Twenty observations may be adequate if researchers merely need a broad overall population estimate and have no subgroup-specific objective. It may be inadequate if the study needs a reasonably precise estimate for the smaller group or intends to compare outcomes across groups.

This is a central reason oversampling is used in survey research. Methodological research on sampling minority populations notes that ordinary designs can yield insufficient subgroup sample sizes for estimating population parameters and that oversampling can improve precision for small-domain estimates.

Oversampling Is Most Defensible When the Subgroup Analysis Matters Before Recruitment

The strongest justification exists when researchers can specify in advance why additional information about the group is needed.

The population may experience the phenomenon differently. Previous research may contain a substantial evidence gap. A policy decision may require subgroup-specific estimates. Researchers may need to assess whether an intervention performs differently across relevant populations. Or a group may have been persistently underrepresented in the existing evidence.

In each case, the sampling decision follows from an analytical purpose.

By contrast, deciding after data collection that a subgroup is interesting does not retroactively give the study enough information about it. If subgroup inference is important, recruitment and sample-size planning should address that need prospectively whenever feasible.

Oversampling Does Not Mean Recruiting an Arbitrary Number of Additional Participants

There is no universal oversampling ratio. Researchers should not automatically double the representation of every small group or aim for equal numbers across all groups.

The appropriate allocation depends on the intended analysis, expected precision, subgroup prevalence, total sample size, sampling frame, recruitment feasibility, cost, design effect, anticipated nonresponse, and analytical approach.

For a quantitative study, researchers may use formal sample-size or precision calculations to determine how many subgroup observations are needed. Survey statisticians may optimize allocation across strata given precision and cost constraints. The appropriate method depends on the design.

Oversampling Can Make the Raw Sample Less Proportional to the Population

This is intentional.

Suppose a target population is 90% Group A and 10% Group B. Researchers recruit 300 participants from Group A and 200 from Group B because they need a sufficiently informative comparison.

Group B now constitutes 40% of the sample even though it constitutes only 10% of the population.

The sample has substantially more information about Group B, but its raw proportions do not reproduce the target population. This illustrates why representation and representativeness should not be treated as synonyms.

Disproportionate Sampling Does Not Automatically Prevent Population Estimation

If participants are selected through a probability sampling design with unequal selection probabilities, researchers can account for those probabilities in analyses intended to estimate population quantities. Survey weights commonly incorporate the inverse of selection probabilities, sometimes together with other adjustments appropriate to the survey design.

In simple terms, an observation from a deliberately oversampled group may represent fewer population members than an observation from a group sampled at a lower rate.

The exact weighting procedure depends on the design. Researchers should therefore preserve information about strata and selection probabilities and use analytical methods appropriate to complex survey data when required.

Oversampling and Weighting Solve Different Problems

Oversampling increases the amount of information collected from a population. Weighting changes how observations contribute to particular population estimates.

Weighting cannot manufacture observations that were never collected. If only a handful of people from an important subgroup participate, giving those observations larger weights does not provide the same subgroup information as actually recruiting enough participants.

Oversampling Is Not the Same as Convenience Recruitment

Researchers may sometimes recruit additional participants from a particular population using targeted community outreach, specialized sampling frames, geographic concentration, multiple sampling frames, or other methods. The statistical implications depend on how selection occurs.

In a probability design, oversampling can involve explicitly different known selection probabilities across strata. In nonprobability research, researchers may also deliberately recruit more members of a particular population, but calling that effort “oversampling” does not create known selection probabilities or automatically support population inference.

Researchers should therefore describe the actual sampling and recruitment mechanism rather than assuming that the label itself establishes methodological rigor.

Oversampling Can Be Useful Even When Population Representativeness Is Not the Main Goal

Some studies seek sufficiently rich evidence from groups whose experiences would otherwise be difficult to examine. The objective may be comparative, exploratory, qualitative, mixed-methods, or focused on understanding variation rather than estimating population prevalence.

In those contexts, deliberately recruiting more participants from an underrepresented population may still be appropriate, although the sampling logic should match the methodology. Qualitative purposive sampling, for example, is conceptually different from disproportionate stratified probability sampling even if both result in greater participation from a particular group.

The common principle is intentionality: researchers should know why additional representation is needed and what conclusions the resulting sample is designed to support.

Oversampling Does Not Fix Barriers That Prevent Participation

A recruitment target is not an accessibility strategy.

If people from a population encounter inaccessible study sites, language barriers, restrictive eligibility criteria, inconvenient scheduling, technology requirements, distrust, or burdensome procedures, simply setting a larger numerical target may not increase enrollment.

Researchers may first need to examine whether participation itself is realistically accessible. FDA's current guidance on enhancing clinical-trial participation, for example, recommends attention to broader eligibility, site location, community engagement, accessible recruitment opportunities, and participant burden as part of increasing enrollment of representative populations.

These recommendations arise from regulated clinical research and should not be generalized as mandatory procedures for every discipline, but they illustrate an important distinction: a sampling objective and a recruitment strategy are not the same thing.

Oversampling Has Costs and Trade-Offs

Targeting a population that is geographically dispersed, relatively small, difficult to identify in a sampling frame, or historically poorly reached by research may require additional sites, screening, recruitment channels, community partnerships, time, and funding.

Disproportionate allocation can also reduce precision for some overall estimates if a fixed total sample is shifted away from other groups, depending on the design and estimator. Weighting may increase variance compared with a design in which selection probabilities are more uniform.

Oversampling should therefore be designed around the estimands and precision requirements that matter most, not added merely because a sample diversity target sounds desirable.

Ethical Recruitment Still Matters When a Group Is Deliberately Targeted

A scientifically justified need for additional participants does not override ordinary ethical obligations. Recruitment should remain voluntary, appropriately communicated, and consistent with applicable ethics requirements.

Researchers should also avoid treating communities merely as sources of participants needed to satisfy a target. The National Academies has emphasized sustained community engagement and attention to structural barriers in efforts to improve representation in clinical research.

Where community engagement is appropriate, it should inform how research is designed and conducted rather than functioning solely as a mechanism for filling quotas.

Sometimes Oversampling Is Not the Right Solution

If the research question does not require subgroup-specific information, deliberately reallocating a limited sample may offer little benefit. If a group is poorly defined or cannot be identified reliably in the sampling frame, the proposed strategy may not work as intended. If the main problem is an exclusionary eligibility criterion or inaccessible procedure, redesigning those features may be more appropriate.

Oversampling is therefore a tool for a specific sampling problem: insufficient information about an important population. It is not a universal remedy for every form of underrepresentation.

Why a Researcher Might Intentionally Sample a Small Group at a Higher Rate

The oversampling did not attempt to make distance learners appear more common. It deliberately exchanged raw proportionality for greater information about a population central to the research question.

Common Misunderstandings About Oversampling

Oversample for an Analysis You Can Name in Advance

Before deliberately increasing recruitment from a population, identify what the additional observations are supposed to accomplish. That purpose should guide both the target sample size and the analysis.

Finally, remember why the strategy exists. The objective is not to make the demographic table look balanced. It is to ensure that an important population contributes enough evidence for the question researchers genuinely need to answer.

Before Oversampling a Group, Check the Sampling Purpose

Questions About Oversampling Underrepresented Groups

Oversample to Obtain Information, Not to Manufacture Demographic Balance

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.