Determination of Sample Size through the Approach: A Practical Research Guide

Quick Answer: How Is Sample Size Determined through the Right Approach?

First, define the primary research question and identify whether the main aim is estimation, hypothesis testing, prediction, or qualitative understanding. Next, choose the sampling design and statistical analysis, set defensible assumptions, calculate the minimum analysable sample, and adjust the recruitment target for expected incomplete data or nonresponse.

For a simple random sample used to estimate a population proportion, a common starting calculation is n = Z²p(1 − p)/d², where Z is a confidence-level critical value, p is the anticipated proportion, and d is the desired absolute margin of error. In an independent two-group comparison, calculate sample size from the target difference, variability, significance level, desired power, and allocation ratio instead.

No universal formula fits all research designs. Finite populations, clustered recruitment, repeated measurements, unequal allocation, qualitative interviews, and complex analyses may require different formulas, software, or simulations. Write down your assumptions before collection begins and report the calculation in enough detail that another researcher could reproduce it.

Key Takeaways

• Choose a sample-size approach to match the primary outcome and study objective, not a fixed percentage of the population.
• Precision-based calculations answer “How accurately can we estimate?”; power calculations answer “How likely are we to detect a specified effect?”
• For simple random-sample proportion estimates, n = Z²p(1 − p)/d² is a starting formula, not an all-purpose rule.
• Use credible inputs for prevalence, effect size, variability, confidence, significance, and power; test plausible alternatives.
• Account separately for clustering, finite-population size where applicable, and expected missing or unusable responses.
• Qualitative sampling is often justified through information needs, diversity, and iterative analysis rather than statistical power.
• Report the initial calculation, any adjustments, the final recruitment target, and the actual analysable sample.

What This Page Covers

• The meaning of sample size, population size, sampling unit, and analysable sample
• How to decide between precision, power, and qualitative information approaches
• Step-by-step sample size calculations for proportions, means, and comparisons
• Finite-population correction, nonresponse allowance, and cluster design effects
• Three worked research situations and a qualitative case
• Common mistakes, decision checks, and reporting standards
• When independent calculation is sufficient and when specialist review is useful

Table of Contents

1. What sample-size determination means
2. Why a defensible sample size matters
3. A step-by-step planning workflow
4. Formulas for precision and power
5. Adjustments for real-world recruitment
6. How to check assumptions and report results
7. Common calculation mistakes
8. Worked examples and case studies
9. Proposal and dissertation checklist
10. Frequently asked questions

Methodology and Academic Sources

This educational guide synthesises standard statistical planning principles with reporting guidance for research studies. The STROBE checklists ask authors of observational research to explain how the study size was reached. The CONSORT 2025 sample-size guidance sets out information to report for trials, including target differences, outcome assumptions, significance, power, and adjustments for missing data.

For analytic planning, the peer-reviewed review of sample-size determination and G*Power explains principles of power analysis, while the G*Power project at Heinrich Heine University Düsseldorf describes the software's supported tests. A more general educational review of sampling methods and sample size discusses prevalence-based calculations and sampling considerations.

Those resources provide methodological context, not permission to apply one number to every study. The research protocol, disciplinary conventions, ethical review requirements, analysis plan, and supervisor or journal instructions take precedence for a particular project. Calculations below are illustrative and assume the stated designs; complex studies may require a statistician to develop or verify the specification.

What Does Determination of Sample Size through the Approach Mean?

It means deriving a defensible sample target from the way evidence will be collected and analysed. A sample is a subset of eligible units drawn from a target population. A unit might be a person, household, school, laboratory specimen, patient visit, document, or study site. The correct unit matters: 300 observations from 30 participants are not necessarily 300 independent participants.

Population size (N) describes the total number of eligible units in the defined population. Sample size (n) is the number selected or analysed, depending on the reporting stage. Recruitment target is the number to approach or enrol so that enough complete observations remain. Sampling method describes how units are selected—such as simple random, stratified, cluster, systematic, purposive, or convenience sampling. These concepts are related but cannot be used interchangeably.

Which approaches are available?

This comparison is a starting point, not a substitute for design-specific methodology. For example, a school survey measuring prevalence might need both a precision-based calculation and a cluster design adjustment because students in the same school can be similar.

Why sample size is not the same as representativeness

A large sample does not automatically remove selection bias. Interviewing 2,000 volunteers from one online group may provide less generalisable information than a smaller probability sample designed to cover the intended population. The size calculation controls selected aspects of uncertainty conditional on its assumptions; recruitment quality, coverage, measurement error, and missing-data patterns also affect trustworthiness.

Why Does Sample-Size Determination Matter for Researchers?

A credible target improves both the scientific value and feasibility of the study. Too few observations may produce wide confidence intervals, unstable estimates, or low probability of finding an effect the study was meant to detect. Recruiting far more participants than necessary can consume time, funds, and participant effort without answering a meaningfully better question.

For a dissertation, the sample-size rationale connects the objectives, hypotheses, recruitment procedure, instruments, and planned analysis. For a journal manuscript, transparent reporting helps reviewers distinguish a prospectively planned design from a number chosen after the results were known. For a clinical or community study, the calculation also contributes to ethical decisions about participant burden and the value of the proposed evidence.

Questions supervisors and reviewers may ask

• Which single primary outcome or main research objective drove the calculation?
• Was the purpose estimation, comparison, model development, or qualitative insight?
• Where did the expected proportion, standard deviation, or meaningful effect size come from?
• Which sampling unit and recruitment design were assumed?
• How were nonresponse, attrition, clustering, and unequal group sizes handled?
• Can another researcher reproduce the assumptions and numeric result?

If you can answer these questions before fieldwork, it becomes easier to defend the design during proposal review and to explain deviations after recruitment.

Can You Determine the Sample Size with Free Tools?

Yes, straightforward designs often can be planned with standard formulas, spreadsheets, or reputable free statistical software. The bigger risk is selecting the wrong formula or supplying unrealistic inputs. A calculator that prints “385” does not know whether you sampled individual students independently, recruited entire classrooms, or need enough observations for a subgroup comparison.

For a simple survey proposal, a well-documented spreadsheet may be enough. For a multi-centre intervention with repeated measures, expert design consultation before recruitment is often the more responsible choice. Editing and formatting assistance can clarify the explanation, but it cannot validate an incorrect statistical model merely by improving the wording.

When Is Self-Service Planning Enough and When Is Specialist Guidance Useful?

Self-service planning is usually reasonable when the design is standard, the primary outcome is clearly defined, and credible assumptions are available. The researcher should still verify calculations independently, ideally using a second method or supervised review.

Specialist statistical guidance is especially useful for multiple primary endpoints, noninferiority or equivalence tests, rare outcomes, regression prediction models, clustered or repeated-measure data, survival analysis, and substantial uncertainty about recruitment losses. In such studies, simulation or dedicated sample-size methods may be more appropriate than textbook approximations.

Separate methodological advice from writing support. A statistician can help evaluate the assumptions and calculations; a research editor can make the approved method understandable, reproducible, and consistent across the abstract, methods, tables, and proposal. The authors decide the design and retain responsibility for the numbers submitted.

How to Determine Sample Size: Step-by-Step Approach

A reliable workflow begins with the primary question and ends with a documented, feasible recruitment target. The following sequence works as a planning framework across many designs, although its statistical implementation must be adapted.

Step 1: Define the population, unit, and primary objective

State exactly who or what qualifies for the study and where the findings are intended to apply. Distinguish sampling units from observations. A district survey of teachers has a different design from a survey that randomly selects schools and interviews every teacher within them. Next write one primary estimand or endpoint: for example, the proportion of students experiencing a particular issue, the mean score in a population, or the between-group difference in an outcome.

Step 2: Select estimation, testing, or qualitative justification

Use precision-based planning when the desired result is an estimate with an acceptable uncertainty interval. Use prospective power analysis when you aim to detect a pre-specified effect. For qualitative work, outline how case selection, diversity, information power, iterative analysis, and conceptual depth will inform adequacy. Do not insert a quantitative formula simply to make an interview study look more rigorous.

Step 3: Choose assumptions and document their sources

For proportions, identify p and the allowable absolute error. For means, identify a plausible standard deviation. For group comparisons, define the smallest worthwhile effect, analysis method, alpha, power, and allocation. Prefer prior studies with comparable populations or credible pilot information. Where estimates are uncertain, plan sensitivity scenarios instead of presenting one speculative number as exact.

Step 4: Calculate the required analysable sample

Use a method that matches the planned analysis and sampling design. Keep intermediate values and units so the calculation can be audited. Record whether the answer is total or per group. Round required counts upward rather than rounding to the nearest convenient number.

Step 5: Adjust for design features and incomplete data

If participants are recruited in clusters, account for intracluster similarity. If the population is small and probability sampling without replacement is used, consider a finite-population correction. Inflate the required completed responses to account for anticipated nonresponse or missing outcomes. Avoid blindly applying adjustments twice: software may already incorporate particular design or attrition assumptions.

Step 6: Check feasibility, run sensitivity scenarios, and report

Compare recruitment requirements with resources, ethics approval, and access to participants. Recalculate using plausible higher and lower estimates of prevalence, standard deviation, attrition, or effect size. If the required target cannot be reached, revise the objectives or research design transparently rather than silently relaxing the assumptions after data collection. Specify the final calculation and any departures in the methods section.

Sample Size Formulas: Precision-Based and Power-Based Approaches

The right formula depends on the primary parameter being estimated or compared. These common equations provide transparent examples under simplified assumptions, not substitutes for design-specific guidance.

Estimating a proportion or prevalence

For a large population sampled using simple random sampling, a common starting formula is n₀ = Z² × p(1 − p) / d². Here Z = 1.96 for a conventional 95% two-sided confidence interval, p is the expected proportion written as a decimal, and d is the desired absolute margin of error. If p is unknown, p = 0.50 yields the maximum p(1 − p) and is a conservative planning choice for this simple formula.

Example: for 95% confidence, p = 0.50, and ±0.05 absolute precision, n₀ = (1.96² × 0.50 × 0.50) / 0.05² = 384.16. Round up to 385 completed responses before further corrections. This calculation does not make a convenience sample representative and does not automatically apply to clustered data.

Estimating a population mean

A simple starting expression for the confidence-interval precision of a mean is n₀ = (Z × σ / d)², where σ is the expected population standard deviation and d is the desired absolute error in the units of the outcome. With σ = 12 points, d = 3 points, and 95% confidence, (1.96 × 12 / 3)² = 61.47; therefore plan at least 62 complete observations. Small-sample t-distribution considerations or unknown variability may make iteration advisable.

Comparing two independent group means

For two equally sized independent groups, a normal-approximation planning formula is n per group ≈ 2(Z_1−α/2 + Z_1−β)² σ² / Δ². Here Δ is the difference worth detecting, σ the common standard deviation, α the two-sided significance level, and 1−β the desired power. This approximation presumes a suitable two-sample comparison and does not represent paired, repeated, or clustered outcomes.

If Δ = 6 points, σ = 15 points, α = 0.05, and power = 80%, n ≈ 2 × (1.96 + 0.84)² × 15² / 6² = 98 participants per group, or 196 analysable participants in total. A test-specific program may return a slightly different integer because of its precise distribution or other settings. Smaller target effects, more variability, or higher power typically increase the sample size.

Confidence level and power are different decisions

A 95% confidence interval describes an inferential procedure's long-run coverage under the model; it is not the same concept as 95% power. Power is the probability that the planned test rejects its null hypothesis when the specified effect and assumptions hold. Selecting “95% confidence and 5% error” does not automatically give an 80%-powered comparison. Make the purpose and intended analysis explicit before selecting settings.

Qualitative and mixed-methods studies

Interview, ethnographic, and case-study research often prioritises information depth, participant diversity, theoretical sampling, and iterative analysis rather than a numerical margin of error. There is no universal number of interviews that guarantees saturation or quality. Explain who will be interviewed, why those participants are information-rich for the question, how analysis will proceed, and what evidence will indicate sufficient coverage. In mixed-methods research, justify the quantitative and qualitative components separately, then explain their integration.

How Do Population Size, Clustering, and Nonresponse Change the Target?

The initial calculation usually estimates the number of analysable units under defined assumptions. Real recruitment often needs additional adjustments. Apply only adjustments that match the design and state the order clearly.

Finite-population correction for a small sampling frame

For simple random sampling without replacement from a known finite population of N units, one common correction is n = n₀ / [1 + (n₀ − 1)/N]. A survey of 1,200 eligible students with n₀ = 384.16 gives n ≈ 291.2, so the target is 292 completed responses. This is a design-based correction, not a reason to reduce a nonprobability convenience sample. For complex designs, obtain the appropriate design-specific calculation.

Design effects for clustered sampling

When units are recruited within schools, hospitals, villages, or other groups, responses may be correlated. An approximate equal-cluster-size design effect is DEFF ≈ 1 + (m − 1)ρ, where m is mean cluster size and ρ is the intraclass correlation coefficient. With m = 20 and ρ = 0.02, DEFF = 1.38; a simple-random-sample target of 385 becomes approximately 532 analysable observations after upward rounding. The number of clusters and variation in cluster sizes also matter; specialised methods may be necessary.

Allowances for incomplete responses or dropouts

If a project needs n valid completed observations and expects response or retention rate r, a common planning calculation is number to invite or enrol = ceiling(n/r). Thus a corrected target of 292 completed surveys and an anticipated 80% response rate means approaching at least 365 eligible individuals. For a comparison needing 98 participants per arm and expecting 15% missing primary outcomes, 98/0.85 = 115.30, so plan 116 per arm.

Do not confuse expected nonresponse with the number of complete cases, and do not assume that inflating a target removes nonresponse bias. Missingness can change the composition of the sample as well as its size. Plans for follow-up, data-quality checks, and statistical handling of missing outcomes belong alongside the sample-size calculation.

Strata, subgroups, and unequal allocation

A survey may need adequately precise estimates within regions, genders, or age bands rather than only for the full population. Each essential subgroup may require its own precision assessment. Similarly, unequal treatment allocation changes power calculations even if the total sample size remains similar. Pre-specify which comparisons are primary; a single “overall” formula may not support every planned subgroup claim.

How Should You Evaluate and Report a Sample-Size Decision?

A defensible justification gives enough detail for another researcher to reconstruct the result, assess the assumptions, and understand what was actually achieved. That means reporting more than the software name or a final participant count.

Illustrative methods wording for a survey: “The target number of completed questionnaires was calculated for a single population proportion using 95% confidence, an assumed prevalence of 50%, and an absolute margin of error of five percentage points. The large-population result was 384.16, rounded to 385. Because the eligible population comprised 1,200 students, a finite-population correction produced a minimum target of 292 completed questionnaires. Anticipating an 80% response rate, we planned to invite at least 365 eligible students.” This description is only valid if the stated sampling assumptions match the actual design.

For a controlled comparison, write the primary outcome, target difference and rationale, assumed variability, test, alpha, power, per-arm target, and missing-outcome allowance. For complex studies, document simulations, software options, and sensitivity analyses. A manuscript should also give the actual recruitment and analysis counts even when they differ from the plan.

What to do when assumptions are uncertain

Present a small range of plausible scenarios, such as p = 0.30, 0.40, and 0.50 for a survey or several meaningful effect sizes for a group comparison. Explain the selected scenario rather than choosing whichever gives the smallest sample. Where assumptions are especially uncertain, obtain pilot or feasibility evidence, consider adaptive planning that is appropriately pre-specified, or seek specialist methodological review.

What not to report

A generic sentence such as “The sample size was calculated at a 95% confidence level” is incomplete; it omits the precision target and sampling assumptions. Likewise, a post hoc power calculation based only on an observed nonsignificant result is not an adequate substitute for reporting the prospective plan and estimate uncertainty. Give the design decisions readers actually need to evaluate the evidence.

Ethical Research Planning and Author Responsibility

Sample size has ethical implications because insufficient data may waste participant contributions and excessive recruitment can impose unnecessary burden. However, scientific value also depends on valid sampling, reliable measurement, fair recruitment, privacy safeguards, and an analysis plan that corresponds to the approved protocol.

When outside experts assist, separate methodological advice from editorial support. A statistician may advise on the underlying mathematical specification; a supervisor or ethics committee may approve the protocol; and an editor may improve the clarity, terminology, tables, and referencing of the account. The named investigators must verify the assumptions, avoid fabricated calculations, acknowledge support where policies require, and describe substantive changes honestly.

Good academic editing can make a dense methods explanation easier to follow without altering findings or claiming methodological validation. If an editor spots a contradiction—for example, 150 subjects in the abstract but 180 in the methods—that should be flagged for the authors rather than silently “fixed” without evidence.

Common Mistakes in the Determination of Sample Size

The most costly mistakes arise when a plausible-looking number is detached from the actual research design. Check these errors before finalising your methodology chapter or submission.

• Starting with a fixed percentage of the population: a “10% rule” generally lacks a sound basis for precision or power calculations.
• Using confidence and margin of error for a hypothesis test: estimation and effect detection need different planning logic.
• Choosing an optimistic effect size: larger assumed effects reduce calculated n, but may leave the study unable to detect the smaller effect that matters.
• Ignoring sampling design: cluster, stratified, or paired data may invalidate a simple independent-observation calculation.
• Forgetting to specify per-arm or total n: a planned 100 per group means 200 overall, before dropout adjustment.
• Applying a finite-population correction to convenience recruitment: the correction assumes an appropriate probability-sampling framework.
• Adding a flat 20% instead of dividing by retention: for 80% expected retention, n/0.80 is the suitable arithmetic, not n × 1.20.
• Ignoring subgroups and missing data: an adequate overall n may not support meaningful precision within small categories.
• Treating qualitative saturation as an automatic interview count: adequacy depends on research purpose, heterogeneity, data quality, and analysis.
• Reporting only the final number: readers need assumptions, calculations, targets, and achieved sample counts.

A practical cross-check is to ask a colleague who has not seen your calculator output to reproduce the result from your methods paragraph. If they cannot, add the missing assumptions rather than extra promotional or methodological jargon.

Worked Examples: Applying the Sample-Size Approach

The following scenarios show how study objectives lead to different targets. All situations and numerical inputs are illustrative, not accounts of actual Contentxprtz clients.

Example 1: A student survey with a known population

Situation: A postgraduate student wants to estimate the proportion of 1,200 eligible students who use a campus research portal. The student initially proposes collecting “10% of the population,” or 120 responses, without showing the expected precision. The correct starting point is a proportion estimate with 95% confidence, p = 0.50, and ±5 percentage points absolute precision under simple random sampling.

Calculation: n₀ = 384.16, rounded up to 385. Because the known frame contains N = 1,200 students, the finite-population calculation yields approximately 291.2, rounded up to 292 complete questionnaires. With an 80% anticipated response rate, the student should plan to invite at least 292/0.80 = 365 eligible students. The invitation strategy must still be capable of producing an appropriate probability sample.

What expert support adds: A supervisor or statistician can verify the frame and design assumptions; ethical editing help can make the methodology and recruitment distinction clear. Neither service can turn a voluntary-response sample into a probability sample after the fact.

Example 2: A PhD scholar comparing two teaching methods

Situation: A doctoral researcher will compare independent groups using a continuous examination score. Instead of copying a prevalence formula, the researcher defines a difference of six points as meaningful, assumes a common standard deviation of 15 points, chooses two-sided α = 0.05, and aims for 80% power. Equal allocation and a basic independent-group comparison are assumed.

Calculation: The normal-approximation result is 98 complete outcomes per arm, 196 overall. Allowing for 15% missing primary outcomes means planning for 116 participants in each arm, 232 total. A test-specific software check should confirm the assumptions. If recruitment occurs by class rather than student, cluster randomisation methods are required and may increase the target.

What expert support adds: The methodologist checks test selection and any clustering. An editor checks that the hypothesis, sample-size paragraph, and analysis methods all refer to the same comparison, without changing the scholarly decisions.

Example 3: A community health team sampling whole villages

Situation: A research team wants an approximate prevalence estimate and first obtains a simple-random-sample target of 385. It plans to sample villagers in clusters of roughly 20 per village, with an anticipated intraclass correlation of 0.02. Applying the original 385 without adjustment would ignore within-village similarity.

Calculation: DEFF ≈ 1 + (20 − 1) × 0.02 = 1.38. Therefore, 385 × 1.38 = 531.3, rounded up to 532 complete observations. At 80% expected completion, the approximate recruitment target becomes 532/0.80 = 665. The actual number of clusters, unequal cluster sizes, analysis weighting, and operational recruitment constraints need a design-aware review.

What expert support adds: Statistical guidance can refine the design effect and cluster counts. Publication-oriented editing can help the team distinguish total recruited people, sampled clusters, and analysable observations.

Example 4: A qualitative dissertation on doctoral supervision

Situation: A student proposes “30 interviews because software said so,” even though the dissertation uses interpretive interviews across diverse academic disciplines. A proportion formula and a numerical power target do not answer how much qualitative evidence is enough for that question.

Better approach: The student defines selection criteria, seeks variation across disciplines and experiences, explains interview depth and analytic strategy, and plans to assess information adequacy as data collection and analysis progress. Any provisional target should be a transparent planning estimate, not a universal saturation guarantee.

What expert support adds: A qualitative methodologist can refine the sampling rationale; editorial support can clarify the distinction between purposive selection and statistical representativeness while preserving the student's interpretation.

Sample-Size Planning and Methodology Writing Checklist

Before a proposal meeting, ethics submission, or journal manuscript, confirm that the design and numerical explanation tell the same story. This checklist helps identify omissions before they become difficult to correct.

• Have I defined the target population, sampling frame, eligibility criteria, and unit of analysis?
• Is one main outcome or research objective explicitly linked to the sample-size calculation?
• Have I selected precision, power, a model-specific method, or qualitative information adequacy for the correct reason?
• Do I have credible sources for p, σ, meaningful difference, or other key inputs?
• Are confidence, precision, alpha, power, allocation, and time point expressed unambiguously where relevant?
• Have I checked the correct total or per-group count and always rounded upward?
• Do finite population, strata, cluster sampling, repeated measures, or multiple endpoints require adjustments?
• Did I account for the expected proportion of incomplete, unusable, or missing outcomes?
• Does the proposed sampling method actually support the uncertainty claims I intend to make?
• Have I run at least one realistic sensitivity analysis for uncertain inputs?
• Can a reader reproduce the calculation using the information in my methods section?
• Will I report enrolled, completed, excluded, and analysed observations separately?

For thesis writers, check that the methods section, proposal tables, ethics documents, recruitment script, and data analysis plan agree. Manuscript assessment and research-language review can identify inconsistencies for author correction, but they should not be presented as substitutes for statistical validation.

How Contentxprtz Can Help You Explain the Method Clearly

Even when the statistical design is sound, a rushed proposal can leave essential assumptions scattered across chapters. Contentxprtz can help researchers improve the structure and readability of the methodology section, clarify technical terminology, organise tables, check reference consistency, and flag discrepancies between objectives, sample planning, and reported participant numbers.

Researchers can explore research support services when they need help organising an existing methodology draft or preparing a clear, ethical account of their work. For manuscript language and presentation, academic editing support may be suitable. If the project needs a new statistical power analysis or complex sampling design, request an appropriately qualified statistical methodologist rather than treating editorial review as statistical certification.

Need help presenting your approved sample-size rationale clearly? Explore research paper writing support for ethical assistance with research communication, structure, and clarity. The study authors should provide verified assumptions, interpret the design, and retain full responsibility for submission.

Summary: Determination of Sample Size through the Approach

Sample-size determination begins with purpose. Use precision-based planning to estimate a population quantity; use prospective power analysis to detect a specified effect; use an explicitly justified qualitative plan when information depth is the core objective. Identify the unit of analysis and sampling design, choose defensible inputs, perform the calculation, and adjust carefully for clustering, finite populations when applicable, and missing data.

A survey with a 95% confidence level and ±5 percentage points precision might require 385 complete observations before corrections, but that number is not a universal rule. The right target changes with the outcome, population, design, effect, variation, response rate, and planned analysis. A defensible methods section reports the assumptions and equations, explains their sources, and distinguishes participants invited, enrolled, and analysed.

Conclusion: Let the Research Question Determine the Sample

The most defensible sample is not the largest affordable number or the most familiar figure from a calculator. It is a target derived from a clear research question, valid design assumptions, and the analysis the researcher intends to perform. Precision and power answer different questions; recruitment losses, clustering, and finite-population sampling can change the initial result; qualitative studies may need a different logic entirely.

Many straightforward survey and mean-estimation problems can be planned using self-service formulas and carefully checked software. When a study has difficult assumptions, specialised tests, multiple levels of sampling, or high ethical stakes, expert methodological review is the safer next step. Contentxprtz can help researchers communicate author-verified sample-size decisions with clear methodology writing, appropriate language, coherent tables, and consistent citations. The investigators remain responsible for calculations, ethics, analysis, and final submission.

“At Contentxprtz, we don’t just edit; we help ideas reach their fullest potential.”