Sampling in research is the systematic process of defining the group of interest (population), specifying the accessible subset that can be contacted or observed (sampling frame), selecting an approach for choosing units from that frame (sampling strategy), and understanding what those choices allow or limit (bias, precision, and transferability). Good sampling aligns the research question with an explicitly defined target population and an appropriate frame, uses a sampling strategy that balances validity with feasibility, and documents limitations so readers can judge the study’s applicability (Barroga and Matanguihan, 2022; Gionfriddo et al., 2024). For projects involving people or sensitive data, researchers must consult their institutional ethics and supervisory support before proceeding.
Introduction
Sampling sits at the heart of empirical research. The logic of inference—extending findings from observed units to broader populations—depends on assumptions about how those observed units were selected. A clear conceptual distinction between target population, sampling frame and sample is essential for transparent reporting and defensible interpretation of results. This article explains the core terms, contrasts probability and non-probability approaches, presents pragmatic decision steps, and outlines common sources of bias and how to mitigate or disclose them. It is intended as general educational guidance; for live studies involving human participants or sensitive data, consult institutional ethics committees or supervisory staff.
Key concepts: population, target population and sampling frame
- Population (or source population): the complete set of entities relevant to the research question (for example, “all registered primary schools in England in 2024”).
- Target population: the population to which the researcher intends the results to apply (for example, “primary schools serving urban areas with >50 pupils”).
- Sampling frame: a practical list or mechanism that identifies units from which a sample can be drawn (for example, the Department for Education registry, a membership database, or an online panel).
Table 1 contrasts these elements and common pitfalls.
| Concept | Definition | Typical example | Risk if poorly specified |
| Population | Entire set of interest | All adults (18+) in a city | Overly broad inferences |
| Target population | Subset intended for inference | Adults with diagnosed diabetes | Misalignment with collected data |
| Sampling frame | Practical list enabling selection | Patient registry, electoral roll | Coverage errors, excluded subgroups |
Table 1. Distinguishing population, target population and sampling frame.
A mismatch between the target population and the sampling frame produces coverage error: groups present in the target population but absent or underrepresented in the frame. Coverage error compromises transferability (external validity) and should be acknowledged and quantified where possible (USC Libraries, 2026).
Sampling strategies: probability and non-probability
Sampling strategies broadly fall into two families.
Probability sampling
- Every unit in the frame has a known, non-zero probability of selection.
- Enables unbiased estimators and straightforward calculation of sampling error when executed correctly.
- Typical methods: simple random sampling, systematic sampling, stratified sampling, cluster (or multi-stage) sampling.
Non-probability sampling
- Selection probabilities are unknown; pragmatic or purposive choices are used.
- Useful for exploratory work, qualitative research, or when frames are inaccessible.
- Typical methods: convenience, purposive, quota, snowball sampling.
Table 2 summarises common methods, strengths and limitations.
| Method | Type | Strengths | Limitations |
| Simple random sampling | Probability | Conceptually straightforward; unbiased | Requires complete frame; may be inefficient for rare subgroups |
| Stratified sampling | Probability | Improves precision for subgroups | Requires stratification variables in frame |
| Cluster sampling | Probability | Cost-effective for dispersed populations | Design effects (intra-cluster correlation) reduce precision |
| Systematic sampling | Probability | Simple to implement | Periodicity in frame can bias sample |
| Quota sampling | Non-probability | Ensures distribution on key variables | Unknown selection probabilities; bias persists |
| Purposive sampling | Non-probability | Focus on relevant or information-rich cases | Not statistically generalisable |
Table 2. Overview of sampling methods.
Probability methods support inferential claims about the target population with quantifiable uncertainty (USC Libraries, 2026). Non-probability methods are defensible when the research seeks depth, theory development, or when probability sampling is infeasible—provided limitations are transparent (Barroga and Matanguihan, 2022).
When to prefer probability over non-probability
Choose probability sampling when:
- The research question requires population-level estimates (prevalence, mean scores).
- Resources allow construction and use of an adequate frame.
- Quantification of sampling error and confidence intervals is essential.
Non-probability may be preferable when:
- The objective is hypothesis generation, in-depth understanding, or testing measurement instruments.
- The frame cannot be constructed or access is restricted.
- Rapid, low-cost data collection is needed, with appropriate caveats.
Gionfriddo et al. (2024) highlight that reproducibility and transparent reporting of sampling methods improves cumulative evidence; where non-probability methods are used, authors should document recruitment mechanisms and sample characteristics thoroughly.
Designing the sampling frame
A sampling frame can be a register, list, database, or an operational mechanism (e.g., all people visiting a clinic during specific hours). Key design considerations:
- Coverage: Does the frame include the target population? Identify omissions and estimate their magnitude if possible.
- Currency: Is the frame up-to-date? Stale frames can inflate non-contact rates.
- Accuracy: Are entries complete and correctly classified?
- Accessibility and permissions: Do GDPR or organisational rules limit use of the frame?
Practical checks at the frame stage reduce downstream bias and facilitate transparent reporting. If the frame is derived from multiple sources, document linkage rules, criteria for inclusion and deduplication strategies.
For a related explanation of why a mismatched frame can weaken a project, read dissertation methodological rigour failures.
Sample size: beyond rule-of-thumb
Sample size depends on the study objective, acceptable precision, expected variation, and design effect. For a simple mean or proportion in probability samples, classical formulae yield required sample sizes given a confidence level and margin of error. In clustered designs, multiply by a design effect (Deff = 1 + (m-1)ρ) where m is cluster size and ρ intra-cluster correlation. For regression models or multivariable analysis, consider events-per-variable or simulation-based approaches.
USC Libraries (2026) provides accessible guidance on sample size logic and the interpretation of uncertainty. For qualitative saturation, sample size is guided by analytic goals rather than statistical formulae.
Practical trade-offs
Cost, time and access shape realistic sampling choices. For example, a nationwide probability sample may be ideal but prohibitive; a stratified quota approach using an online panel could meet the need for timely subgroup estimates, provided panel composition is compared to known population benchmarks and weights applied. Document decisions and justify compromises.
For a related discussion of scope and quantitative project framing, see statistics project ideas.
Fictional planning application
Below is a condensed, fictional planning application for a cross-sectional survey intended to illustrate transparent sampling decisions.
Project title: Urban Active Travel Survey 2026
Research question: What proportion of adults aged 18–65 in City X cycle for commuting at least once per week?
Target population: Adults aged 18–65 residing in City X.
Sampling frame: Adult residents registered with the municipal electoral roll (most complete public register) supplemented by a household address file for residents not on the roll (e.g., recent movers, non-registered residents).
Sampling strategy:
- Stage 1 (frame construction): Merge electoral roll with household address file, deduplicate using address and name heuristics.
- Stage 2 (stratification): Stratify by ward (n=20) and neighbourhood deprivation quintile to ensure geographic and socio-economic coverage.
- Stage 3 (selection): Randomly select 50 addresses per ward using simple random sampling; within households, use Kish grid to select one adult aged 18–65.
- Stage 4 (weights): Compute survey weights to adjust for selection probability and non-response, calibrate to known age–sex and ward distributions.
Estimated sample size: 1,000 completed interviews (50 per ward × 20 wards). Power calculation: With an expected cycling prevalence of 15% and 95% confidence, margin of error ≈ 2.5% (ignoring design effect). Anticipate design effect 1.2 due to household clustering; inflate sample accordingly.
Ethics and data protection: Secure approval from the Institutional Review Board. Use opt-out initial contact letter with clear information and a dedicated response mechanism. Store personal identifiers separately from survey responses and follow data minimisation principles. For live studies involving people, consult institutional ethics and supervisory support.
This fictional plan demonstrates explicit links between question, target population, frame and sampling method; it shows how stratification and weighting can improve precision and reduce bias while noting design effects and ethical oversight.
Practical sequence
Implement sampling in a stepwise, auditable way.
- Define the research question and intended inference precisely.
- Specify the target population in operational terms (age, geography, inclusion/exclusion).
- Identify candidate frames and evaluate them on coverage, currency and access.
- Choose sampling strategy aligned with inference needs and constraints (probability if population estimates are required).
- Develop a documented selection algorithm (random number seeds, systematic interval, Kish grid).
- Pilot recruitment procedures (contact modes, response rates) to refine non-response adjustments.
- Collect data according to protocol; log attempts and non-response reasons.
- Compute design weights and adjust for non-response and calibration targets.
- Analyse with methods appropriate to the sampling design (use survey weights, account for clustering and strata).
- Report sampling methods, participation rates, limitations and the likely direction and magnitude of biases.
This sequence makes sampling decisions reproducible and helps readers judge the strength of inferences (Gionfriddo et al., 2024).
Selection bias, non-response and transferability
Selection bias arises when inclusion in the sample is associated with the outcome of interest after conditioning on design variables. Key contributors:
- Coverage error: parts of the target population are not in the frame.
- Non-response bias: eligible units decline or cannot be contacted.
- Self-selection: volunteers differ systematically from non-volunteers.
Mitigation strategies:
- Use probability sampling when feasible.
- Improve frame coverage or document excluded groups.
- Implement follow-up to reduce non-response and collect paradata on non-respondents.
- Apply weighting and calibration to known population margins.
- Sensitivity analyses: examine robustness of findings under plausible non-response scenarios.
Transferability (external validity) refers to whether results apply beyond the studied sample. For quantitative probability samples, transferability to the target population is defensible within sampling uncertainty and documented frame coverage. For non-probability samples, transferability is weaker and relies on theoretical or substantive arguments plus empirical comparisons to benchmarks.
Critical limitations and safeguards
Every sampling approach has limits. Below are important constraints and suggested safeguards.
- Imperfect frames: Acknowledge omissions, estimate their size, and discuss likely direction of bias. Use auxiliary sources or dual-frame designs where feasible.
- Measurement and sampling confounding: Avoid procedures where selection is driven by the outcome variable or its proxies.
- Small subgroup precision: Provide confidence intervals and, where necessary, aggregate categories or plan oversamples with proper weighting.
- Design effects: Account for clustering and stratification in variance estimation.
- Data protection and ethics: Follow GDPR (or equivalent), anonymise data where possible, minimise identifiable data collection, and obtain ethics approvals. For any research involving people or sensitive data, seek institutional ethics and supervisory guidance before data collection.
- Transparency: Pre-register sampling plans where possible and report deviations and non-response transparently.
These safeguards reduce risks associated with inference and improve the interpretability of results (Barroga and Matanguihan, 2022; Gionfriddo et al., 2024).
Tables to support reporting
The following table is an example template for reporting sample flow. Researchers should adapt it to their study.
| Step | Number | Notes |
| Frame size (initial records) | 50,000 | After deduplication |
| Eligible records after screening | 45,000 | Excluded non-residents |
| Selected for contact | 5,000 | Random selection |
| Contacted (response attempts) | 4,200 | Contacts made at least once |
| Refusals | 1,800 | Document reason where given |
| Ineligible upon contact | 200 | Moved away, age-out |
| Completed interviews | 2,200 | Final sample |
Table 3. Example sample flow table.
A second useful table is a coverage comparison: list major demographic benchmarks and sample proportions.
| Benchmark (census) | Sample proportion | Difference |
| Age 18–34 | 30% | -2% |
| Female | 51% | +1% |
| Employed full-time | 55% | -4% |
| Ethnic minority | 18% | +3% |
Table 4. Example coverage comparison for transparency.
Provide both raw and weighted estimates in reporting so readers can assess how weighting changed distributions.
FAQs
Q1: Can I generalise results from a non-probability sample?
A1: Generalisation is weaker with non-probability samples because selection probabilities are unknown. You can present descriptive results and argue for transferability using theoretical justification, sample benchmarking against population data, and sensitivity analyses; however, probability samples remain the gold standard for population-level inference (USC Libraries, 2026).
Q2: What is the difference between non-response and selection bias?
A2: Non-response is a specific mechanism where some sampled units do not provide data; if non-respondents differ systematically on outcomes, non-response induces selection bias. Selection bias more generally covers any systematic difference between the sampled units and the target population due to the selection process or frame omissions.
Q3: How should I report sample weighting?
A3: Report the weighting steps: base weights (inverse selection probabilities), non-response adjustments, and post-stratification or calibration targets. Provide descriptive statistics before and after weighting, and use survey-aware analysis that accounts for weights, strata and clusters.
Q4: When is cluster sampling preferable?
A4: Use cluster sampling when the population is geographically dispersed and in-person data collection costs are high. Clusters (e.g., schools, households) reduce fieldwork costs but increase intra-cluster correlation; plan for higher sample sizes or more clusters to preserve precision.
For related decision support, use Statistics and SPSS resources and dissertation-methodology support.
References
Barroga, E. and Matanguihan, G.J. (2022) ‘A practical guide to writing quantitative and qualitative research questions and hypotheses in scholarly articles’, Journal of Korean Medical Science, 37(16), e121. Available at: https://doi.org/10.3346/jkms.2022.37.e121 (Accessed: 26 August 2026).
Gionfriddo, M.R., McClendon, C., Nolfi, D.A., Kalarchian, M.A. and Covvey, J.R. (2024) ‘Back to the basics: Guidance for designing good literature searches’, Research in Social and Administrative Pharmacy, 20(4), pp. 463–468. Available at: https://doi.org/10.1016/j.sapharm.2024.01.009 (Accessed: 26 August 2026).
USC Libraries (2026) ‘Quantitative methods’, Organizing Your Social Sciences Research Paper. Available at: https://libguides.usc.edu/writingguide/quantitative (Accessed: 26 August 2026).