How Do You Choose a Sampling Strategy for Hidden or Hard-to-Reach Populations?

How Can You Study a Population Whose Members Are Difficult to Identify or Recruit?

Suppose you want to investigate how independent academic writers use generative AI when preparing manuscripts for clients. Some writers work through formal platforms, others obtain projects through personal networks, and many do not publicly disclose their professional activities.

There is no complete directory. Public advertisements may reach only the most visible practitioners, while institutional recruitment channels miss people working outside universities.

Should you ask participants to refer others? Recruit through online communities? Identify professional gathering places? Or attempt to construct a probability sampling design?

Similar challenges arise when researching informal workers, displaced populations, undocumented professional activities, people experiencing homelessness, or individuals whose participation could expose them to stigma or other risks.

The difficulty is not simply finding enough participants. A recruitment strategy may systematically reach well-connected, visible, or institutionally affiliated individuals while overlooking people whose circumstances differ.

How should researchers choose a sampling method when the population is difficult to enumerate, access, or engage?

Choose the Strategy According to Why the Population Is Difficult to Reach

Designing Research Sampling for Populations That Are Difficult to Identify or Access

What Is a Hidden or Hard-to-Reach Population?

A hidden population is generally one whose members cannot be readily identified or enumerated through conventional sampling frames. A hard-to-reach population is one whose members are difficult to contact, recruit, or retain using ordinary research procedures.

These categories overlap but are not identical.

For example, a professional population may be identifiable through a register but difficult to recruit because members have demanding schedules. Another population may be willing to participate but lack a comprehensive directory.

Some populations are difficult to access because participation could expose sensitive identities, activities, or circumstances.

Researchers should identify the actual barrier rather than treating hidden and hard-to-reach as interchangeable descriptions.

Why Does the Source of Inaccessibility Matter?

Different barriers require different sampling responses.

If no individual-level register exists, researchers may investigate geographic, institutional, or venue-based selection frames.

If the population is socially connected but not publicly listed, network-based recruitment may be feasible.

If participants are identifiable but reluctant to disclose sensitive experiences, confidentiality and trusted recruitment arrangements may matter more than changing the sampling method.

If people are geographically dispersed, online recruitment or multistage selection may reduce access costs.

The method should address the actual barrier rather than merely carry a label commonly associated with difficult populations.

Which Sampling Methods Can Be Considered?

These methods do not provide equivalent inferential guarantees. Their suitability depends on the target population, research objective, and actual selection procedures.

When Is Purposive Sampling Appropriate?

Purposive sampling is particularly useful when researchers need participants with specific experiences, roles, or characteristics.

For example, a qualitative study examining how independent academic editors negotiate the disclosure of AI assistance may deliberately recruit editors who have encountered relevant decisions.

The selection may emphasize contrasting professional arrangements, disciplinary contexts, or levels of experience.

Purposive sampling can support conceptual depth, but it does not ordinarily establish population inclusion probabilities.

Researchers should explain why selected participants are informative and how the recruitment process may have overlooked other perspectives.

When Should Conventional Snowball Sampling Be Used?

Conventional snowball sampling asks initial contacts or participants to identify additional people who may satisfy the eligibility criteria.

For example, an independent editor may refer another editor who works outside formal professional associations.

This approach can reveal individuals who would otherwise be difficult to identify.

However, referrals tend to follow existing social relationships. Participants with larger or more visible networks may have greater opportunities to be recruited.

People who are socially isolated or disconnected from initial recruitment networks may remain absent.

Researchers should not assume that several referral waves automatically eliminate these limitations.

How Can Snowball Sampling Be Strengthened?

Researchers can begin with initial contacts drawn from different relevant networks rather than relying on one highly connected individual.

For example, recruitment may begin with editors from different professional communities, employment arrangements, and geographic locations.

Referral procedures should be documented, and researchers should examine whether recruitment repeatedly reaches similar types of participants.

Where appropriate, participants may be invited to share study information with others rather than disclose another person's identity directly.

These measures may improve diversity and privacy, but they do not transform conventional snowball sampling into probability sampling.

What Is Respondent-Driven Sampling?

Respondent-driven sampling, or RDS, is a structured network-based recruitment approach developed for studying certain populations that are difficult to enumerate.

It typically begins with initial participants, often called seeds, who recruit eligible peers through controlled referral procedures.

Some RDS estimators use information about participants' reported network connections and recruitment patterns to estimate population characteristics.

However, these estimates depend on assumptions about network structure, recruitment behavior, participation, and the estimator used.

Researchers should distinguish the recruitment procedure from the statistical assumptions required for population inference.

When Is Respondent-Driven Sampling Suitable?

RDS may be considered when population members have meaningful social connections through which recruitment can proceed and when the research objective justifies its complexity.

For example, a population whose members know and can contact other eligible individuals may provide a plausible recruitment network.

However, network fragmentation, inaccurate reports of network size, preferential recruitment, and differential participation may undermine estimation.

RDS is not automatically appropriate for every population lacking a register.

Researchers should assess network feasibility before recruitment and evaluate assumptions during analysis.

Does Respondent-Driven Sampling Guarantee Representative Estimates?

No. RDS can produce useful estimates under suitable conditions, but the relevant assumptions may be difficult to satisfy or verify.

For example, initial participants may recruit peers who resemble themselves, while individuals outside connected networks remain unreachable.

Long recruitment chains do not necessarily eliminate all dependence on initial recruitment choices.

Researchers should examine recruitment patterns, network information, subgroup coverage, and the sensitivity of estimates to analytical assumptions.

RDS should not be presented as equivalent to conventional probability sampling without careful qualification.

What Is Time-Location Sampling?

Time-location sampling identifies venues and time periods where eligible population members may be encountered.

Researchers construct a venue-time sampling frame and select combinations of places and times for recruitment or further sampling.

For example, a study of informal workers may identify relevant gathering places and operating periods.

However, individuals who never attend the covered locations or times cannot enter through that frame.

Attendance frequency can also affect the probability of encountering particular individuals.

Researchers should distinguish probability selection of venue-time units from probability selection of individuals.

Can Venue-Based Sampling Be Probability-Based?

Potentially, when the selection of venue-time units and eligible individuals follows an appropriate random design and the necessary inclusion probabilities can be established.

However, randomizing venue visits alone does not guarantee known inclusion probabilities for individual population members.

Researchers may need information about attendance patterns, repeated visits, and eligibility at selected venues.

The target population may also need to be defined as people who attend the covered venues during the relevant period.

Claims about people who never attend those venues require additional evidence or assumptions.

Can Probability Sampling Work for Hidden Populations?

Sometimes. The absence of a complete individual-level register does not necessarily make probability sampling impossible.

For example, researchers may randomly select geographic areas and enumerate eligible individuals within selected areas.

Multistage designs may similarly select institutions, establishments, or other clusters before identifying individuals.

However, the approach must provide appropriate population coverage and known inclusion probabilities under the intended design.

Rare, mobile, or highly dispersed populations may make these approaches expensive or inefficient.

Researchers should examine sampling alternatives when no complete frame exists before assuming that only nonprobability methods are available.

How Does Network Structure Affect Sampling?

Network structure influences which participants can be reached through referrals.

For example, independent academic editors may form separate networks according to discipline, geographic location, language, or employment arrangement.

If recruitment begins in one network and few connections exist to others, the sample may remain concentrated within that group.

Highly connected participants may also have greater opportunities to enter the sample.

Researchers should examine whether network boundaries correspond to characteristics relevant to the study outcomes.

Recruitment diversity cannot be inferred merely from the number of referral chains.

What Is Homophily in Network-Based Recruitment?

Homophily refers to the tendency for social connections to occur among people who share certain characteristics.

For example, academic editors who specialize in medical manuscripts may be more likely to know other medical editors than editors working in unrelated disciplines.

Referral recruitment may therefore reproduce similarities among connected participants.

This can affect subgroup coverage and the composition of the achieved sample.

Researchers should consider whether recruitment networks are sufficiently connected across characteristics relevant to the research question.

How Can Online Recruitment Help?

Online recruitment can reach geographically dispersed participants who are difficult to contact through conventional institutional channels.

For example, researchers may recruit independent professionals through relevant online communities or professional platforms.

However, participation in these spaces is not universal.

Platform algorithms, membership requirements, language, internet access, and self-selection can influence who encounters and responds to invitations.

Using several online channels may improve reach, but it does not automatically establish population representativeness.

How Should Researchers Protect Participants in Sensitive Populations?

Recruitment should minimize unnecessary disclosure of identities, activities, and personal circumstances.

For example, asking participants to provide identifiable information about colleagues engaged in sensitive professional activities may create privacy risks.

Researchers should consider whether participants can share study invitations directly with peers rather than disclosing their contact details.

Consent procedures should explain recruitment, confidentiality protections, data handling, and any limits to confidentiality.

Applicable ethics and data protection requirements should guide recruitment and recordkeeping.

Methods that improve access should not expose participants or nonparticipants to disproportionate risks.

Can Incentives Affect Network-Based Recruitment?

Yes. Incentives may influence participation and referral behavior.

In some structured network-based studies, compensation may be provided for participation and eligible referrals under approved procedures.

However, incentives can affect which peers are recruited and may create undue pressure in particular circumstances.

Researchers should evaluate incentive amounts, local conditions, and participant vulnerability.

Ethics review and transparent reporting are important when incentives form part of the recruitment design.

How Can Researchers Reduce Gatekeeper Bias?

Gatekeepers are individuals or organizations that control or facilitate access to prospective participants.

For example, a professional association administrator may decide which members receive a research invitation.

Gatekeepers may unintentionally favor visible, cooperative, or institutionally connected individuals.

Researchers should document gatekeeper roles and consider multiple access routes when appropriate.

However, bypassing legitimate organizational permissions or privacy protections is not an acceptable solution.

Access arrangements should be evaluated for both ethical legitimacy and selection consequences.

How Do You Know Whether the Sample Is Becoming Too Narrow?

Researchers should monitor recruitment sources, participant characteristics, and the types of experiences represented.

For example, if nearly all recruited editors belong to one professional association, the sample may omit independent practitioners outside that organization.

In qualitative research, additional purposive recruitment may seek contrasting experiences.

In population estimation, however, adding convenient participants does not necessarily correct unknown inclusion probabilities.

Researchers should distinguish improving conceptual coverage from establishing statistical representativeness.

Can Sampling Weights Correct Network Selection?

Specialized network estimators may use reported network sizes and other information to adjust for unequal recruitment opportunities.

However, such adjustments depend on assumptions about network structure, reporting accuracy, and recruitment processes.

Weighting cannot automatically correct disconnected networks or systematic exclusion of particular groups.

Researchers should report the estimator used and evaluate the sensitivity of findings to relevant assumptions.

Weights should not be treated as proof that a referral sample is representative.

What If Population Size Is Unknown?

Unknown population size complicates some sampling designs and population estimates, but it does not make all research impossible.

Qualitative studies may investigate relevant experiences without estimating the total population size.

Some specialized population estimation methods may be considered when suitable data and assumptions exist.

However, researchers should not infer the size of a hidden population simply from the number of people recruited.

The research objective should determine whether population size estimation is necessary.

How Should Sample Size Be Determined?

Sample size planning should reflect the study's objective and selection method.

For population estimation, researchers may consider precision, design effects, anticipated recruitment patterns, and uncertainty associated with the estimator.

For qualitative research, sample adequacy may depend on the research aim, relevant variation, information power, and analytical strategy.

Network-based designs may require attention to recruitment waves, network structure, and subgroup coverage.

No universal sample size formula applies to every hidden population study.

How Can Selection Bias Be Assessed?

Researchers should identify who can enter the sample through the selected recruitment mechanism and who cannot.

For example, conventional snowball sampling may exclude individuals disconnected from initial referral networks.

Venue-based sampling may omit people who never attend selected locations.

Online recruitment may miss people without access to relevant platforms.

Where credible external information exists, sample characteristics may be compared with population benchmarks.

However, demographic similarity does not guarantee the absence of unmeasured selection bias.

When Should Multiple Recruitment Methods Be Combined?

Combining recruitment methods may improve access to different segments of a population.

For example, researchers may use professional directories, online communities, and participant referrals to identify independent academic editors.

However, combining nonprobability recruitment channels does not automatically create a probability sample.

Duplicate recruitment opportunities and differing participation mechanisms may complicate interpretation.

Researchers should document each source and distinguish broader reach from statistically defensible population coverage.

What If No Method Can Support Population-Wide Estimates?

Researchers may need to reconsider the intended inference.

For example, a study initially designed to estimate the prevalence of AI use among all independent academic editors may be unable to justify that estimate using the available recruitment mechanisms.

The study may still investigate patterns, experiences, or relationships among a clearly described recruited population.

Alternatively, the team may invest in constructing a more suitable sampling frame.

Researchers should not present an unsupported population estimate merely because the original question was ambitious.

How Should Hidden Population Sampling Be Reported?

Methods sections should describe the target population, eligibility criteria, recruitment sources, initial contacts, referral procedures, and achieved sample.

For network-based methods, relevant information may include recruitment waves, network measures, and estimator assumptions.

For venue-based sampling, report the venue-time frame and selection procedures.

For probability designs, explain how inclusion probabilities were established and how clustering or unequal probabilities were handled.

Researchers should identify likely coverage gaps, privacy protections, and limitations on population inference.

Recruiting Independent Academic Editors Across Disconnected Professional Networks

Suppose the researcher interviews 28 editors, including individuals from professional associations, freelance platforms, and informal networks.

The sample may support a rich qualitative analysis if participants provide relevant, sufficiently varied evidence.

However, the researcher cannot conclude that a particular percentage of all freelance academic editors follows a given AI disclosure practice.

The sample was selected for experiential relevance, not through known population inclusion probabilities.

Had the objective been population prevalence estimation, the team would have needed a different sampling design or additional assumptions and information supporting statistical inference.

Common Misconceptions About Sampling Hidden Populations

How to Select a Sampling Strategy for a Difficult-to-Reach Population

Identify whether the primary challenge concerns enumeration, physical access, social networks, willingness to participate, or confidentiality.

The appropriate strategy is the one that provides access to relevant participants while preserving the ethical and inferential conditions necessary for the research question.

What to Verify Before Sampling a Hidden or Hard-to-Reach Population

Frequently Asked Questions About Sampling Hidden Populations

Sampling Hidden Populations Requires a Defensible Access and Selection Strategy

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.