Sampling and Fieldwork Management

Introduction

Business research rarely permits an entrepreneur or analyst to study every member of a market. A population may contain millions of consumers, thousands of firms, or a constantly changing group of online users. Sampling provides a defensible way to obtain evidence from a manageable subset while making carefully qualified statements about the wider population.

Sampling is not simply the act of selecting respondents. It is a chain of decisions involving the target population, sampling frame, sampling design, sample size, recruitment, data collection, quality control, and analysis. Errors at any stage can reduce the credibility of the findings. A large sample selected from a poor frame may be less useful than a smaller, carefully designed sample.

For entrepreneurs, sampling competence is especially important. Market research may guide product design, pricing, location, customer segmentation, investor presentations, and resource allocation. Decisions based on unrepresentative or poorly managed fieldwork can create costly false confidence.

Learning Objectives

By the end of this chapter, students should be able to:

• define a target population, accessible population, sampling frame, sampling unit, and sample;
• distinguish probability from non-probability sampling;
• select appropriate sampling techniques for different research purposes;
• estimate sample size and explain the assumptions behind common formulas;
• understand sampling error, nonsampling error, statistical power, and representativeness;
• design recruitment and fieldwork procedures;
• monitor nonresponse, interviewer performance, data quality, and ethical compliance; and
• evaluate the strengths and limitations of sampling decisions in business research.

1. The Architecture of Sampling

1.1 Population and target population

A population is the complete set of units relevant to a research question. A unit may be a person, household, business, transaction, store, advertisement, or document. The target population is the population to which the researcher intends to generalize.

A target population must be defined by content, geographic scope, and time. “Small-business owners” is incomplete because it does not specify the type of business, location, legal status, or reference period. A stronger definition might be: “All independently owned retail businesses employing fewer than 50 people and operating in Accra during January–June 2026.”

The accessible population is the portion of the target population that can realistically be reached. Differences between the target and accessible populations create coverage limitations. For example, a study of all urban consumers may use an online panel that excludes people without reliable internet access.

1.2 Sampling frame

A sampling frame is the operational list or procedure used to identify members of the accessible population. Examples include a customer database, business registry, electoral register, professional association list, school enrollment file, or geographically defined household listing.

A frame should be:

• complete, covering eligible units;
• current, reflecting recent entries, exits, and contact details;
• nonduplicated, so that units do not have unequal selection chances merely because they appear repeatedly; and
• eligible, excluding units outside the target definition.

Frame problems include undercoverage, overcoverage, duplication, outdated information, and incorrect contact details. A frame is not necessarily a neutral mirror of the population. For instance, a registry may list formal firms but omit informal enterprises.

1.3 Sampling units and stages

The element is the basic unit about which information is collected, such as an individual customer. A sampling unit is the unit selected at a particular stage. In a multistage design, a researcher might select districts, then stores, then customers. The selected unit at the first stage is a primary sampling unit; later selections form subsequent stages.

The distinction matters because the unit of selection may differ from the unit of analysis. A study may select households but analyze individual adults. Researchers must specify who is eligible within a household and how one person is chosen, otherwise interviewers may select the most available or cooperative person.

2. Probability and Non-Probability Sampling

2.1 Probability sampling

In probability sampling, every eligible unit has a known, non-zero probability of selection. Selection is governed by a random mechanism rather than interviewer preference. Known selection probabilities permit estimation of sampling error and, when the design is sound, support statistical generalization.

Common probability designs are summarized below.

In simple random sampling, each sample of a given size has equal probability of selection. A random-number generator, rather than a convenient list position, should determine selection.

In systematic sampling, the interval is approximately:

where N is the number of eligible units in the frame and n is the desired sample size. After selecting a random starting position between 1 and k, every kth unit is selected. This method is efficient, but the researcher must inspect the frame for periodic patterns.

Stratified sampling divides the population into mutually exclusive and collectively exhaustive strata, such as region, firm size, or customer type. Proportionate allocation selects from each stratum in proportion to its population size. Disproportionate allocation may deliberately oversample a small but strategically important group; analysis then usually requires weights.

Cluster sampling selects naturally occurring groups. It is economical when a population is geographically dispersed. However, people within the same cluster often resemble one another. This reduces the amount of independent information supplied by each additional respondent.

A multistage design combines these procedures. For example, a national entrepreneurship survey may randomly select regions, then municipalities, then firms, and finally an owner or manager within each firm.

2.2 Non-probability sampling

In non-probability sampling, selection probabilities are unknown or unequal in a way that cannot be fully calculated. It is often appropriate for exploratory research, hard-to-reach populations, qualitative studies, rapid concept testing, or situations without a usable frame.

Important forms include:

• Convenience sampling: selecting accessible respondents, such as visitors to a campus or followers of a social-media page.
• Purposive or judgment sampling: selecting cases because they possess relevant characteristics, such as experienced founders or specialist investors.
• Quota sampling: filling predetermined subgroup quotas without random selection within each subgroup.
• Snowball sampling: asking participants to refer other eligible participants; useful for hidden or networked populations.
• Consecutive sampling: recruiting every eligible case available during a defined period until the target is reached.

Non-probability samples can produce valuable insight, but a sample percentage cannot automatically be treated as a population estimate. A quota sample with the same age distribution as a population is not necessarily representative on unmeasured variables such as political interest, income, or brand loyalty. Researchers should describe the recruitment process transparently and avoid overstating generalizability.

3. Sampling Error, Bias, and Representativeness

Sampling error is the difference between a sample estimate and the corresponding population value that arises because only part of the population is observed. It is a property of sampling variability, not evidence that the researcher made a procedural mistake. Probability theory permits its estimation under an appropriate design.

Nonsampling error includes coverage error, nonresponse error, measurement error, processing error, and interviewer error. Increasing the sample size does not necessarily solve these problems. Indeed, a larger biased sample can produce a highly precise estimate of the wrong quantity.

A useful framework distinguishes four questions:

1. Coverage: Who could enter the sample, and who could not?
2. Selection: Were eligible units selected according to the stated design?
3. Response: Did selected units participate, and do respondents differ from nonrespondents?
4. Measurement: Did the questionnaire and fieldwork produce accurate answers?

Representativeness is not a single visual property. It concerns whether the sample supports the intended inference for the variables and population of interest. A sample may represent customers by age but not by purchasing frequency. Researchers should compare sample distributions with credible population benchmarks where available, while recognizing that matching a few margins does not eliminate all bias.

4. Estimating Sample Size

Sample size depends on the research objective, design, desired precision, confidence level, expected variability, population size, response rate, and available resources. There is no universally correct sample size.

4.1 Estimating a proportion

For a simple random sample intended to estimate a population proportion, the familiar initial sample-size formula is:

Here, n₀ is the initial required sample size, z is the standard-normal critical value associated with the desired confidence level, p is the anticipated population proportion, and e is the maximum tolerated margin of error expressed as a proportion. The formula assumes a probability sample, a sufficiently large population, independent selection, and a simple random design.

If no prior estimate of p is available, p = 0.5 is commonly used because it produces the largest product p(1−p) and therefore a conservative sample size.

For a 95% confidence level, z ≈ 1.96. If p = 0.5 and e = 0.05:

The result is rounded upward to 385 respondents under the stated idealized assumptions. This does not mean that 385 completed interviews guarantee an unbiased estimate.

For a finite population of size N, the finite population correction can be applied:

This correction is useful when the sample is a substantial fraction of the population, commonly when sampling fraction is above approximately 5%, although the precise decision depends on the design and objective.

4.2 Adjusting for design effect and nonresponse

Clustered or complex designs often require a design effect, represented by DEFF. A practical planning equation is:

where n_ideal is the number of completed interviews required under simple random sampling, DEFF reflects efficiency loss or gain from the actual design, and r is the anticipated response rate expressed as a proportion. This is a planning approximation; final analysis may require design weights and specialized variance estimation.

For example, if 385 completed interviews are needed, the anticipated design effect is 1.5, and the expected response rate is 70%:

The researcher should therefore approach approximately 825 eligible units to obtain about 578 completed cases? No: this interpretation would be incorrect because the equation as written estimates contacts needed to obtain the design-adjusted completed sample. The required completed sample is 385 × 1.5 = 577.5, and dividing by 0.70 gives approximately 825 sampled or contacted eligible units. This example illustrates why researchers must distinguish the ideal simple-random completed sample from the design-adjusted completed sample.

4.3 Power and hypothesis testing

Statistical power is the probability of detecting an effect of a specified size when that effect truly exists, under a defined analysis and significance threshold. It is commonly expressed as:

where β is the probability of a Type II error, failing to reject a null hypothesis when a specified alternative is true. Power depends on sample size, effect size, outcome variability, significance level, and study design. A conventional target such as 80% is a planning convention, not a law.

Power analysis should be conducted before data collection. Researchers should specify the primary outcome, minimum practically important effect, significance level α, desired power, and statistical test. A statistically significant result may be commercially trivial, while a nonsignificant result from a small sample may be inconclusive rather than evidence of no effect.

Precision-based sample size and power-based sample size answer different questions. Precision asks how narrow an estimate or confidence interval should be. Power asks how likely a planned test is to detect a specified effect. Researchers should not select a sample size solely by applying a universal rule such as “10% of the population.”

5. Recruitment and Fieldwork Management

Fieldwork converts a design into observed data. A fieldwork plan should specify eligibility, contact procedures, recruitment channels, consent, interviewer training, call-back rules, incentives, supervision, data security, and escalation procedures.

5.1 Eligibility and screening

Screening questions should be short, neutral, and logically ordered. Eligibility criteria must be stated before recruitment begins. For example, a study of business software might require that the respondent be involved in purchase decisions and that the firm has used the software for at least six months. Screening should not be used to manufacture a preferred result by excluding inconvenient opinions.

A screening script should define how ambiguous cases are handled. “Decision maker” may include an owner, procurement manager, or operational lead; the field protocol must identify which roles qualify.

5.2 Recruitment modes

Telephone, face-to-face, web, mobile, email, social media, and mixed-mode recruitment each create different coverage and measurement effects. Online recruitment may be fast and inexpensive but can overrepresent digitally engaged users. Face-to-face interviewing can reach people with limited internet access but requires stronger safety and interviewer controls.

Incentives can improve participation, but they should be proportionate and clearly disclosed. Excessive incentives may attract ineligible or professional respondents. Recruitment messages should state the sponsor where ethically appropriate, study purpose in non-leading language, duration, eligibility, confidentiality arrangements, and voluntary nature.

5.3 Interviewer training and standardization

Training should cover the research purpose without encouraging interviewer expectations, exact question wording, probing rules, consent, privacy, device use, safety, and refusal conversion. Interviewers should not paraphrase questions inconsistently or imply that one answer is preferred.

Standardization reduces interviewer effects, but qualitative or exploratory studies may intentionally permit flexible probing. The choice must match the research design. Audio recording, where lawful and consented, can support quality review; it should never be introduced secretly.

5.4 Monitoring fieldwork

A live fieldwork dashboard can track invitations, contacts, eligible cases, completes, refusals, break-offs, callbacks, quota or stratum progress, interview duration, and item nonresponse. Monitoring should occur by interviewer, mode, geography, and subgroup—not merely in aggregate.

Useful quality checks include:

• duplicate device, telephone, email, or respondent identifiers;
• implausibly short completion times;
• straight-lining in matrix questions;
• contradictory answers;
• excessive missing data;
• unusual response patterns by interviewer;
• failed attention checks where justified; and
• verification callbacks to a legitimate subsample.

Quality flags should trigger investigation, not automatic deletion. A short interview may be legitimate for an uncomplicated questionnaire, and an unusual answer may be substantively correct.

6. Managing Nonresponse

Nonresponse occurs when sampled units do not provide usable data. Unit nonresponse means an entire selected unit does not participate. Item nonresponse means a participant omits particular questions.

The crude response rate can be described as:

In practice, professional standards distinguish completed, partial, ineligible, unknown-eligibility, and refusal cases. The definition should be reported because different classifications produce different rates.

Nonresponse creates bias when respondents and nonrespondents differ in relevant ways. A high response rate does not prove absence of bias, and a lower rate does not automatically prove serious bias. Researchers should use multiple strategies:

• make several contact attempts at different times;
• offer suitable modes or languages;
• keep the questionnaire concise;
• communicate legitimacy and confidentiality;
• use trained interviewers and respectful refusal procedures;
• provide ethical incentives; and
• compare early and late respondents or sample estimates with external benchmarks.

Weighting adjustment may compensate for unequal selection probabilities or differential response. A base weight is often related to the inverse of a unit’s selection probability. Nonresponse and calibration adjustments may then align weighted sample totals with known population totals. Weighting can reduce some biases but may increase variance and cannot repair missing variables or fundamentally defective coverage.

7. Ethical, Legal, and Governance Requirements

Participants must receive sufficient information to make a voluntary, informed decision. Consent should cover the study purpose, procedures, expected duration, risks, benefits, confidentiality, data use, withdrawal rights, and contact information where applicable.

Researchers should collect only data necessary for the research purpose, restrict access, secure devices and files, establish retention periods, and remove direct identifiers when feasible. Sensitive information requires stronger safeguards. Applicable privacy and data-protection law must be followed, including rules concerning lawful processing, children, cross-border transfer, and recorded communications.

Ethical fieldwork prohibits deception unless justified by a recognized research protocol and approved through appropriate review. Participants must not be pressured by employers, teachers, lenders, or community leaders. Incentives should not be so large that they undermine voluntary participation.

Research teams should also govern vendors. Contracts should specify sampling procedures, data ownership, confidentiality, subcontracting, quality standards, breach notification, and audit rights. AI-generated or automated recruitment can introduce hidden exclusion, duplicate participants, profiling, or discriminatory targeting. Automated tools should be validated, monitored, and subject to human oversight.

8. An Illustrative Case: Sampling for a Digital Payments Venture

Background

This fictional case concerns PayLink, an early-stage venture considering a digital payment service for independent retailers in three cities. The founders want to estimate interest, identify barriers, and compare small shops with medium-sized shops.

Problem

The founders initially propose an online survey distributed through their social-media followers. This would be inexpensive but would exclude retailers with limited digital engagement and would overrepresent people already interested in PayLink. The team also has an incomplete list of businesses from a commercial directory.

Analysis

The researchers define the target population as independently owned retail businesses with one to 49 employees, operating in the three cities during the study period. They combine a municipal business register, a market-association list, and a geographic listing exercise. Duplicates are removed, and informal retailers identified through the listing are added where legally and operationally feasible.

They stratify by city and firm size because both are expected to influence adoption. Within each stratum, businesses are selected systematically after a random start. The owner or the employee principally responsible for payment decisions is screened for eligibility.

The desired simple-random completed sample is 385 for a broad proportion estimate at approximately 95% confidence and 5% margin of error, assuming maximum variability. Because the multistage field plan is expected to have a design effect of 1.3 and the anticipated response rate is 65%, the team plans approximately:

eligible sampled units or contact attempts, subject to the exact operational definition of response and design.

Approach and solution

The team uses face-to-face recruitment for markets and telephone follow-up for listed businesses. Each selected business receives up to four contact attempts at varied times. Interviewers use a standardized script, while supervisors review duration, missingness, and geographic progress daily. A small pilot identifies ambiguous questions about transaction fees and distinguishes awareness from actual use.

The researchers document refusals and unknown eligibility, compare response rates across strata, and apply design weights in analysis. They report that the sample is a probability-based sample of the constructed frame, not automatically a perfect representation of all retailers in the cities.

Outcome

The survey finds that stated interest is high among medium-sized retailers but that perceived fraud risk is the leading adoption barrier among small shops. The founders revise the product to include transaction alerts and a fraud-reimbursement policy. They avoid claiming that the results represent all informal retailers because coverage of that group remains incomplete.

Learning points

The case demonstrates that sampling begins with population definition and frame construction. Stratification can protect important subgroups, while mixed-mode fieldwork can reduce coverage problems. Sample-size arithmetic cannot compensate for poor eligibility rules, weak recruitment, or inadequate measurement. Transparent limitations improve the credibility of entrepreneurial decisions.

9. Contemporary Developments and Best Practices

Digital panels, mobile sampling, geolocation, administrative data linkage, and machine-assisted quality monitoring are changing fieldwork. These tools can reduce cost and improve speed, but they also create risks involving opaque recruitment, panel conditioning, privacy, bot responses, and algorithmic exclusion.

Best practice is to:

1. define the estimand and target population before choosing a sample;
2. document frame construction, exclusions, duplicates, and unknown cases;
3. match probability or non-probability methods to the inference intended;
4. conduct a pilot and revise unclear instruments;
5. pre-specify sample-size assumptions and primary outcomes;
6. monitor fieldwork continuously by subgroup and mode;
7. retain an auditable record of contact attempts and dispositions;
8. distinguish sampling error from nonsampling error;
9. disclose weighting, missing data, deviations, and limitations; and
10. protect participants throughout recruitment, collection, storage, analysis, and reporting.

Chapter Summary

Sampling is a structured process linking a target population to an observable sample. A sampling frame operationalizes the accessible population, while sampling units define what is selected at each stage. Probability sampling uses known non-zero selection probabilities and supports estimation of sampling error; non-probability sampling is often useful for exploration and hard-to-reach groups but requires cautious claims about generalization.

Sample size depends on precision or statistical power, variability, confidence or significance requirements, population size, design effect, and expected response. The proportion formula using n₀ = (z² × p × (1 − p)) ÷ e² is a simple-random starting point, not a universal rule. Fieldwork quality depends on ethical recruitment, clear screening, interviewer standardization, nonresponse management, monitoring, verification, and secure data governance.

Representativeness cannot be inferred from sample size alone. Coverage, selection, response, and measurement must all be considered. Sound entrepreneurial research therefore combines appropriate design, transparent limitations, careful execution, and disciplined interpretation.

Key Terms

Target population: The complete population to which the researcher intends to generalize.

Accessible population: The portion of the target population that can realistically be reached.

Sampling frame: The list or operational procedure identifying eligible units for selection.

Sampling unit: The unit selected at a particular stage of sampling.

Element: The basic unit about which information is collected.

Probability sampling: Sampling in which eligible units have known, non-zero selection probabilities.

Non-probability sampling: Sampling in which selection probabilities are unknown or not fully calculable.

Simple random sampling: Random selection giving each eligible unit an equal selection opportunity.

Stratified sampling: Sampling separately within defined population subgroups.

Cluster sampling: Sampling naturally occurring groups rather than isolated elements.

Multistage sampling: Sampling conducted through two or more sequential levels.

Sampling error: Difference from observing a sample rather than the full population.

Nonsampling error: Error caused by coverage, response, measurement, processing, or fieldwork problems.

Nonresponse: Failure of selected units to provide usable data.

Design effect: The relative efficiency of a complex design compared with simple random sampling.

Statistical power: Probability of detecting a specified effect when it exists.

Weighting: Adjusting observations to reflect selection probabilities or population benchmarks.

Quota sampling: Filling subgroup targets without random selection within groups.

Fieldwork: The operational process of recruiting participants and collecting data.

Coverage error: Bias arising when the frame excludes or incorrectly includes population units.

Review Questions

Short Answer Questions

1. What is the difference between a target population and an accessible population?
2. Why can an outdated sampling frame create coverage error?
3. Distinguish probability sampling from non-probability sampling.
4. What is the purpose of a design effect in sample-size planning?
5. Name two strategies for reducing nonresponse.

Long Answer Questions

1. Compare simple random, stratified, cluster, and multistage sampling, including one appropriate business application for each.
2. Explain the assumptions and limitations of the basic sample-size formula for estimating a population proportion.
3. Discuss why a large sample may still fail to be representative.
4. Design a fieldwork management plan for a survey of small-business owners.
5. Explain how ethical, privacy, and data-security obligations shape recruitment and fieldwork.

Analytical / Application Questions

1. A researcher plans a simple-random survey requiring 385 completed responses. The design effect is estimated at 1.4 and the response rate at 60%. Calculate the approximate number of eligible units to approach and explain the assumptions.
2. An online customer survey matches the population’s age and gender distributions but receives responses mainly from frequent buyers. Evaluate the representativeness of the sample and propose two improvements.
3. A national study selects regions, then stores, then customers within stores. Identify the sampling stages, explain one likely source of increased variance, and recommend a fieldwork quality-control procedure.

Multiple Choice Questions with Answers

Question 1

Which statement best defines a sampling frame?

A. The complete theoretical population of interest

B. The list or procedure used to identify eligible units for selection

C. The number of completed interviews required for analysis

D. The questionnaire used to screen respondents

Correct Answer: B. The list or procedure used to identify eligible units for selection

Question 2

Which technique gives every eligible unit a known, non-zero probability of selection?

A. Convenience sampling

B. Snowball sampling

C. Probability sampling

D. Judgment sampling

Correct Answer: C. Probability sampling

Question 3

A researcher divides firms into micro, small, and medium categories, then randomly samples within each category. What design is being used?

A. Cluster sampling

B. Stratified sampling

C. Convenience sampling

D. Snowball sampling

Correct Answer: B. Stratified sampling

Question 4

Using z = 1.96, p = 0.5, and e = 0.05, the approximate initial sample size for a proportion is:

A. 96

B. 196

C. 385

D. 960

Correct Answer: C. 385

Question 5

In the formula Power = 1 − β, what does β represent?

A. The probability of a Type II error

B. The population mean

C. The sampling interval

D. The response rate

Correct Answer: A. The probability of a Type II error

Question 6

Which problem occurs when a business registry excludes informal enterprises that belong to the target population?

A. Processing error

B. Coverage error

C. Randomization error

D. Recall improvement

Correct Answer: B. Coverage error

Question 7

A researcher selects every 20th business after choosing a random starting point. This is:

A. Systematic sampling

B. Quota sampling

C. Purposive sampling

D. Cluster sampling

Correct Answer: A. Systematic sampling

Question 8

If 385 completed interviews are required, the design effect is 1.5, and the response rate is 70%, approximately how many eligible units should be approached?

A. 385

B. 550

C. 578

D. 825

Correct Answer: D. 825

Question 9

Which statement about nonresponse is most accurate?

A. A high response rate guarantees an unbiased sample

B. Nonresponse matters only in qualitative research

C. Nonresponse can bias results when respondents differ from nonrespondents

D. Increasing sample size always eliminates nonresponse bias

Correct Answer: C. Nonresponse can bias results when respondents differ from nonrespondents

Question 10

Why can cluster sampling produce larger sampling error than simple random sampling at the same nominal sample size?

A. Clusters always contain ineligible units

B. Units within a cluster may be more similar to one another

C. Cluster sampling never uses random selection

D. Clusters eliminate the need for weighting

Correct Answer: B. Units within a cluster may be more similar to one another