Sampling Design and Fieldwork

Introduction

Marketing research rarely studies every member of a market. A census of all customers, households, businesses, or users may be too costly, slow, or operationally impossible. Sampling provides a disciplined way to learn about a larger population by studying a smaller group. The quality of the resulting evidence depends not only on the number of respondents, but also on how the population is defined, how units are selected, how fieldwork is managed, and how bias and uncertainty are assessed.

Sampling design is therefore a bridge between research objectives and empirical evidence. A carefully designed sample can support defensible estimates with limited resources. A poorly designed sample may produce precise-looking results that systematically misrepresent the market.

This chapter explains population definition, sampling frames, sampling units, probability and non-probability sampling, sample-size planning, sampling error, confidence, nonresponse and coverage bias, recruitment, fieldwork quality control, weighting, and research logistics.

Learning Objectives

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

1. Define a target population, sampling frame, sampling unit, element, and respondent.
2. Distinguish probability from non-probability sampling and select suitable designs.
3. Estimate sample size for proportions and explain the assumptions behind the calculation.
4. Interpret sampling error, confidence intervals, design effects, and finite-population corrections.
5. Diagnose coverage error, nonresponse bias, and other threats to representativeness.
6. Develop a recruitment and fieldwork plan with appropriate quality controls.
7. Explain when weighting is necessary and identify its limitations.
8. Make practical and ethical decisions about sampling in marketing research.

1. The Logic of Sampling

A population is the complete set of units about which a researcher wishes to draw conclusions. In marketing, the population might be all current users of a brand, all adults who purchased a category in the last six months, or all retail outlets carrying a product.

The target population is the population specified by the research objectives. It should identify the relevant people or organizations, geographical area, time period, and eligibility conditions. “Consumers” is usually too vague; “residents of the country aged 18 or above who purchased packaged coffee at least once during the previous three months” is more operational.

A sampling frame is the practical list, database, map, panel, or procedure used to identify and select population units. Examples include a customer database, electoral register, business directory, postal address file, online panel, or list of retail stores. The frame may not perfectly match the target population.

A sampling unit is the unit selected at a particular stage of sampling. It may be a person, household, store, organization, geographic area, or online account. The element is the basic unit about which information is collected. In a household survey, a household may be selected as the sampling unit, while one adult within it is the final element. A respondent is the person who provides the information.

These concepts must be separated. For example, a study of restaurant patrons could have:

1.1 Inclusion and exclusion criteria

Eligibility rules should be established before recruitment. They might specify age, location, purchase history, employment status, decision-making responsibility, or product usage. Exclusion criteria may remove professional respondents, employees of the sponsor, duplicate records, or individuals who participated in a closely related study.

Ambiguous criteria create inconsistent screening. For instance, “recent customer” should be defined by a precise period and a verifiable behavior. Screening questions should be brief, neutral, and ordered so that ineligible people are not exposed unnecessarily to the full questionnaire.

1.2 Sampling frames and frame quality

A frame can be evaluated using four questions:

• Coverage: Does it include all, or nearly all, eligible units?
• Duplicates: Does one population unit appear more than once?
• Accuracy: Are contact details and eligibility information current?
• Accessibility: Can units be contacted lawfully and practically?

Frame undercoverage occurs when eligible units are absent. Overcoverage occurs when ineligible units are included. Duplication gives some units a higher chance of selection unless corrected. Frame error is especially important in digital research: an online panel may exclude people with limited internet access, while a customer database may exclude nonmembers and former customers.

2. Probability and Non-Probability Sampling

The central distinction is whether the probability of selection is known and greater than zero for every population unit.

2.1 Probability sampling

In probability sampling, selection follows a random mechanism, and each eligible unit has a known, non-zero probability of inclusion. This supports statistical estimation of sampling error and more defensible population inference.

#### Simple random sampling

Every element has an equal chance of selection. A random-number generator selects units from a complete frame. The method is conceptually simple, but it requires a good frame and can be expensive when the population is geographically dispersed.

#### Systematic sampling

After arranging the frame, the researcher selects every kth unit after a random start. If the frame contains N units and the desired sample is n, the approximate interval is:

Here, N is the number of frame units and n is the required sample size. Systematic sampling is efficient, but periodic patterns in the frame can create bias. For example, selecting every tenth customer could overrepresent a recurring transaction pattern.

#### Stratified random sampling

The population is divided into mutually exclusive strata, such as region, age group, or customer type, and a random sample is drawn within each stratum. Stratification can improve precision when members within strata are relatively similar and the strata differ from one another. It also ensures representation of strategically important groups.

Proportional allocation selects sample sizes in proportion to stratum population sizes. Disproportionate allocation deliberately oversamples smaller or high-priority groups so that they can be analyzed. Such samples generally require weighting when producing population estimates.

#### Cluster and multistage sampling

In cluster sampling, naturally occurring groups—such as neighborhoods, stores, schools, or postal sectors—are selected, and units within selected clusters are surveyed. It reduces travel and administrative costs but often reduces statistical efficiency because units in the same cluster tend to resemble one another.

Multistage sampling combines stages, for example: select regions, then districts, then households, then one adult per household. It is practical for large populations but requires careful calculation of selection probabilities and fieldwork procedures.

#### Area sampling

Area sampling uses geographic boundaries as sampling units. It is useful when no comprehensive individual-level frame exists. Maps, address points, or census enumeration areas may be used to select locations and then households.

2.2 Non-probability sampling

In non-probability sampling, selection probabilities are unknown. It is often cheaper and faster, and it may be appropriate for exploratory research, hard-to-reach groups, qualitative work, or situations where no usable frame exists. However, conventional margins of sampling error cannot be justified solely from the sample size.

Common forms include:

• Convenience sampling: recruiting accessible participants, such as students or visitors to a website.
• Judgment or purposive sampling: selecting cases that meet a researcher’s substantive criteria.
• Quota sampling: filling category quotas, such as age and gender targets, without random selection within categories.
• Snowball sampling: asking participants to refer other eligible participants; useful for hidden or networked populations.
• Volunteer or opt-in sampling: allowing people to choose whether to participate, often through an open link.

Non-probability methods should not automatically be treated as invalid. Their value depends on the research purpose, transparency, recruitment quality, and limits placed on generalization. A quota sample matching the population on age and gender can still differ on unmeasured characteristics such as political interest, digital behavior, or brand involvement.

2.3 Choosing a design

The design decision should consider the objective, population, available frame, required precision, subgroup analysis, budget, time, and ethical constraints.

3. Sample Size and Precision

Sample size should be determined by the precision required, the variability of the measure, the confidence level, design complexity, expected nonresponse, and analytical objectives. “Larger is always better” is incomplete: a large biased sample can be worse than a smaller well-designed sample.

3.1 Estimating a proportion

For a simple random sample estimating a population proportion, the initial sample-size formula is:

where:

• n₀ = required sample size before finite-population adjustment;
• z = standard-normal critical value for the chosen confidence level;
• p = anticipated population proportion;
• e = desired maximum margin of error, expressed as a proportion.

The formula assumes a probability sample, independent observations, a sufficiently large population, and a simple random design. It is most conservative when p = 0.50 because p(1 − p) is then largest. Common two-sided critical values are approximately 1.645 for 90% confidence, 1.96 for 95%, and 2.576 for 99% confidence.

For a 95% confidence level, p = 0.50, and e = 0.05:

The result is rounded upward to 385 respondents. This is a planning value, not a guarantee that every estimate will have exactly a ±5 percentage-point error. The margin varies by estimate and is affected by design and nonresponse.

If the population is finite and relatively small, the finite-population correction can be applied:

where N is the population size and n is the adjusted sample size. If N = 2,000 and n₀ = 385:

Thus, approximately 323 completed responses are required under the stated assumptions.

3.2 Means and other measures

For a population mean, a preliminary formula is:

where σ is the anticipated population standard deviation and e is the desired margin of error in the measurement’s units. For a satisfaction scale, for example, e must be expressed in scale points. Since σ is often unknown, researchers use a prior study, a pilot survey, or a conservative estimate.

3.3 Allowing for nonresponse

Sample-size formulas usually refer to completed cases. If the expected completion rate is r, the number of invitations required is:

If 400 completes are required and the expected completion rate is 0.25, approximately 1,600 eligible invitations are needed. This calculation assumes the rate is defined consistently; in practice, eligibility, contact, and completion rates should be tracked separately.

3.4 Design effect and clustering

Complex designs may produce less precision than a simple random sample of the same size. The design effect is commonly represented as deff. A planning approximation is:

For equal-sized clusters, a frequently used approximation is:

where m is the average number of sampled elements per cluster and ρ is the intracluster correlation. This approximation assumes roughly equal cluster sizes and a comparable outcome correlation within clusters. If m = 10 and ρ = 0.05, deff ≈ 1.45. A simple-random target of 385 would therefore require approximately 559 completed cases before other adjustments. Researchers should use survey-statistics software for complex designs.

3.5 Subgroup requirements

A total sample may be adequate for an overall estimate but inadequate for subgroup comparisons. If managers need separate estimates for regions, customer types, or age groups, each subgroup needs enough observations for the intended analysis. Oversampling can improve subgroup precision, but it changes the sample’s raw composition and normally requires weights for overall estimates.

4. Sampling Error, Confidence, and Bias

Sampling error is random variation arising because a sample rather than a census is observed. It decreases, generally, as sample size increases, but it does not disappear entirely unless a census is conducted.

For a simple random sample estimating a proportion p̂, the estimated standard error is approximately:

A large-sample two-sided confidence interval is:

The interval expresses a procedure’s long-run coverage property: across repeated samples, approximately 95% of intervals constructed using z = 1.96 would contain the true parameter under the model assumptions. It does not mean that there is a 95% probability that a fixed parameter lies inside one particular interval in the classical interpretation.

If 240 of 400 respondents select a brand, p̂ = 0.60. Then:

The approximate 95% interval is 0.60 ± 1.96 × 0.0245, or approximately 0.552 to 0.648. For proportions near zero or one, or for small samples, methods such as the Wilson interval are generally preferable to the simple Wald interval.

Sampling error is only one source of uncertainty. Nonsampling errors include coverage error, nonresponse, measurement error, processing error, interviewer effects, and questionnaire problems. A probability sample does not automatically remove these problems.

4.1 Coverage error

Coverage error occurs when the frame excludes eligible units, includes ineligible units, or gives some units unequal access to selection. Undercoverage of low-income households, for example, may bias estimates if income is associated with product usage.

Possible remedies include combining frames, updating contact information, using mixed modes, screening for omitted groups, and reporting limitations. Combining frames requires attention to duplicates and unequal selection probabilities.

4.2 Nonresponse and self-selection

Unit nonresponse occurs when selected eligible units do not participate. Item nonresponse occurs when respondents omit particular questions. Nonresponse becomes a bias problem when participants and nonparticipants differ on the outcomes of interest and response behavior is related to those outcomes.

A high response rate is desirable but is not a sufficient measure of quality. A low response rate can sometimes yield limited bias if respondents resemble nonrespondents; a high response rate can still be biased if the frame excludes an important group.

Researchers can reduce nonresponse through credible invitations, short and mobile-friendly questionnaires, appropriate incentives, reminders, flexible modes, clear privacy information, and convenient scheduling. Refusal conversion should be conducted only where ethically and operationally appropriate.

4.3 Measurement and interviewer effects

Recruitment and fieldwork can influence responses. Interviewers may unintentionally signal expectations, and participants may give socially desirable answers. Interviewer training, standardized scripts, neutral probing, privacy, and quality monitoring help reduce these effects.

5. Recruitment and Fieldwork Planning

A fieldwork plan converts the design into controlled operations. It should specify:

1. target population and eligibility rules;
2. frame source and selection procedure;
3. sample quotas or probability allocations;
4. contact modes and call or invitation schedule;
5. incentive policy;
6. interviewer or recruiter training;
7. expected productivity and timeline;
8. quality-control checks;
9. data-security procedures;
10. escalation rules and reporting responsibilities.

5.1 Recruitment materials

An invitation should state who is conducting the research, the general purpose without leading participants, expected duration, eligibility, voluntary nature, confidentiality or privacy arrangements, incentive terms, and contact information. It should not misrepresent the sponsor or imply that participation is required.

For business-to-business studies, the sampling unit may be an organization while the respondent is a knowledgeable employee. Screening should verify both organizational eligibility and respondent knowledge. Multiple respondents from one organization may violate independence assumptions unless the design explicitly accommodates them.

5.2 Fieldwork monitoring

A live fieldwork dashboard should monitor issued contacts, attempted contacts, eligible cases, completes, refusals, breakoffs, quota cells, mode, interviewer, and date. Rates should be defined clearly. For example:

• Contact rate: proportion of sampled units successfully reached.
• Eligibility rate: proportion of contacted units found eligible.
• Completion rate: proportion of eligible units completing the study.
• Breakoff rate: proportion beginning but not finishing.

Organizations use several formal response-rate definitions, and rates should not be compared unless their numerators and denominators are equivalent.

Quality checks may include duplicate detection, response-time review, straight-line or invariant-pattern checks, attention checks used cautiously, geolocation or device checks where lawful, open-ended response review, audio monitoring with consent, and back-checks of a sample of interviews. Quality control should identify poor data without excluding unusual but genuine respondents merely because their answers are inconvenient.

5.3 Research logistics

Logistics include questionnaire programming, translation, device compatibility, interviewer schedules, sample release batches, incentive distribution, data transfer, backup, and incident management. A pilot should test recruitment, screen-out rates, question comprehension, survey length, routing, and operational assumptions—not merely spelling.

Translation should use a structured process such as independent translation, review, reconciliation, and pretesting. Cultural adaptation may be needed for examples, currencies, product categories, and response scales.

6. Weighting and Adjustment

Weighting assigns different influence to observations. It is used to compensate for unequal selection probabilities, disproportionate stratification, differential response, or known differences between the achieved sample and a reference population.

The base weight for unit i is commonly the inverse of its inclusion probability:

where πᵢ is the unit’s probability of inclusion. In multistage designs, πᵢ may be the product of selection probabilities across stages. A nonresponse adjustment may then be applied within reasonably homogeneous adjustment cells. Finally, calibration or post-stratification may align weighted totals with known population totals for variables such as age, region, or gender.

Weights should be documented, including their source, construction, trimming rules, and effect on estimates. Extreme weights can increase variance. Weight trimming reduces instability but may introduce some residual discrepancy. Weighting also cannot repair variables that were completely absent from the frame or correct inaccurate measurement.

For a weighted proportion, an estimator can be written as:

where yᵢ equals 1 when unit i has the characteristic and 0 otherwise. Standard errors for weighted and clustered data should use appropriate survey procedures rather than simple-random formulas.

7. Ethics, Privacy, and Governance

Sampling and fieldwork involve personal data, vulnerability, and power relationships. Researchers should obtain informed consent where required, collect only necessary information, protect contact lists, restrict access, and specify retention and deletion procedures. Sensitive questions require a legitimate purpose and careful wording.

Participants should understand that research participation is voluntary and distinct from sales promotion. “Sugging”—selling under the guise of research—damages trust and violates professional standards. Incentives must not be so large that they create undue pressure, especially for financially vulnerable groups.

Children and other protected populations require additional safeguards, including appropriate permissions and age-sensitive procedures. AI-assisted recruitment, fraud detection, or respondent profiling should be transparent, proportionate, tested for discriminatory effects, and subject to human oversight. Automated exclusion should not rely on opaque signals that systematically remove legitimate participants.

8. Contemporary Developments and Critical Evaluation

Modern researchers increasingly use mixed-mode designs combining web, mobile, telephone, in-person, and app-based data collection. Mixed modes can improve coverage and convenience, but mode effects may alter answers because question presentation, privacy, and interviewer presence differ.

Online access panels enable rapid recruitment but may suffer from professional respondents, panel conditioning, duplicate participation, and uncertain population coverage. Passive behavioral data and transaction records can complement survey data, yet they measure observed behavior rather than attitudes and may contain consent, identity, and interpretation problems.

Machine learning can support sample management by predicting contact success or identifying likely duplicates. It should assist—not replace—methodological judgment. A model optimized for completion may over-recruit highly responsive people and worsen representativeness. Every automated procedure should be validated against the research objective and audited for subgroup performance.

The most important critical principle is that representativeness is multidimensional. Matching a sample to population benchmarks on a few demographics does not prove that the sample is representative of opinions, behaviors, or unmeasured characteristics. Researchers should make claims no broader than the design and evidence justify.

Applied Case: Launch Research for a Plant-Based Beverage

Background

A beverage company plans to launch a plant-based drink in three metropolitan regions. It wants to estimate trial intention, identify promising segments, and compare the regions. The target population is adults aged 18 or above who purchased a non-dairy beverage in the past three months.

Problem

The client initially proposes an open social-media survey because it is inexpensive. However, an opt-in survey would likely overrepresent highly engaged consumers and people interested in plant-based products. The client also needs regional comparisons, while the available loyalty database covers only existing customers.

Analysis

The research team combines two approaches. For a population estimate, it uses a stratified probability sample of households and screens for category purchase. Region is the primary stratum, with additional monitoring of age and gender. Because the company needs regional estimates, the regions receive sufficient sample sizes even though their population sizes differ. A separate purposive component recruits heavy category users for exploratory interviews; those interviews are not used to estimate population incidence.

A pilot indicates a 22% eligibility rate among contacted adults and a 30% completion rate among eligible people. To obtain 1,200 completed surveys, the team therefore plans approximately:

The team releases the sample in batches rather than all at once, allowing it to inspect response patterns and correct operational problems without altering the selection rules improperly.

Approach and quality control

Invitations explain that the study is independent of sales promotion. The questionnaire is mobile-tested and includes neutral descriptions of the product. Interviewers receive standardized screening and probing instructions. The dashboard monitors region, age, gender, contact outcomes, breakoffs, and completion time. A small number of cases are back-checked with consent. Duplicate records are removed according to predefined rules.

Because some age-region cells respond less frequently, nonresponse adjustments are calculated within appropriate cells, and final weights align the completed sample with known regional demographic totals. The report distinguishes weighted population estimates from findings from the purposive interviews.

Outcome

The company obtains more credible regional estimates than it would have received from the open link. The exploratory interviews provide language and motivations useful for positioning, but the team does not claim that interview participants represent the market. The final report presents confidence intervals for probability-based estimates, describes fieldwork limitations, and warns that weighting cannot correct possible undercoverage of adults without stable addresses.

Learning points

The case demonstrates that sample quality depends on design coherence. The target population, frame, selection method, recruitment operations, weighting, and claims must fit together. Separate objectives may require separate samples, and inexpensive recruitment is not necessarily low-cost if it produces misleading decisions.

Best-Practice Checklist

Before launching fieldwork, confirm that:

• the target population is specific and defensible;
• the sampling frame and its exclusions are documented;
• every selection stage has a clear procedure;
• sample size reflects precision and subgroup needs;
• expected nonresponse is included in recruitment planning;
• screening and consent materials are neutral and lawful;
• pilot testing has examined both measurement and logistics;
• fieldwork indicators and stop rules are defined;
• quality checks protect data integrity without arbitrary exclusion;
• weighting and variance estimation match the design;
• limitations are reported with the findings;
• personal data and incentives are governed ethically.

Chapter Summary

Sampling design determines how a researcher moves from a defined population to evidence about that population. A target population must be distinguished from the frame, sampling unit, element, and respondent. Probability sampling provides known selection probabilities and supports statistical inference, while non-probability sampling can be useful for exploration, qualitative research, and hard-to-reach groups but requires cautious claims.

Sample size depends on desired precision, confidence level, variability, finite population size, design effect, subgroup requirements, and expected nonresponse. The familiar proportion formula is valid under stated probability-sampling assumptions; clustering, weighting, and complex designs require specialized analysis. Confidence intervals describe uncertainty from sampling, but they do not account automatically for coverage, nonresponse, measurement, or processing errors.

Effective fieldwork combines sound recruitment, pilot testing, trained personnel, monitoring, quality control, privacy protection, and documented logistics. Weighting can address unequal selection and some response differences, but it cannot create information about completely omitted groups or repair poor measurement. Sound marketing research therefore depends on matching design, operations, analysis, and the strength of the claims made.

Key Terms

Target population: Complete population about which the study seeks conclusions.

Sampling frame: Operational list or procedure used to identify population units.

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

Element: Basic unit about which data are collected.

Respondent: Person who provides information in a study.

Probability sampling: Sampling with known, non-zero selection probabilities.

Non-probability sampling: Sampling in which selection probabilities are unknown.

Simple random sampling: Equal-probability random selection from a frame.

Stratified sampling: Random sampling separately within defined population strata.

Cluster sampling: Selection of groups followed by study of units within them.

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

Quota sampling: Non-random recruitment to fill specified category totals.

Sampling error: Random difference between a sample estimate and population value.

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

Nonresponse bias: Bias arising when participation relates to study outcomes.

Design effect: Loss or gain in precision caused by a complex sample design.

Weighting: Assigning unequal analytical influence to observations.

Confidence interval: Interval produced by a procedure designed to cover a parameter at a stated long-run rate.

Finite-population correction: Adjustment recognizing reduced sampling variability when a substantial fraction of a finite population is sampled.

Fieldwork quality control: Procedures used to monitor recruitment, interviewing, data integrity, and compliance.

Review Questions

Short Answer Questions

1. What is the difference between a target population and a sampling frame?
2. Distinguish a sampling unit, an element, and a respondent.
3. Why might a researcher use stratified sampling?
4. What is the difference between coverage error and nonresponse bias?
5. Why does weighting not automatically make a sample representative?

Long Answer Questions

1. Compare probability and non-probability sampling, including their advantages, limitations, and appropriate marketing applications.
2. Explain the main probability sampling designs and discuss how cost and precision influence their selection.
3. Derive and interpret the main components of the sample-size formula for estimating a population proportion.
4. Develop a fieldwork quality-control plan for a national consumer survey.
5. Explain how sampling error, nonsampling error, confidence intervals, and weighting should be presented in a research report.

Analytical / Application Questions

1. A company has a database of 50,000 customers but knows that 15% of current customers are not enrolled. Identify likely coverage problems and propose a sampling strategy.
2. A survey requires 600 completed interviews, and the expected eligibility rate is 40% with a 25% completion rate among eligible people. Calculate the approximate number of initial contacts required and identify operational assumptions.
3. A regional study oversamples a small rural region to obtain reliable subgroup results. Explain why weighting is needed for overall estimates and identify two risks associated with extreme weights.

Multiple Choice Questions with Answers

Question 1

Which statement best defines a sampling frame?

A. The complete theoretical population about which conclusions are desired

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

C. The number of completed questionnaires required for analysis

D. The person who answers the research questionnaire

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

Question 2

Which sampling method divides a population into groups and randomly samples within each group?

A. Cluster sampling

B. Snowball sampling

C. Stratified sampling

D. Convenience sampling

Correct Answer: C. Stratified sampling

Question 3

Using a 95% confidence level, p = 0.50, and a margin of error of 0.05, the initial sample size for a simple random sample is approximately:

A. 100

B. 196

C. 385

D. 1,000

Correct Answer: C. 385

Question 4

A study needs 400 completed interviews and expects a completion rate of 25% among eligible invitations. How many eligible invitations are required?

A. 100

B. 400

C. 1,000

D. 1,600

Correct Answer: D. 1,600

Question 5

What is the principal statistical advantage of probability sampling?

A. It guarantees no measurement error

B. It makes every sample exactly representative

C. It provides known selection probabilities that support inference

D. It always costs less than non-probability sampling

Correct Answer: C. It provides known selection probabilities that support inference

Question 6

Which situation is an example of coverage error?

A. An eligible person refuses to participate

B. A respondent skips a satisfaction question

C. A customer database excludes nonmembers of the loyalty program

D. An interviewer records an answer incorrectly

Correct Answer: C. A customer database excludes nonmembers of the loyalty program

Question 7

If the average cluster size is m = 10 and the intracluster correlation is ρ = 0.05, the approximate design effect is:

A. 0.50

B. 1.05

C. 1.45

D. 2.00

Correct Answer: C. 1.45

Question 8

The base weight for a unit with inclusion probability πᵢ is commonly defined as:

A. πᵢ

B. 1 ÷ πᵢ

C. πᵢ²

D. 1 − πᵢ

Correct Answer: B. 1 ÷ πᵢ

Question 9

Which statement about a 95% confidence interval is most accurate in classical statistical terms?

A. There is a 95% probability that the fixed parameter lies in this particular interval

B. Across repeated samples, the procedure would capture the parameter approximately 95% of the time

C. It includes all forms of survey bias

D. It proves that the sample is representative

Correct Answer: B. Across repeated samples, the procedure would capture the parameter approximately 95% of the time

Question 10

Why might a researcher use a purposive sample of heavy category users alongside a probability survey?

A. To replace all population estimates with anecdotal evidence

B. To obtain exploratory insight from strategically relevant participants

C. To eliminate the need for screening

D. To guarantee a known margin of sampling error

Correct Answer: B. To obtain exploratory insight from strategically relevant participants