Sampling, Fieldwork and Research Quality

Introduction

Market research rarely studies every member of a market. A firm may want to understand thousands of potential customers, distributors, employees, or competitors, but investigating the entire population is usually too costly, slow, or impractical. Sampling provides a disciplined way to select a smaller group whose information can support conclusions about a larger target population.

Sampling is not simply a technical exercise in choosing a number of respondents. It involves defining the population, identifying or constructing a sampling frame, selecting cases, determining sample size, managing data collection, and evaluating the credibility of the resulting evidence. A poorly selected large sample can be less useful than a smaller, carefully designed sample. Conversely, a probability sample can still produce misleading results if non-response, coverage problems, measurement error, or fieldwork failures are ignored.

For entrepreneurs, these issues have direct commercial consequences. Decisions about product positioning, pricing, market entry, customer experience, and investment depend on the quality of evidence. This chapter develops the concepts and practical procedures required to design samples, manage fieldwork, estimate uncertainty, and judge research quality.

Learning Objectives

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

1. Define a target population, sampling frame, sampling unit, and sample.
2. Distinguish probability from non-probability sampling.
3. Select appropriate sampling designs for different research purposes.
4. Estimate sample size for proportions and means, subject to practical constraints.
5. Explain sampling error, non-response, coverage error, and representativeness.
6. Design and control fieldwork using professional and ethical procedures.
7. Evaluate the reliability, validity, and overall quality of market research.
8. Apply these principles to an entrepreneurial market-research decision.

1. The Logic of Sampling

1.1 Population, sample and inference

The target population is the complete set of units about which a study seeks to make conclusions. For a subscription service, it might be all adults in a defined city who use online payment services. The sampling unit is the basic element selected at a stage of sampling, such as a person, household, business, store, or geographic area.

A sampling frame is the operational list or system from which the sample is selected. Examples include a customer database, a business directory, a list of registered firms, or a database of residential addresses. The sampling frame may not perfectly match the target population. Some eligible units may be missing, duplicated, outdated, or incorrectly classified.

The sample is the set of units actually selected for the study. Researchers use sample statistics—such as a sample proportion or mean—to estimate population parameters. For example, the sample proportion of users who prefer a new product concept estimates the corresponding population proportion.

Sampling involves two broad forms of uncertainty:

• Sampling uncertainty arises because only part of the population is observed.
• Non-sampling error arises from coverage problems, non-response, poor measurement, interviewer effects, processing mistakes, and other failures.

Increasing sample size generally reduces sampling uncertainty, but it does not automatically correct systematic non-sampling errors.

1.2 Census and sample survey

A census attempts to collect information from every unit in the target population. It may be appropriate when the population is small, legal completeness is required, or the cost of missing a unit is extremely high. A sample survey is preferable when the population is large, dispersed, changing rapidly, or when resources are limited.

A census is not necessarily error-free. A national census may suffer from non-response, inaccurate records, or measurement problems. Therefore, the relevant comparison is not “sample versus no error,” but the expected quality, cost, speed, and usefulness of each design.

2. Probability and Non-Probability Sampling

2.1 Probability sampling

In probability sampling, every eligible population unit has a known, non-zero probability of selection. The probabilities need not be equal, but they must be known or estimable. This property permits design-based estimation of sampling error and supports statistical generalisation, provided the target population and fieldwork are appropriately managed.

The major probability designs are as follows.

#### Simple random sampling

Under simple random sampling, every possible sample of a given size has the same probability of selection. A random-number generator can select units from a complete sampling frame.

Its strengths are conceptual simplicity and straightforward estimation. Its weaknesses include the need for a good frame, possible high travel or contact costs, and poor representation of small but important subgroups by chance.

#### Systematic sampling

In systematic sampling, the researcher selects every kth unit after a random starting point. If the frame contains N units and the desired sample size is n, the approximate sampling interval is:

Here, N is the number of frame units, n is the planned sample size, and k is the interval. Systematic sampling works well when the frame has a useful order and does not contain periodic patterns that coincide with k. A random start is important; selecting every kth unit from the first record can introduce bias.

#### Stratified sampling

The population is divided into mutually exclusive and collectively exhaustive subgroups called strata, such as regions, customer types, firm sizes, or age categories. A probability sample is then selected within each stratum.

Stratification is useful when subgroups differ meaningfully or when management requires reliable estimates for each subgroup. Proportional allocation selects sample sizes in proportion to stratum population sizes. Disproportionate allocation deliberately oversamples small or strategically important strata, but weights may be required when producing population-level estimates.

#### Cluster sampling

In cluster sampling, naturally occurring groups—such as neighbourhoods, branches, schools, or retail outlets—are selected, and units within selected clusters are studied. It reduces travel and administrative costs but often increases sampling variance because units within a cluster tend to resemble one another.

A multistage sample combines stages, for example selecting regions, then towns, then households, then individuals. It is practical for geographically dispersed populations, but design effects and unequal selection probabilities must be considered.

#### Probability proportional to size

When clusters differ substantially in size, clusters may be selected with probability proportional to their size. This can make the final selection probability more nearly equal when a fixed number of units is then sampled within each selected cluster. The full selection probability must nevertheless be documented and, where necessary, used to calculate weights.

2.2 Non-probability sampling

In non-probability sampling, selection probabilities are unknown or zero for some members of the target population. Statistical margins of sampling error based on random sampling theory should not be reported as if they applied to the sample.

Common forms include:

• Convenience sampling: selecting accessible respondents, such as visitors to a website or people in a nearby location.
• Purposive or judgement sampling: selecting cases because they possess relevant characteristics or expertise.
• Quota sampling: filling specified subgroup quotas without random selection within each quota.
• Snowball or respondent-driven recruitment: asking participants to identify other eligible participants.
• Self-selection or volunteer sampling: allowing individuals to opt into the study.

Non-probability methods are often appropriate for exploratory research, hard-to-reach populations, qualitative interviews, concept development, and early-stage entrepreneurial inquiry. They can be faster and less expensive than probability designs. However, they are vulnerable to selection bias, unknown inclusion probabilities, and overrepresentation of highly engaged or easily reached respondents.

The choice should reflect the inference intended. If the goal is to discover motivations among experienced users, purposive sampling may be defensible. If the goal is to estimate the percentage of all potential customers who will purchase, a probability design—or a carefully justified alternative with explicit limitations—is generally preferable.

2.3 Comparing sampling designs

3. Sample Size Estimation

Sample size should be determined by the research objective, population variability, desired precision, confidence level, design, subgroup requirements, expected response rate, budget, and consequences of error. There is no universally correct sample size.

3.1 Estimating a population proportion

For a simple random sample used to estimate a population proportion, the approximate sample size is:

where:

• n₀ = initial required sample size;
• z = standard-normal critical value corresponding to the chosen confidence level;
• p = anticipated population proportion;
• e = maximum tolerated margin of error, expressed as a proportion.

This formula assumes a probability sample, independent observations, a sufficiently large population relative to the sample, and a proportion estimator with the stated precision. Common z values are 1.645 for 90% confidence, 1.96 for 95%, and 2.576 for 99% confidence.

When p is unknown, p = 0.50 is commonly used because p(1 − p) is then largest and therefore produces the most conservative sample size.

Worked example. An entrepreneur wants to estimate the proportion of urban customers who would try a new service. The desired confidence level is 95%, the margin of error is ±5 percentage points, and no prior estimate of p is available.

Using z = 1.96, p = 0.50, and e = 0.05:

The required sample is rounded upward to 385 completed responses. The result is not a guarantee that the estimate is within five percentage points: it describes the performance expected under the stated random-sampling assumptions and does not account for non-response or coverage error.

3.2 Finite population correction

When the sample is a substantial fraction of a known finite population, the initial sample can be adjusted:

where N is the population size and n₀ is the initial sample size. The correction is most relevant when the sampling fraction is large; for very large populations, n is nearly equal to n₀.

Example. If the eligible customer population is N = 2,000 and n₀ = 385:

Thus, approximately 323 completed interviews may provide the stated precision under the same assumptions. The calculation is relevant only when the population size and frame are credible and sampling is without replacement.

3.3 Adjusting for design effect and non-response

Clustered, stratified, weighted, or otherwise complex designs may have a different variance from simple random sampling. The design effect, usually denoted DEFF, captures this relative variance change. An operational target for completed interviews can be expressed as:

where n_SRS is the simple-random-sampling requirement. If expected response rate is r, expressed as a proportion, the number of invitations or eligible contacts should be approximately:

For example, if 385 completed responses are needed, the design effect is 1.5, and the expected response rate is 50%:

These figures are planning values. The design effect should ideally be estimated from prior studies or an appropriate intracluster-correlation model, rather than chosen arbitrarily. In general, increasing the number of clusters is often more effective than interviewing many additional units within the same cluster.

3.4 Estimating a mean

When estimating a population mean, an approximate simple-random sample size is:

where σ is the anticipated population standard deviation in the measurement units and e is the desired margin of error in those same units. The formula assumes a quantitative variable, an approximately unbiased sample, and a reliable estimate of σ. If σ is unknown, researchers may use a pilot study, prior research, or a conservative planning estimate.

3.5 Subgroup analysis and practical judgement

A total sample adequate for an overall estimate may be inadequate for subgroup comparisons. If the research must compare three customer segments, each segment needs enough observations for stable estimates and meaningful analysis. Oversampling a small segment may be justified, but population estimates must then be weighted appropriately.

Statistical power is also relevant when testing differences or relationships. A sample-size calculation based only on a margin of error may not provide sufficient power to detect a commercially important difference. Researchers should specify the primary outcome, minimum meaningful effect, significance level, desired power, and analysis method before data collection.

4. Sampling Error and Representativeness

4.1 Sampling error

Sampling error is the difference between a sample estimate and the corresponding population value that occurs because the sample does not include every unit. It is a random phenomenon: repeated probability samples would produce somewhat different estimates.

For a proportion under simple random sampling, the standard error is approximately:

where p̂ is the observed sample proportion and n is the completed sample size. An approximate 95% confidence interval is:

This approximation is most suitable when the sample is probability-based and the expected counts of successes and failures are sufficiently large. For small samples or extreme proportions, exact or alternative interval methods may be preferable.

A confidence interval is not a probability statement that a fixed parameter moves randomly after the study. Under repeated sampling, a procedure producing 95% intervals would contain the true parameter approximately 95% of the time, assuming the model and sampling design are appropriate.

4.2 Non-response and coverage error

Unit non-response occurs when a selected eligible unit provides no usable response. Item non-response occurs when a respondent omits particular questions. Non-response can create bias when respondents and non-respondents differ on variables related to the outcome.

A simple response rate can be expressed as:

Industry standards may use more detailed disposition codes, distinguishing ineligible cases, refusals, non-contacts, and partial interviews. Researchers should report the definition used rather than presenting an unexplained percentage.

Coverage error occurs when the sampling frame excludes eligible units, includes ineligible units, or gives some units multiple chances of selection. For example, an online-only frame may exclude people with limited internet access. A customer database may exclude prospective customers who have never purchased.

4.3 Representativeness is not a single statistic

A sample is representative in relation to a specified target population, variables, and purpose. Matching population proportions on age and gender does not prove that the sample is representative on purchasing behaviour, attitudes, or unobserved characteristics. A sample can look demographically balanced while still being biased by recruitment channel, motivation, or timing.

Researchers should compare the sample with credible population benchmarks, investigate unusual patterns, and document all weighting procedures. Weighting can correct known imbalances under assumptions, but it cannot guarantee correction of unmeasured bias. Extreme weights can increase variance and make estimates unstable.

5. Fieldwork Design and Management

Sampling design becomes evidence only through fieldwork. Fieldwork includes recruitment, interviewing, observation, administration of instruments, quality control, respondent protection, and data management.

5.1 Prepare the fieldwork protocol

A professional protocol should specify:

1. target population and eligibility criteria;
2. sampling frame and selection procedure;
3. contact modes and contact attempts;
4. questionnaire or interview guide;
5. instructions for interviewers and supervisors;
6. consent and privacy procedures;
7. quotas, if used, and rules for replacement;
8. call-back and refusal-conversion procedures;
9. quality checks and escalation rules;
10. data security, coding, and delivery requirements.

Eligibility questions should be clear and should not allow interviewers to manipulate inclusion. Replacement rules must be defined in advance. Replacing a difficult respondent with an easier one can change the sample and conceal non-response bias.

5.2 Questionnaire and interviewer preparation

A questionnaire should use understandable language, a logical sequence, neutral wording, appropriate recall periods, and response options that are exhaustive and mutually exclusive where applicable. Pilot testing should examine comprehension, length, skip logic, respondent burden, and technical performance.

Interviewers require training in recruitment, consent, neutrality, probing, recording answers, handling distress, and maintaining confidentiality. They should not suggest answers, fabricate responses, coach participants, or selectively omit difficult cases. Certification through role-play or observed practice is preferable to merely distributing written instructions.

5.3 Monitoring fieldwork

Useful fieldwork indicators include completion rates by interviewer and location, contact outcomes, average duration, break-offs, item non-response, quota progress, unusual answer patterns, duplicate cases, and the proportion of interviews completed at unusual times. Monitoring should be used for corrective action, not simply retrospective reporting.

Quality assurance may include back-checking a sample of respondents, audio review where consent permits, GPS or timestamp checks for in-person work, paradata analysis, logic checks, and comparison of interviewer distributions. Such controls must be proportionate and must not undermine respondent privacy.

5.4 Digital and platform-based fieldwork

Online panels, mobile surveys, social media recruitment, and customer-relationship-management databases can reduce cost and speed delivery. They also introduce risks: duplicate accounts, professional survey takers, bots, device sharing, panel conditioning, and uncertain coverage. CAPTCHA or bot detection, unique survey links, attention checks used carefully, response-time review, and transparent recruitment documentation can improve quality.

An attention check should not be treated as a substitute for a sound sample. Excessive screening can remove legitimate respondents, and very short completion times may reflect familiarity rather than careless responding. Automated quality rules should be reviewed by a researcher before cases are deleted.

6. Evaluating Research Quality

Research quality is multidimensional. A useful evaluation distinguishes reliability, validity, precision, credibility, and practical usefulness.

6.1 Reliability

Reliability concerns consistency. A reliable instrument produces stable results under comparable conditions. Sources of unreliability include ambiguous wording, inconsistent interviewer behaviour, unstable constructs, and technical errors. Reliability does not establish that the measure is correct: a scale can consistently measure the wrong construct.

6.2 Validity

Content validity asks whether the instrument adequately covers the construct. Construct validity asks whether the measure behaves as theory predicts and relates to other measures appropriately. Criterion validity considers association with a relevant external criterion, such as actual purchase or retention. External validity concerns the extent to which findings can be transferred or generalised to the intended population and settings.

Sampling is especially important for external validity, while questionnaire design strongly affects measurement validity. Both must be addressed; neither compensates for the other.

6.3 A practical quality framework

A research team can assess quality through five questions:

• Coverage: Could eligible units be reached and selected?
• Selection: Was the selection process consistent with the intended inference?
• Response: Who participated, who did not, and how might this affect results?
• Measurement: Were questions, interviewers, and instruments capable of measuring the constructs accurately?
• Processing and reporting: Were data cleaned, weighted, analysed, and communicated transparently?

A quality report should disclose the population, frame, dates, mode, sample design, achieved sample, response outcomes, weighting, exclusions, limitations, and uncertainty calculations. Claims should not be stronger than the design supports.

7. Ethics, Governance and Responsible Practice

Research participants should receive understandable information about purpose, procedures, voluntary participation, risks, benefits, data use, and withdrawal rights. Consent should be appropriate to the context, and sensitive studies may require formal ethics review.

Researchers must minimise collection of personal data, restrict access, apply retention limits, and protect identifiers. Where data-protection law applies, the research team should establish a lawful basis, provide required notices, and avoid using research data for unrelated marketing without appropriate permission. Special care is required with children, vulnerable groups, health information, financial information, and location data.

Incentives should compensate time without becoming coercive. Deception, covert observation, and recontact require explicit justification and safeguards. Researchers should not manipulate sampling or reporting to produce a desired commercial conclusion. Sponsors should be informed when the design cannot support the requested claim.

8. Illustrative Case: Sampling for a Sustainable Meal-Service Venture

Background

GreenPlate, an illustrative early-stage venture, plans to launch a subscription meal service using reusable packaging in a large metropolitan area. The founders need to estimate likely trial among adults who purchase delivered meals at least once per month. They also want to compare two customer segments: young professionals and households with children.

Problem

The founders initially propose posting a survey link in a sustainability-focused social-media group. The resulting responses would be inexpensive, but they would likely overrepresent people already interested in environmental issues. The founders need an estimate useful for market sizing and a comparison of segment preferences.

Analysis

The target population is defined as adults in the metropolitan area who ordered a delivered meal at least once during the previous month. The team identifies a commercial research panel with screening variables and supplements it with a customer database from a pilot delivery partner. The two sources do not cover the population identically, so frame overlap and eligibility are documented.

A stratified design is selected, with strata based on the two priority segments and residential zones. The team plans 600 completed surveys: 300 per priority segment, allowing stable descriptive comparisons. Because the segments are not equally common in the population, the analysis plan specifies population weights for the overall estimate.

The questionnaire is piloted with 30 respondents. The pilot reveals that respondents interpret “sustainable packaging” inconsistently, so the concept is defined with a short neutral description. A randomized concept test is added to reduce order effects. Screening, consent, response-duration checks, duplicate detection, and partial-interview rules are documented before launch.

Approach and solution

The fieldwork team sends invitations in multiple time windows and monitors completion by segment and zone. Non-respondents receive two reminders, but no respondent is replaced merely because the survey is inconvenient. The team records invitation outcomes and compares early and late respondents as a diagnostic, while recognising that this is not a definitive test of non-response bias.

After data collection, 612 usable responses are available. Twelve cases fail a pre-specified duplicate rule and are removed. The final sample contains 600 cases. Weighting adjusts the segment and zone distribution to external population benchmarks. The report presents both weighted estimates and unweighted sample counts, explains the weighting assumptions, and avoids describing the online panel as a random sample of all city residents.

Outcome

The weighted estimate indicates moderate overall trial interest but a substantially higher rate among young professionals. However, willingness to try is much higher than willingness to pay the required price. The founders therefore postpone a citywide launch, conduct a price-and-packaging experiment with the priority segment, and revise the financial model.

Learning points

The case demonstrates that sample quality depends on alignment between the target population, frame, selection process, measurement instrument, and intended claim. A large online sample would not automatically solve coverage or self-selection problems. Stratification improved the ability to compare important segments, while weighting made the overall estimate more aligned with the population. Most importantly, the research separated stated trial interest from commercially relevant willingness to pay.

9. Contemporary Developments and Best Practices

Market research increasingly combines probability samples, online panels, administrative data, behavioural records, and qualitative recruitment. Hybrid designs can improve coverage but require careful deduplication, harmonised measurement, and explicit modelling of source differences. Calibration and weighting methods may align samples with known benchmarks, but they remain dependent on the quality of those benchmarks and assumptions about unobserved variables.

Artificial intelligence can support address validation, fraud detection, interviewer monitoring, coding of open-ended responses, and adaptive scheduling. It should not silently determine eligibility, exclude respondents, or generate synthetic responses without governance and disclosure. Automated decisions require auditability, bias testing, human review, and secure handling of personal data.

Best practice is therefore to:

1. define the population before selecting respondents;
2. match the design to the intended inference;
3. use probability sampling when population estimation is central;
4. justify non-probability sampling when exploration or access is the priority;
5. calculate sample size from precision, power, design effect, and response expectations;
6. pilot both the instrument and the fieldwork process;
7. pre-specify eligibility, replacement, exclusion, and weighting rules;
8. monitor fieldwork continuously;
9. report limitations and uncertainty honestly; and
10. protect participants throughout the research lifecycle.

Chapter Summary

Sampling enables researchers to study a manageable part of a market while making carefully qualified statements about a larger population. Probability sampling provides known selection probabilities and a foundation for estimating sampling error; simple random, systematic, stratified, cluster, multistage, and probability-proportional-to-size designs serve different operational needs. Non-probability methods are valuable for exploration and hard-to-reach groups but do not support conventional margins of sampling error without strong additional assumptions.

Sample size depends on confidence, precision, expected variability, population size, design effect, subgroup requirements, and response rate. The proportion formula n₀ = z²p(1 − p) ÷ e² is useful under specified probability-sampling assumptions, while finite-population, design-effect, and non-response adjustments improve planning. Sampling error is only one component of uncertainty. Coverage error, non-response, measurement error, interviewer effects, processing mistakes, and weighting can materially affect conclusions.

Effective fieldwork requires a written protocol, trained personnel, pilot testing, ethical recruitment, quality monitoring, and transparent reporting. Research quality is multidimensional: coverage, selection, response, measurement, processing, reliability, validity, and generalisability must be considered together. Entrepreneurs should ensure that research claims do not exceed what the sample and design can support.

Key Terms

Target population: The complete group about which the research seeks to draw conclusions.

Sampling unit: The basic element selected at a stage of sampling.

Sampling frame: The operational list or system from which a sample is selected.

Sample: The units selected for participation in a study.

Probability sampling: Sampling in which every eligible unit has a known, non-zero selection probability.

Non-probability sampling: Sampling in which selection probabilities are unknown or zero for some units.

Simple random sampling: A design in which every possible sample of a specified size has equal selection probability.

Stratified sampling: Dividing the population into strata and sampling within each stratum.

Cluster sampling: Selecting groups of population units and studying units within selected groups.

Multistage sampling: Sampling carried out through two or more successive stages.

Quota sampling: Filling subgroup targets without necessarily selecting respondents randomly within groups.

Sampling error: Random difference between a sample estimate and the corresponding population parameter.

Non-response: Failure of a selected eligible unit to provide a usable response.

Coverage error: Error caused by mismatch between the sampling frame and target population.

Design effect: The relative variance of a complex sample compared with a simple random sample.

Margin of error: The planned maximum difference between an estimate and population value under stated assumptions.

Representativeness: The degree to which a sample supports inference about its defined target population.

Reliability: Consistency or stability of a measurement process.

Validity: The extent to which a study or measure supports accurate interpretation and intended use.

Weighting: Assigning analysis factors to observations to reflect unequal selection probabilities or population benchmarks.

Review Questions

Short Answer Questions

1. Distinguish a target population from a sampling frame.
2. What is the principal difference between probability and non-probability sampling?
3. Explain why increasing sample size does not necessarily eliminate non-response bias.
4. Define coverage error and give one market-research example.
5. What information should a fieldwork protocol contain?

Long Answer Questions

1. Compare simple random, stratified, and cluster sampling, including their advantages, limitations, and appropriate applications.
2. Explain how confidence level, margin of error, expected proportion, finite population size, design effect, and response rate affect sample-size planning.
3. Discuss the meaning and limitations of a 95% confidence interval for a sample proportion.
4. Develop a quality-control plan for a multi-interviewer telephone survey.
5. Evaluate the ethical and governance responsibilities involved in using online panels and customer databases for market research.

Analytical / Application Questions

1. A start-up needs to estimate a population proportion at 95% confidence with a ±4% margin of error and no prior estimate of the proportion. Calculate the initial simple-random sample size using p = 0.50, and explain how the target should change if the expected response rate is 40%.
2. A survey achieves demographic quotas but recruits entirely through a sustainability-focused social-media group. Assess the likely threats to representativeness and propose a stronger design.
3. A national retailer wants reliable results for four regions, including two small regions. Recommend a sampling design, allocation strategy, fieldwork controls, and reporting approach.

Multiple Choice Questions with Answers

Question 1

Which feature defines probability sampling?

A. Respondents select themselves into the study

B. Every eligible unit has a known, non-zero probability of selection

C. The researcher selects only expert respondents

D. The sample is always proportional to population size

Correct Answer: B. Every eligible unit has a known, non-zero probability of selection

Question 2

A population contains 10,000 units and a researcher wants a systematic sample of 500 units. What is the approximate sampling interval?

A. 5

B. 10

C. 20

D. 50

Correct Answer: C. 20

Question 3

When no prior estimate of a population proportion is available, which value of p commonly gives the most conservative sample-size estimate?

A. 0.01

B. 0.25

C. 0.50

D. 0.90

Correct Answer: C. 0.50

Question 4

Which problem occurs when an online-only frame excludes eligible people with limited internet access?

A. Coverage error

B. Design effect

C. Item non-response

D. Randomisation

Correct Answer: A. Coverage error

Question 5

A sample requires 400 completed interviews, the expected response rate is 50%, and no other adjustment is made. Approximately how many eligible contacts are needed?

A. 200

B. 400

C. 600

D. 800

Correct Answer: D. 800

Question 6

Which sampling design is most directly intended to ensure that important subgroups are included in planned numbers?

A. Stratified sampling

B. Convenience sampling

C. Snowball sampling

D. Unrestricted volunteer sampling

Correct Answer: A. Stratified sampling

Question 7

What is the main statistical consequence of interviewing many similar respondents within the same selected cluster?

A. It always eliminates sampling error

B. It can increase variance through a design effect

C. It guarantees population representativeness

D. It converts the study into a census

Correct Answer: B. It can increase variance through a design effect

Question 8

Which statement about non-response is most accurate?

A. It matters only when the response rate is below 10%

B. It is harmless whenever the sample is large

C. It can bias results when respondents differ from non-respondents on relevant variables

D. It is the same as coverage error

Correct Answer: C. It can bias results when respondents differ from non-respondents on relevant variables

Question 9

A research team uses purposive sampling to recruit experienced industry buyers for exploratory interviews. What is the most defensible interpretation?

A. The findings automatically estimate all buyers in the market

B. The design is suitable for accessing relevant expertise but has limited statistical generalisability

C. The sample has a calculable random-sampling margin of error

D. The study is equivalent to a census

Correct Answer: B. The design is suitable for accessing relevant expertise but has limited statistical generalisability

Question 10

Which practice best supports research quality during fieldwork?

A. Allowing interviewers to replace difficult respondents freely

B. Changing eligibility criteria after seeing early results

C. Pre-specifying procedures, training interviewers, and monitoring outcomes

D. Deleting all unusually short interviews without documented rules

Correct Answer: C. Pre-specifying procedures, training interviewers, and monitoring outcomes