Sampling and Fieldwork Design

Introduction

Market research rarely studies every member of a market. Instead, it uses a carefully designed subset to make defensible statements about a larger population. Sampling therefore connects research questions to evidence. A weak sample can invalidate an otherwise sophisticated questionnaire, statistical model, or brand strategy; a well-designed sample can produce useful insight at substantially lower cost than a census.

In brand management, sampling decisions affect estimates of awareness, consideration, preference, purchase, loyalty, brand associations, willingness to pay, and response to communications. Fieldwork design then determines whether the intended sample is actually reached, screened, interviewed, monitored, and documented. Sampling is consequently both a statistical problem and a managerial governance problem.

This chapter develops the logic of population definition, sampling frames, sampling units, probability and non-probability designs, sample-size determination, statistical power, recruitment, weighting, nonresponse, fieldwork quality, and representation. It also considers mixed-method research, online panels, mobile research, and ethical requirements.

Learning Objectives

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

• distinguish a target population, study population, sampling frame, sampling unit, element, and respondent;
• select and justify probability or non-probability sampling designs;
• calculate and interpret sample size for proportions and means under stated assumptions;
• distinguish precision, confidence, power, effect size, and practical significance;
• design recruitment, screening, quotas, incentives, interviewer procedures, and quality controls;
• diagnose coverage error, nonresponse, sampling error, and weighting effects;
• integrate qualitative purposive sampling with quantitative representation requirements; and
• evaluate sampling plans ethically, operationally, and analytically.

1. The Architecture of a Sample

1.1 Population, frame, unit, and respondent

The target population is the complete group to which the researcher intends to generalise. For example, a study may target adults aged 18–64 who purchased a category at least once during the previous six months in a specified country. The study population is the portion of the target population that can actually be identified and reached under the research design.

A sampling frame is the operational list, database, platform, map, registry, or procedure used to identify potential members of the study population. Examples include a customer CRM file, a business directory, a probability-based online panel, an electoral register, or a geographic list of households. A frame is not automatically equivalent to the population: it may omit eligible people, include ineligible people, or contain duplicates.

The sampling unit is the unit selected at a particular stage. It may be a person, household, store, firm, social-media account, geographic area, or transaction. The element is the ultimate member about whom information is required. In a household survey, households may be selected first, but the element may be an adult purchaser within the household. The respondent is the element who actually provides data.

These distinctions matter because errors arise when researchers treat a convenient frame as the population or confuse the unit contacted with the unit analysed. A survey of CRM customers, for instance, cannot automatically represent all category buyers.

1.2 Coverage and the inferential chain

A valid inference requires an explicit chain:

Target population → study population → sampling frame → selected units → respondents → analysed cases.

At each transition, some members may be lost or misclassified. Undercoverage occurs when eligible population members have no chance of selection. Overcoverage occurs when the frame contains ineligible or duplicate units. Coverage error is the resulting difference between the target population and the frame-based population.

A research brief should therefore specify inclusion and exclusion criteria, geography, time period, category behaviour, age or eligibility boundaries, and the unit of analysis. Ambiguous definitions produce ambiguous samples.

1.3 Sampling error and nonsampling error

Sampling error is the random difference between a sample estimate and the corresponding population value that arises because only some units are observed. It can be quantified under probability sampling. Nonsampling error includes coverage error, nonresponse error, measurement error, processing error, fraudulent participation, interviewer effects, and specification error. Increasing the sample does not necessarily solve nonsampling error; a large biased sample can be less useful than a smaller, carefully controlled one.

2. Probability Sampling Designs

Probability sampling gives each eligible unit a known, non-zero probability of selection. The probabilities need not be equal, but they must be known or estimable. This feature supports design-based estimation, sampling-error calculations, and transparent population inference.

2.1 Simple random sampling

In simple random sampling (SRS), every sample of a specified size has the same chance of selection. It is conceptually straightforward and useful when a complete, accurate frame exists. A random-number generator—not convenience or interviewer judgement—selects units.

Its limitations include the cost of contacting geographically dispersed units and the practical difficulty of obtaining a complete frame. SRS also may produce too few cases in small but strategically important subgroups.

2.2 Systematic sampling

Systematic sampling selects every kth unit after a random start. If the frame has N units and n are required, the approximate interval is:

where k is the sampling interval. Systematic sampling is efficient and easy to administer, but periodic ordering in the frame can create bias. For example, a customer list sorted by purchase frequency may produce a distorted sample if the interval interacts with that pattern.

2.3 Stratified sampling

In stratified sampling, the population is divided into mutually exclusive strata—such as region, age group, customer status, or usage level—and a sample is drawn within each stratum. Stratification can improve precision when members within strata are relatively homogeneous and strata differ from one another. It also guarantees representation of specified groups.

Proportional allocation sets each stratum’s sample in proportion to its population size. Disproportionate allocation deliberately oversamples smaller or high-priority groups, after which weights may be required. Stratification is not the same as setting uncontrolled quotas: probability selection must still occur within strata if probability inference is claimed.

2.4 Cluster and multistage sampling

Cluster sampling selects naturally occurring groups, such as regions, stores, schools, or households, and then studies all or some units within selected clusters. It reduces travel and operational cost but usually decreases statistical efficiency because units in the same cluster tend to resemble one another.

Multistage sampling combines stages—for example, select regions, then postal areas, then households, then one eligible adult. It is common in national research. The design must record selection probabilities at every stage. The overall probability of selection for a unit is the product of its conditional probabilities across stages:

Here π_i is unit i’s overall inclusion probability, π_{hi} is its conditional probability at stage h, and H is the number of stages. This expression assumes the stages are correctly specified and the relevant conditional probabilities are known or estimable.

2.5 Design effect

Clustered or unequal-probability samples often contain less information than an SRS of the same nominal size. A common approximation for equal-sized clusters is:

where deff is the design effect, m is the average number of sampled respondents per cluster, and ρ is the intraclass correlation coefficient (ICC), the correlation among observations within clusters. The approximation assumes a reasonably common cluster size and a simple clustering structure. If ρ = 0, clustering adds no variance under this approximation; as m or ρ increases, precision declines.

The effective sample size is approximately:

where n is the achieved sample size. Design effects also arise from unequal weights, stratification, repeated measures, and complex estimators, so software-based variance estimation is preferable for final analysis.

3. Non-Probability Sampling

Non-probability sampling does not provide known selection probabilities for all eligible units. It is not inherently inferior for every purpose, but it requires more cautious claims. It is particularly common in exploratory qualitative work, difficult-to-reach populations, rapid concept testing, and commercial online panels.

3.1 Major forms

Convenience sampling recruits whoever is readily available. It is inexpensive but vulnerable to severe coverage and self-selection bias.

Purposive sampling deliberately selects information-rich cases according to theoretically or managerially relevant criteria. In qualitative brand research, researchers may recruit category enthusiasts, lapsed users, loyalists, switchers, or culturally distinctive segments.

Quota sampling fills predetermined subgroup counts, such as 50% women and 50% men or specified age-by-region cells. Quotas improve visible composition but do not remove selection bias within cells, because participants may still be selected by convenience or panel availability.

Snowball or respondent-driven recruitment uses referrals from existing participants. It can reach hidden or networked populations, but participants may share characteristics and networks. Respondent-driven sampling has specialised assumptions and estimators; ordinary snowball recruitment should not be presented as probability sampling.

Judgement sampling relies on expert selection, often for business-to-business informants, key opinion leaders, or specialist purchasers.

3.2 When non-probability designs are appropriate

Non-probability designs are defensible when the objective is discovery rather than population estimation, when no adequate frame exists, when the population is rare or hidden, or when qualitative depth is more important than prevalence measurement. They can also be used for experimental studies where random assignment—not population sampling—is the main basis for causal identification.

Researchers should state exactly what the design supports. “Among recruited panel participants” is a narrower and more honest claim than “among all consumers.” Calibration, benchmarking, sensitivity analyses, and probability-based validation subsamples can strengthen, but do not magically eliminate, selection concerns.

4. Sample Size, Precision, and Power

Sample size is determined by the decision purpose, outcome variability, desired precision, confidence level, design, subgroup requirements, expected response, and available resources. There is no universally correct sample size.

4.1 Estimating a proportion

For an SRS estimating a population proportion, a commonly used planning formula is:

where n₀ is the initial required sample size, z is the standard-normal critical value for the selected confidence level, p is the anticipated population proportion, and e is the desired maximum margin of error expressed as a proportion. The formula assumes an SRS, independent observations, a sufficiently large population, and a proportion estimate based on a simple unweighted analysis.

If no prior estimate of p is available, p = 0.50 is conservative because p(1 − p) is then maximal. For 95% confidence, z ≈ 1.96. If e = 0.05:

Thus, at least 385 completed interviews are required before allowing for design effects, subgroup analysis, or nonresponse.

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

This correction is relevant when the sample is a substantial fraction of the population, commonly around 5% or more, and selection is without replacement. It does not correct coverage or nonresponse bias.

4.2 Adjusting for design and response

If the planned design effect is deff and the expected response or completion rate is r, the number of contacts or invitations required can be approximated by:

Here n_target is the desired number of completed cases under SRS-equivalent precision, and r is expressed between 0 and 1. For example, if n_target = 385, deff = 1.5, and r = 0.60:

The field team should plan approximately 963 eligible invitations or contacts, subject to additional screening incidence and quota constraints. If only 70% of contacted people are eligible, the required initial contacts would be approximately 963 ÷ 0.70 ≈ 1,376.

4.3 Means and other outcomes

For estimating a population mean under SRS, the approximate sample-size formula is:

where σ is the anticipated population standard deviation in the outcome’s measurement units and e is the desired margin of error in those same units. The formula assumes a continuous outcome, independent observations, and a reasonably stable estimate of σ from prior research or a pilot. For a mean brand-attitude score, e must be expressed in score points, not percentages.

4.4 Statistical power

Precision concerns estimation; power concerns detecting an effect if it exists. For a two-group comparison, planning requires a significance level α, desired power 1 − β, effect size, outcome variability, allocation ratio, and test specification. A generic standardised effect size for a difference in means is:

where μ₁ and μ₂ are the group means and σ is a relevant common standard deviation. A small effect may be strategically important even when difficult to detect, while a statistically significant effect may be commercially trivial in a very large sample.

Power calculations should be conducted with a recognised statistical procedure or software appropriate to the design. Cluster sampling, weighting, repeated observations, multiple testing, attrition, and subgroup analyses generally require larger samples than a simple two-group calculation suggests. Post hoc power calculations based only on an observed p-value are usually uninformative; confidence intervals and the size and uncertainty of the estimated effect are more useful.

4.5 Subgroups and precision trade-offs

A total sample of 1,000 does not imply 1,000 observations for every segment. If a subgroup represents 10% of the sample, it may contain only about 100 cases, and weighting or missing data may reduce effective precision further. Sample planning should begin with the minimum decision-relevant cell sizes, not merely the headline total.

5. Recruitment and Fieldwork Design

A fieldwork plan converts the sampling design into controlled operations. It should specify the sample source, recruitment channel, eligibility screener, quotas or strata, contact attempts, interview mode, language, dates, incentives, interviewer instructions, quality checks, escalation procedures, data security, and stopping rules.

5.1 Screening and eligibility

Screeners should be short, logically ordered, and designed to reduce false eligibility. Questions about recent category purchase, decision involvement, employment in research or advertising, geographic location, age, and prior participation may be necessary. Sensitive questions should be justified and handled proportionately. Eligibility logic should be programmed and tested, with impossible combinations blocked.

Overly transparent screeners can encourage strategic misrepresentation. Where appropriate, use neutral wording, verification questions, and cross-checks. However, deception must not violate informed consent or create material risk.

5.2 Contact and recruitment protocols

Probability fieldwork requires standardised contact attempts, varied days and times, call dispositions, and documented substitutions. Substituting a convenient respondent for a noncontacted sampled unit can destroy the intended design. Online studies should monitor invitations, duplicate devices, suspicious completion speed, straight-lining, repeated open-text responses, and panel overlap.

Qualitative recruitment should seek variation relevant to the research question rather than statistical proportionality. Researchers may use purposive matrices combining life stage, category involvement, brand relationship, geography, and cultural context. Recruitment should continue until the research team judges that additional interviews are producing limited conceptual novelty—often called thematic saturation—but saturation is a reasoned analytic judgement, not a mechanical sample-size law.

5.3 Interviewer and mode effects

Interviewers can influence responses through probing, tone, identity, or selective recording. Training should cover neutral probing, translation, handling distress, consent, and escalation. Back-checks and audio review may be used where lawful and consented.

Mode effects occur when telephone, face-to-face, web, mobile, or messaging modes produce different participation or answers. Sensitive questions may elicit greater disclosure in self-completion modes, whereas complex visual stimuli may work better on larger screens. A mode decision should consider coverage, accessibility, respondent burden, privacy, stimulus requirements, and measurement comparability.

6. Nonresponse, Weighting, and Representation

6.1 Nonresponse

Unit nonresponse occurs when a selected or invited unit provides no usable interview. Item nonresponse occurs when a respondent omits particular questions. Nonresponse bias arises when respondents and nonrespondents differ on variables related to the estimate, and the difference is not fully corrected by weighting or adjustment.

A response rate is useful for fieldwork monitoring but is not itself a complete measure of bias. A high response rate can still be biased if the frame excludes a major group; a lower rate may produce limited bias if respondents resemble nonrespondents on relevant variables. Report contact outcomes, eligibility, completion, break-offs, and response-rate definitions transparently.

6.2 Weighting

A basic design weight is the inverse of inclusion probability:

where w_i is unit i’s base weight and π_i is its probability of selection. If a person has a 0.02 inclusion probability, the base weight is 50. Nonresponse adjustment may multiply this weight by an adjustment factor for cells or propensity classes, and calibration or raking may then align weighted totals with trusted population margins.

For a weighted estimate of a mean, a common estimator is:

where y_i is the observed value and w_i is the final analysis weight. Weights must be appropriate to the estimand and analysis. Extreme weights can increase variance, so trimming may be considered, but trimming changes the estimator and should be documented.

Weighting corrects known compositional differences under assumptions; it cannot reliably repair unmeasured selection bias. Analysts should report weighted and unweighted results, examine sensitivity, and incorporate weights in variance estimation.

6.3 Representation beyond demographics

Matching a sample to age, gender, and region does not guarantee representativeness. Brand outcomes may also depend on category incidence, usage frequency, income, education, language, digital access, disability, cultural identity, and media behaviour. The relevant variables depend on the research objective and the quality of available benchmarks. Representation should be evaluated against the population that the claim concerns, not against generic census variables alone.

7. Fieldwork Quality, Governance, and Ethics

Quality assurance should operate before, during, and after data collection. Before launch, pilot the instrument, test routing, verify translations, estimate completion time, check quotas, and review consent language. During fieldwork, monitor incidence, dropout, duration, straight-lining, duplicate identifiers, unusual geographies, interviewer patterns, and open-ended quality. After fieldwork, audit exclusions, cleaning rules, weighting, and deviations from protocol.

A useful fieldwork dashboard distinguishes operational indicators from validity indicators. Operational indicators include completed interviews, cost per completion, and quota progress. Validity indicators include coverage, response patterns, fraud flags, item nonresponse, and design consistency. Stopping fieldwork solely because quotas are full may be unsafe if one source supplies unusually homogeneous respondents.

Ethical practice requires voluntary informed consent, truthful explanation of purpose at an appropriate level, the right to withdraw, minimisation of personal data, secure storage, restricted access, retention limits, and transparent use of incentives. Researchers must avoid coercive incentives, discriminatory exclusion, unnecessary collection of sensitive attributes, and undisclosed reuse of personal data. Children and vulnerable groups require additional safeguards and appropriate permissions. Applicable privacy and research-governance laws must be followed; legal compliance is a minimum, not a substitute for ethical judgement.

AI-assisted recruitment, fraud detection, translation, and synthetic data can improve efficiency, but they create risks. Automated screening may reproduce historical exclusion, facial or voice analysis may be intrusive, and synthetic respondents cannot be treated as evidence of real population attitudes without validation. Human review, documentation, auditability, and disclosure are essential.

8. An Integrated Design Workflow

A defensible sampling plan can be developed through the following sequence:

1. Define the estimand or insight objective: what population, outcome, time period, and comparison matter?
2. Specify the target population and eligibility rules.
3. Identify the best available frame and document coverage gaps.
4. Choose the sampling unit and selection design.
5. Decide whether probability inference is required or whether purposive exploration is sufficient.
6. Determine key strata, quotas, subgroup cells, and minimum analytic sizes.
7. Calculate a starting sample size using an appropriate outcome, confidence or power requirement, design effect, and expected completion rate.
8. Design recruitment, contact, incentive, consent, and quality procedures.
9. Pilot the full process, not only the questionnaire.
10. Monitor fieldwork and document deviations.
11. Clean, weight, and analyse data using design-consistent methods.
12. State the scope and limitations of the conclusions.

Applied Case: Illustrative National Brand-Choice Study

Background

A fictional packaged-beverage company wants to estimate unaided awareness and understand switching between its brand and two competitors among adults who purchased the category in the past three months. Management requires national estimates and comparisons among three age groups.

Problem

The company’s CRM includes only its loyalty-program members, who are more engaged than ordinary category buyers. A cheap online convenience sample would be fast but could overrepresent heavy users and digitally active consumers. The team also needs approximately 400 completed interviews in each age group for planned comparisons.

Analysis and approach

The researchers define the target population as category purchasers aged 18 or above in the country during the previous three months. They use a probability-based panel with demographic profiling, stratify invitations by region and age, and screen for recent category purchase. Because the three age groups are sampled disproportionately to meet analytic requirements, final estimates will be calibrated to population benchmarks for age, region, and gender.

A pilot indicates a completion rate of 55%. The team anticipates a design effect of 1.3 and requires 1,200 completed interviews overall: 400 per age group. The approximate invitation requirement is:

They therefore plan approximately 2,837 eligible invitations, with additional initial invitations because not every contacted panel member will pass the purchase screen. The protocol includes device and duplicate checks, minimum open-text standards, speed review, and a small number of callbacks for verification where permitted.

Outcome

The weighted estimate shows lower overall awareness than the unweighted result because the achieved sample contains too many heavy category users. The age-group comparison remains directionally similar but has wider uncertainty after weighting. Management decides to use the national weighted estimate for market sizing, while treating exploratory open-ended explanations as hypothesis-generating rather than population prevalence evidence.

Learning points

The case demonstrates that a frame is not automatically the target population, subgroup requirements drive sample planning, disproportionate selection requires weighting, and quality controls must address both fraudulent participation and substantive respondent differences. It also shows why a response rate alone cannot establish representativeness.

Challenges and Limitations

No sampling design removes all uncertainty. Frames become outdated, population definitions change, and benchmarks may themselves be imperfect. Probability methods can be expensive and still suffer nonresponse; non-probability methods can be highly informative but often cannot support strong prevalence claims. Weighting can increase variance and may conceal rather than resolve selection problems. Online access, language, disability, privacy concerns, and digital literacy can create systematic exclusion. Finally, statistical representativeness does not guarantee conceptual relevance: a precisely measured question may still fail to capture the meanings through which consumers experience a brand.

Best practice is therefore not to seek a mythical perfect sample but to align the design, claim, evidence, and uncertainty. Research reports should state who was eligible, how people were reached, how many were invited and completed, what exclusions occurred, how weights were constructed, what design effects were considered, and which conclusions are exploratory or generalisable.

Chapter Summary

Sampling translates a research objective into a defensible set of observations. Target populations, study populations, frames, sampling units, elements, and respondents must be distinguished because each transition can introduce error. Probability designs—including simple random, systematic, stratified, cluster, and multistage sampling—support known selection probabilities and design-based inference. Non-probability designs are valuable for qualitative discovery, rare populations, and exploratory work, but claims must be limited to what the recruitment process supports.

Sample size depends on precision or power, confidence, variability, effect size, subgroup needs, design effect, and expected completion. Weighting uses inverse selection probabilities and calibration to address known compositional differences, but it cannot guarantee removal of unmeasured selection bias. Fieldwork quality requires tested screeners, standardised recruitment, interviewer training, monitoring, fraud detection, transparent deviations, and ethical data governance. Strong research aligns the population, design, fieldwork, analysis, and final claim.

Key Terms

Target population: The complete group to which a study intends to generalise.

Study population: The portion of the target population accessible under the research design.

Sampling frame: The list or operational procedure used to identify eligible units.

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

Element: The ultimate member about whom information is required.

Respondent: An eligible person or unit that provides usable data.

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

Sampling error: Random difference caused by observing a sample rather than a census.

Simple random sampling: Selection in which every sample of a specified size has equal probability.

Stratified sampling: Sampling separately within defined population strata.

Cluster sampling: Sampling naturally occurring groups rather than individual units directly.

Design effect: The variance ratio associated with a complex design relative to SRS.

Non-probability sampling: Sampling without known non-zero selection probabilities for all units.

Quota sampling: Filling prescribed subgroup counts without necessarily using random selection within cells.

Nonresponse bias: Bias arising when respondents differ from nonrespondents on relevant characteristics.

Base weight: The inverse of a unit’s selection probability.

Calibration: Adjustment of weights so sample totals match trusted population benchmarks.

Statistical power: The probability of detecting a specified effect when that effect exists.

Effective sample size: The approximate information equivalent of an SRS after design effects.

Thematic saturation: A qualitative judgement that additional data are producing limited new conceptual insight.

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. Why can cluster sampling reduce statistical efficiency?
4. Define coverage error and give one example.
5. What does a base weight represent?

Long Answer Questions

1. Compare stratified, cluster, and multistage sampling for a national brand study. Discuss cost, representation, and precision.
2. Explain how sample size for a proportion is determined and discuss the assumptions behind the standard formula.
3. Discuss the causes of nonresponse bias and evaluate strategies for reducing or assessing it.
4. Explain why quotas do not automatically make a non-probability sample representative.
5. Develop an ethical and operational fieldwork protocol for an online consumer survey.

Analytical / Application Questions

1. A study requires 500 completed interviews, expects a design effect of 1.4, and estimates a completion rate of 50%. Calculate the approximate number of eligible invitations required and explain one reason the actual number may need to be higher.
2. A CRM-based survey reports much higher loyalty than a population benchmark. Diagnose at least three possible sources of bias and propose a revised sampling design.
3. A qualitative brand project seeks insight from loyalists, switchers, lapsed users, and category rejectors. Construct a purposive recruitment matrix and explain why proportional demographic sampling is not the primary criterion.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes a sampling frame from a target population?

A. A frame is the theoretical group, while the target population is the contact list

B. A frame is the operational source used to identify units, while the target population is the group to which inference is intended

C. A frame contains only respondents, while the target population contains only nonrespondents

D. There is no distinction when a survey is conducted online

Correct Answer: B. A frame is the operational source used to identify units, while the target population is the group to which inference is intended

Question 2

A study uses simple random sampling and seeks a 95% confidence estimate of a proportion with p = 0.50 and margin of error 0.05. Approximately how many completed cases are required before other adjustments?

A. 96

B. 196

C. 385

D. 1,000

Correct Answer: C. 385

Question 3

Which design deliberately samples separate population groups such as age bands and then selects units within each group?

A. Stratified sampling

B. Snowball sampling

C. Convenience sampling

D. Unrestricted cluster sampling

Correct Answer: A. Stratified sampling

Question 4

If the inclusion probability for a sampled unit is 0.02, what is its basic inverse-probability weight?

A. 0.02

B. 2

C. 20

D. 50

Correct Answer: D. 50

Question 5

Under the approximation deff ≈ 1 + (m − 1)ρ, what happens when the intraclass correlation ρ increases while cluster size remains fixed?

A. The design effect increases

B. The design effect becomes zero

C. The effective sample size necessarily doubles

D. Sampling becomes equivalent to a census

Correct Answer: A. The design effect increases

Question 6

Which statement about quota sampling is correct?

A. It guarantees random selection within every quota cell

B. It eliminates all nonresponse bias

C. It can improve visible subgroup composition but does not by itself establish probability-based representation

D. It is identical to stratified random sampling

Correct Answer: C. It can improve visible subgroup composition but does not by itself establish probability-based representation

Question 7

A researcher wants to explore the experiences of lapsed users and category rejectors who are difficult to identify in a general population frame. Which approach is most directly appropriate for exploratory qualitative work?

A. Purposive recruitment

B. Systematic sampling with a fixed interval

C. A census of all consumers

D. Proportional allocation from a complete electoral register

Correct Answer: A. Purposive recruitment

Question 8

Which statement about weighting is most accurate?

A. Weighting always removes every form of selection bias

B. Weighting can adjust known compositional differences but may increase variance and cannot guarantee correction of unmeasured bias

C. Weighting is required only for qualitative interviews

D. Weighting makes sample size irrelevant

Correct Answer: B. Weighting can adjust known compositional differences but may increase variance and cannot guarantee correction of unmeasured bias

Question 9

A project needs 1,200 completed interviews, has a design effect of 1.3, and expects a 55% completion rate. Approximately how many eligible invitations are required?

A. 660

B. 1,200

C. 2,182

D. 2,837

Correct Answer: D. 2,837

Question 10

Which is an example of item nonresponse?

A. An eligible person never receives an invitation

B. A sampled household refuses the entire interview

C. A respondent completes the survey but leaves income unanswered

D. A frame omits people without internet access

Correct Answer: C. A respondent completes the survey but leaves income unanswered