03 · What You Need to Know
How Target Populations, Accessible Populations, and Samples Differ
What Is the Target Population?
The target population is the collection of individuals or other entities about which a research study intends to produce knowledge.
It is defined by the research question and the population boundaries relevant to the intended inference.
For example, a study estimating generative AI use among undergraduate students might define its target population as all undergraduate students enrolled in recognized higher education institutions in the Philippines during a specified academic year.
The target population includes eligible individuals whether or not they can be contacted, are willing to participate, or belong to institutions cooperating with the researcher.
Importantly, the target population need not consist of people. A study examining institutional AI governance might target universities, while another examining healthcare delivery might target hospitals.
The appropriate definition follows the intended research question and target population boundaries.
What Is the Accessible Population?
The accessible population is the portion of the relevant population that can realistically be identified, contacted, or otherwise accessed through the available recruitment arrangements.
Accessibility may depend on institutional permission, geographic location, available registers, recruitment channels, resources, and practical constraints.
For example, a researcher interested in all undergraduate students nationally may have permission to recruit only from five universities.
Eligible students within those institutions who can be reached through the approved procedures form an accessible population for that study.
Accessibility should not be confused with willingness to participate. A student may be reachable and eligible but decline the invitation.
The accessible population is therefore conceptually distinct from the responding sample.
What Is the Study Sample?
The study sample is the subset of units selected or recruited from the relevant population to participate in the research.
However, the term sample is sometimes used ambiguously in research reporting.
A probability survey may distinguish selected individuals, eligible contacted individuals, respondents, and the final analytic sample.
For example, 1,000 students might be selected from an enrollment register, 900 might be eligible and successfully contacted, 700 might respond, and 680 might have the information required for a particular analysis.
These counts represent different stages.
Researchers should identify which count they mean rather than describing every stage simply as the sample size.
How Do the Three Concepts Compare?
| Dimension |
Target population |
Accessible population |
Study sample |
| Central question |
Who do you want to learn about? |
Who can you realistically reach? |
Who is selected or recruited? |
| Defined primarily by |
Research question and estimand |
Access and operational arrangements |
Sampling and recruitment procedures |
| Typical scope |
Broadest intended inference population |
Reachable portion of that population |
Selected subset of eligible units |
| Principal concern |
Conceptual and population relevance |
Coverage and access limitations |
Selection, participation, and sampling error |
| Illustrative educational study |
All eligible undergraduate students nationally |
Eligible students reachable in participating universities |
Students recruited through the implemented procedure |
The table describes a common arrangement, not a universal rule that the three groups must always be perfectly nested. Eligibility changes, overlapping recruitment channels, and frame errors can produce more complicated relationships.
Is the Accessible Population Always a Subset of the Target Population?
Conceptually, the accessible population should consist of reachable members of the intended target population.
Operationally, however, recruitment systems may include individuals who do not satisfy the target population definition.
For example, an institutional mailing list may contain former students, postgraduate students, or individuals who have already withdrawn.
These individuals may be reachable but ineligible.
Researchers should distinguish the broader contactable population from the eligible accessible population and verify eligibility before selection or analysis.
This distinction becomes particularly important when administrative records are incomplete or outdated.
Where Does the Sampling Frame Fit?
The sampling frame is the operational list or mechanism used to identify units available for selection.
It is related to the accessible population but is not necessarily identical to it.
For example, a university may have 12,000 eligible undergraduate students, all of whom could theoretically be approached through institutional channels. Yet the enrollment register supplied to the researcher might contain only 11,000 eligible records.
The register would underrepresent the accessible population.
Alternatively, it might include duplicate or ineligible records.
Researchers should therefore assess whether the sampling frame matches the intended population instead of assuming that every available list is complete.
Can the Target and Accessible Populations Be Identical?
Yes. In some studies, the entire target population may be identifiable and reachable.
For example, a researcher might investigate all 35 faculty members employed in a particular academic department during a specified semester.
If every eligible faculty member can be contacted, the accessible population may coincide with the target population.
However, the final responding sample may still be smaller if some faculty members decline or do not provide usable data.
Complete access therefore does not guarantee complete participation.
Even when all eligible individuals respond, the study's conclusions remain bounded by how the target population was originally defined.
Why Does the Difference Matter for Generalizability?
Generalizability concerns whether findings can reasonably apply beyond the particular observations analyzed.
If the accessible population differs systematically from the target population, findings from the accessible group may not represent the broader population.
For example, universities willing to participate in an educational technology study may have stronger digital infrastructure than universities that decline.
Students from those institutions may consequently have different experiences with AI-supported learning.
Even if the researcher randomly samples students within participating universities, the results may not provide an unbiased estimate for all university students nationally.
Generalizability depends on the selection process and relevant differences between populations, not simply on whether individual respondents were randomly chosen.
Does a Large Sample Solve a Narrow Accessible Population?
Not necessarily. A large sample can reduce sampling variability within the population represented by the sampling design.
It cannot automatically eliminate systematic differences between accessible and inaccessible members of the target population.
For example, surveying 5,000 students from two technologically advanced universities may produce precise estimates for those institutions while providing limited evidence about students in institutions with different resources.
The issue is not whether 5,000 is numerically sufficient. It is whether the selection process supports the intended population inference.
This explains why a large sample can still be seriously biased.
How Does Probability Sampling Affect the Relationship?
Probability sampling uses a defined selection mechanism in which relevant units have known or appropriately calculable inclusion probabilities.
When the sampling frame adequately covers the target population and the design is correctly implemented, probability sampling can support design-based population inference.
However, random sampling within a restricted accessible population does not automatically justify inference to a broader target population.
Suppose students are randomly selected from five conveniently recruited universities.
The selection may be probability-based within those institutions, but the universities themselves were not selected through a national probability design.
Researchers should distinguish the level at which random selection occurred and the population to which that selection mechanism applies.
What About Nonprobability Sampling?
Nonprobability sampling does not provide known inclusion probabilities for all relevant population units.
Convenience, purposive, quota, and some network-based recruitment strategies fall within this broad category.
These methods can be appropriate for research questions emphasizing particular experiences, conceptual understanding, feasibility, or specific populations.
However, conventional design-based margins of sampling error do not automatically apply to nonprobability samples.
Statistical population estimates may require additional modeling assumptions and external information.
Researchers should avoid treating a large nonprobability sample as equivalent to a probability sample merely because it includes many respondents.
How Can Selection Bias Occur Between These Populations?
Selection bias can arise at several stages.
First, institutional access may exclude certain groups from the accessible population.
Second, the sampling frame may omit eligible individuals within participating institutions.
Third, selected individuals may differ in their likelihood of responding.
Finally, analytical exclusions or missing data may remove participants whose characteristics are related to the study variables.
These processes can influence population estimates and, under certain conditions, estimated relationships between variables.
Researchers should evaluate each stage separately rather than describing all discrepancies as sampling bias without identifying their mechanisms.
What Is the Difference Between Coverage Error and Nonresponse Error?
Coverage error concerns whether the sampling mechanism includes the intended population units.
Nonresponse error concerns the failure to obtain information from eligible units selected or invited through that mechanism.
For example, a student missing from the enrollment register represents a coverage problem.
A student included in the register, selected for the survey, and subsequently declining participation represents nonresponse.
The distinction matters because the remedies differ.
Updating the sampling frame may improve coverage, while reminders, accessible survey formats, or appropriate nonresponse adjustments may address some participation problems.
Neither remedy guarantees elimination of bias when important differences remain unmeasured.
Can Statistical Weighting Bridge the Differences?
Weighting may help when selection probabilities or relevant population characteristics are known.
For example, survey weights can account for unequal inclusion probabilities, while calibration adjustments may align sample distributions with known population totals.
However, weighting cannot guarantee that an accessible population represents a broader target population.
If entire categories of institutions are absent, their outcomes may not be recoverable from the observed data without additional assumptions.
Weighting is therefore a methodological tool rather than a substitute for appropriate population coverage.
What Does Transportability Add to This Discussion?
In causal research, transportability concerns whether an effect estimated in one population can be applied to another population under specified assumptions.
For example, an AI tutoring intervention evaluated among students in well-resourced universities may have different effects among students with limited connectivity.
Transporting the effect requires considering characteristics that modify treatment effects and whether those characteristics are adequately represented or measured.
Randomization can strengthen internal causal inference within a study, but it does not automatically establish transportability to all members of a broader target population.
Researchers should distinguish estimating an effect credibly in the study population from demonstrating that the same effect applies elsewhere.
Do Qualitative Studies Need These Distinctions?
Yes, although their inferential purposes may differ.
A qualitative study may aim to understand experiences among a particular category of participants rather than estimate a population proportion.
For example, researchers might investigate how first-generation university students experience AI-supported academic writing.
The relevant participant group should be defined clearly, even when statistical representativeness is not the objective.
Purposive recruitment may emphasize experiential relevance, variation, and conceptual depth.
Researchers should nevertheless distinguish the broader group of interest from the individuals accessible through particular institutions and those ultimately interviewed.
Claims about transferability should be supported by contextual description and analysis rather than unsupported assertions of population representativeness.
How Should These Populations Be Reported?
Methods sections should define the target population, describe the accessible population, identify the sampling frame where applicable, and explain how the study sample was selected.
Report the number of units eligible, invited, selected, responding, and analyzed when those counts are available and relevant.
Explain important institutional, geographic, temporal, and eligibility restrictions.
Where the accessible population differs from the target population, identify the likely implications for inference.
Reporting guidance such as STROBE emphasizes clear descriptions of study settings, eligibility criteria, participant selection, and analytical populations.
