Quantitative Data Collection and Survey Management

Introduction

Quantitative market research converts observations, attitudes, behaviours and market outcomes into systematically structured data that can be compared, modelled and used for managerial decisions. In brand management, quantitative research supports decisions about segmentation, brand positioning, communication effectiveness, customer experience, pricing, product development and brand equity.

A survey is not simply a questionnaire distributed to respondents. It is a complete measurement and operational system involving a target population, sampling frame, recruitment method, instrument, mode of administration, fieldwork controls, data-processing rules and quality-assurance procedures. A poorly designed survey can produce precise-looking but misleading results; a well-designed survey makes the limitations of its evidence explicit and manages them systematically.

This chapter examines online, mobile, telephone and face-to-face surveys, as well as experiments embedded in survey programmes. It also addresses panel management, programming, piloting, response quality, fraud detection, missing data, cleaning and fieldwork-vendor governance.

Learning Objectives

After studying this chapter, students should be able to:

1. Select an appropriate survey or experimental mode for a research problem.
2. Translate research objectives into measurable variables and a programmed instrument.
3. Design sampling, recruitment and panel-management procedures.
4. Plan pilots and fieldwork using quality, cost, speed and ethical criteria.
5. Detect low-quality, fraudulent or ineligible responses without unjustified exclusion.
6. Apply defensible procedures for missing data, coding, cleaning and documentation.
7. Manage external fieldwork vendors through specifications, controls and deliverables.
8. Evaluate the validity, reliability and managerial usefulness of survey evidence.

1. The Survey as a Total Research Process

A quantitative survey normally follows this sequence:

Decision problem → information requirements → constructs and variables → sample design → instrument design → programming → pilot → fieldwork → quality control → cleaning and coding → analysis → reporting.

Errors may enter at every stage. A useful distinction is between sampling error, which results from observing a sample rather than the entire population, and nonsampling error, which includes coverage error, nonresponse, measurement error, processing error and fraudulent participation. Increasing sample size reduces sampling variability but does not automatically correct a biased questionnaire, an incomplete sampling frame or fabricated interviews.

1.1 Population, frame and sample

The target population is the group to which the researcher intends to generalise. The sampling frame is the operational list or access mechanism used to reach members of that population. The sample is the set of selected or recruited units that provide data.

For example, “all adults who purchased a premium skincare product in the last six months” is a target population. A consumer panel claiming to contain such adults is an access mechanism, not necessarily a perfect frame. Screening questions must establish eligibility, but self-reported eligibility can itself be inaccurate.

1.2 Probability and non-probability approaches

In probability sampling, each population unit has a known, non-zero probability of selection. Simple random, systematic, stratified and cluster sampling are examples. Probability designs support stronger design-based inference, although nonresponse and coverage problems remain.

Non-probability approaches include quota sampling, convenience sampling, purposive recruitment and opt-in online panels. They can be efficient and appropriate for exploratory or commercially time-sensitive research, but sample composition and recruitment mechanisms must be disclosed. Weighting may improve alignment with known population characteristics, but weighting cannot guarantee removal of unmeasured selection bias.

1.3 Sample-size planning

For a simple random sample estimating a population proportion, the approximate margin of error is:

where E is the margin of error, z is the standard-normal critical value for the chosen confidence level, p is the anticipated population proportion, and n is the completed sample size. The formula assumes a probability sample, independent observations and a large enough population for the finite-population correction to be negligible. It describes sampling variability only; it does not quantify nonresponse or measurement bias.

When no prior estimate of p is available, p = 0.5 gives the largest variance and therefore a conservative sample-size requirement. For a 95% confidence level, z ≈ 1.96. If p = 0.5, E = 0.05, then:

This is not a universal target. Subgroups, design effects, expected incidence, quota constraints, weighting and the decision consequences of error must also be considered. Clustered or heavily weighted samples generally have greater variance than an equally sized simple random sample.

2. Choosing the Survey Mode

Mode affects coverage, cost, response behaviour, privacy, stimulus presentation and data quality. The correct choice follows from the population and task rather than from fashion.

2.1 Online and mobile surveys

Online surveys may use probability-based recruitment, customer databases, social recruitment or access panels. The researcher should record recruitment source, device type, completion time, location or region where appropriate, and consent status.

A mobile-first questionnaire uses short screens, readable fonts, touch-compatible scales, limited matrix questions and visible progress information. Long grids, hover-dependent instructions and dense stimuli should be avoided. A responsive design is not merely a smaller desktop questionnaire: it changes how respondents read, remember and answer.

Mobile research can collect contextual data such as photographs, barcode scans or geolocation, but these features require explicit consent, data minimisation and security controls. Location data should not be collected simply because the technology permits it.

2.2 Telephone surveys

Telephone interviewing can be computer-assisted. The programme displays the question, records answers, manages routing and may schedule callbacks. Interviewer training should cover neutral probing, pronunciation, refusal conversion, privacy and the handling of sensitive answers.

Call attempts should follow a documented protocol that varies time and day where appropriate. A contact outcome taxonomy distinguishes completed interviews, partial interviews, refusals, ineligible contacts, no answer, busy signals and answering machines. These outcomes are more informative than reporting only the number of completed interviews.

2.3 Face-to-face surveys

Face-to-face studies may be conducted in homes, workplaces, retail locations or controlled facilities. Location can shape the sample: a shopping-centre intercept reaches visitors, not all category users. Interviewer selection of respondents, visible branding and the physical presence of other people may influence answers. Private administration and standardised introductions reduce—but do not eliminate—these effects.

3. From Research Objectives to a Measurement Instrument

A high-quality instrument begins with a construct map. A construct is an abstract concept such as trust, perceived quality or brand attachment. It must be translated into observable indicators. The researcher should distinguish:

• Behavioural measures: purchase, usage, search or recommendation behaviour.
• Attitudinal measures: beliefs, evaluations, preferences and intentions.
• Descriptive variables: demographics, firmographics, usage context and category involvement.
• Derived variables: indices, segments or scores calculated from multiple items.

3.1 Measurement quality

Validity concerns whether a measure supports the intended interpretation. Content validity asks whether relevant aspects of a construct are represented. Criterion-related validity compares a measure with an appropriate external criterion. Construct validity considers convergent and discriminant relationships with other measures.

Reliability concerns consistency. Internal consistency may be assessed for multi-item scales, but a high consistency coefficient does not prove that the scale measures the correct construct. Test–retest reliability may be inappropriate when attitudes are expected to change.

A scale should have clear anchors and a justified number of response categories. Balanced agree–disagree scales can be useful, but agreement formats may encourage acquiescence. Frequency, likelihood and importance should not be treated as interchangeable dimensions.

3.2 Question wording and order

Questions should be specific, answerable and neutral. Avoid double-barrelled questions such as “How satisfied are you with the product’s quality and value?” Use a defined reference period, such as “in the past 30 days,” when recall is important. Provide “not applicable,” “don’t know” or “prefer not to answer” when these are substantively meaningful—not merely to avoid difficult design decisions.

Question order can create context effects. Early category questions may prime later brand evaluations; asking about complaints may reduce subsequent satisfaction ratings. A common design is screening, behavioural context, evaluation, intentions, demographics and debriefing, but the order should be tested against the research purpose.

3.3 Experimental elements

Surveys may contain experiments, including randomised concept exposure, message tests, conjoint tasks or A/B comparisons. Random assignment improves causal interpretation when treatment is properly implemented and other conditions are held constant. The instrument must record treatment assignment, exposure and completion status.

For a simple difference in mean outcomes between treatment and control groups, the estimated effect is:

where ȳ₁ is the sample mean for the treatment group and ȳ₀ is the sample mean for the control group. Causal interpretation requires valid randomisation, adequate compliance, comparable measurement and no influential interference between participants. Randomisation does not repair a biased outcome measure or a sample that cannot support the intended population claim.

4. Panel and Sample Management

A panel is a managed pool of individuals or organisations available for research participation. Panel management involves recruitment, profiling, consent, identity control, incentive administration, engagement, deduplication, suspension and retirement.

4.1 Recruitment and profiling

Recruitment sources should be documented because source affects motivation and composition. Profile variables should be limited to information that supports sampling or routing. Sensitive characteristics require a clear purpose, lawful basis or valid consent, restricted access and retention controls.

Panels should avoid excessive invitation frequency. High burden can produce professional respondents, rushed answers and systematic attrition. Incentives should compensate time fairly without becoming coercive, particularly for vulnerable participants.

4.2 Quotas and weighting

Quotas can control achieved composition by age, gender, region, category usage or other variables. Quotas are operational controls, not proof of representativeness. If the sample is weighted to population targets, the weighting variables, trimming rules and effective sample-size implications should be reported.

A basic weighted estimate of a mean is:

where yᵢ is respondent i’s observed value and wᵢ is that respondent’s analysis weight. Weights must be defined consistently; arbitrary or extreme weights can increase variance and make results unstable.

5. Instrument Programming and Piloting

Programming is part of measurement, not an administrative afterthought. The programmed survey should implement the approved questionnaire exactly, while making technical constraints visible to the research team.

A programming specification should include:

• question IDs and variable names;
• response labels and value codes;
• mandatory, optional and “prefer not to answer” fields;
• skip logic and display logic;
• randomisation and rotation rules;
• piping and calculated fields;
• validation limits and error messages;
• stimulus files and version numbers;
• device and browser requirements;
• termination conditions;
• data exports and audit metadata.

5.1 Cognitive and technical pilots

A cognitive pilot examines comprehension, recall, judgement and response mapping through interviews or close observation. A technical pilot examines routing, rendering, data capture, timing, quotas, links, randomisation and export integrity. A soft launch collects a limited number of live cases before full release.

Pilot results should be reviewed against predefined acceptance criteria. For example, a high break-off rate on one screen may indicate burden, confusion, a programming defect or an ineligible sample source. It should not be dismissed automatically as respondent carelessness.

Version control is essential. Changes after launch should be logged with date, reason, affected cases and whether earlier responses remain comparable. If a wording change alters measurement, the data may need to be segmented by instrument version.

6. Fieldwork Management and Vendor Governance

An external fieldwork vendor should receive a written statement of work. It should specify population, sample source, quotas, expected incidence, target completes, mode, questionnaire version, timetable, incentives, quality rules, privacy obligations, deliverables and escalation procedures.

The client should not rely solely on a vendor’s statement that interviews are “validated.” Quality must be auditable. Useful deliverables include anonymised respondent IDs, recruitment source, timestamps, paradata, disposition codes, treatment assignment, exclusion flags, weighting information and a fieldwork report.

A fieldwork dashboard may monitor:

• completes and remaining quota cells;
• incidence and screen-out rates;
• break-off rates;
• median and distribution of completion time;
• duplicate and suspicious-device rates;
• open-text quality;
• treatment-group balance;
• callback outcomes for telephone work;
• respondent complaints and privacy incidents.

Vendor incentives can create undesirable behaviour if payment is based only on volume. Contracts should reward valid, auditable completes and permit quality audits, recontact where ethically appropriate, corrective action and removal of invalid cases.

7. Response Quality and Fraud Detection

Low-quality responding is not identical to fraud. A respondent may misunderstand a question, rush because of distraction, use a translation tool, or provide an unusual but genuine answer. Fraud implies deceptive participation or manipulation, such as duplicate identities, automated submissions or fabricated eligibility.

No single flag should normally determine exclusion. A defensible approach combines independent indicators and documents the decision rule. Potential indicators include:

• implausibly short completion time relative to the instrument;
• repeated identical selections across long grids;
• contradictory screening and demographic answers;
• nonsensical or copied open-text responses;
• repeated device, network or identity patterns;
• failed attention checks;
• impossible timestamps or routing sequences;
• excessive participation frequency;
• straightlining where the item design makes it implausible.

An attention check should test whether the respondent followed an instruction, not whether the respondent selected the researcher’s preferred opinion. Location or device signals are probabilistic and may produce false positives in shared networks, privacy-protected browsers or legitimate travel situations.

A practical quality-review protocol is:

1. Define quality indicators before inspecting outcomes.
2. Separate eligibility failures, technical failures and quality concerns.
3. Examine the full response pattern rather than one flag.
4. Preserve the original record and create an exclusion field.
5. Apply the same rule across treatment groups and sample sources.
6. Record counts, reasons and sensitivity results.

Machine-learning fraud models may identify complex patterns, but they require labelled data, monitoring for subgroup bias and human review. Automated exclusion without explainability is poor governance.

8. Missing Data, Coding and Cleaning

Missingness has different meanings. A respondent may refuse, select “don’t know,” be legitimately skipped, abandon the survey or experience a technical failure. These states should not be collapsed into one blank value.

Three common assumptions are:

• MCAR (Missing Completely At Random): missingness is unrelated to observed and unobserved values.
• MAR (Missing At Random): conditional on observed information, missingness is unrelated to the missing value.
• MNAR (Missing Not At Random): missingness depends on the unobserved value even after considering observed information.

These are assumptions about the missing-data mechanism, not labels that can be proven merely by inspecting a dataset. For example, income nonresponse may plausibly be related to income itself, making a simple complete-case analysis potentially biased.

Cleaning should occur through reproducible transformations rather than manual overwriting. A cleaning log should record variable, original value, rule, changed value, reason, date and analyst. Typical checks include range validation, legal skip patterns, duplicate IDs, inconsistent dates, impossible combinations, code-frame integrity and treatment assignment.

For a binary variable, a missingness rate may be reported as:

where M is the number of eligible cases missing the variable and N is the number of eligible cases for which the variable was expected. The denominator must be stated; using all records when a question was legitimately skipped misstates missingness.

Possible treatments include complete-case analysis, available-case analysis, explicit missing categories, single imputation and multiple imputation. The choice depends on the estimand, measurement scale, missingness assumptions and analysis model. Multiple imputation is generally preferable to simple mean substitution when its assumptions and implementation are defensible, because mean substitution artificially reduces variability and can distort relationships.

9. Ethics, Privacy and Governance

Survey research requires informed participation, proportionality and respect for autonomy. Participants should understand who is conducting the research where disclosure is required, what participation involves, how data will be used, whether incentives are provided and how withdrawal or complaints can be handled.

Data minimisation means collecting only data needed for a defined purpose. Personally identifiable information should be separated from analytical data wherever possible, protected in transit and at rest, accessed according to role and deleted according to a documented retention schedule. Cross-border transfers and sensitive categories require jurisdiction-specific review.

Deception, covert observation, recording, biometric data, geolocation and recontact require additional scrutiny. Research teams should distinguish market research from direct marketing: using survey responses to contact participants for sales without appropriate permission undermines trust and may violate applicable law or professional codes.

Researchers should also consider accessibility. Screen-reader compatibility, sufficient contrast, plain language, translated instruments and alternatives to visual or hearing-dependent tasks improve inclusion and reduce construct-irrelevant exclusion.

10. Contemporary Developments

Survey operations increasingly use APIs, real-time quota dashboards, device-responsive design, automated translation, synthetic testing and AI-assisted coding of open text. These tools can improve efficiency but do not replace methodological judgement.

Generative AI may assist with draft wording, programming checks, anomaly review or thematic coding. Human researchers must verify cultural appropriateness, preserve version control, protect confidential data and test whether AI systems introduce stereotypical or unequal treatment. Synthetic respondents are useful for debugging and scenario exploration, but they are not substitutes for empirical participants.

Privacy-enhancing technologies, first-party recruitment and authenticated panels are becoming more important as third-party identifiers decline. At the same time, stronger identity controls must be balanced against exclusion of legitimate respondents who use privacy tools or shared devices.

Applied Case: Illustrative Brand-Concept Test

Background

A fictional consumer-goods company, Northstar Foods, plans to launch a premium low-sugar snack under an existing brand. Management wants to compare two advertising concepts and estimate whether the stronger concept improves purchase intention among regular category buyers.

Problem

The initial proposal is a 20-minute desktop survey recruited through an unrestricted online link. It has a quota for age and gender but no defined purchase-eligibility rule, no treatment randomisation audit and no plan for suspicious responses. This design risks duplicate participation, poor mobile completion and ambiguous interpretation.

Analysis and approach

The research team defines the target population as adults who purchased a relevant snack category in the past three months. It specifies a mobile-first online instrument with a screening question, category behaviour, baseline brand familiarity, random assignment to Concept A or Concept B, advertisement exposure, purchase intention, perceived fit and demographics.

A pilot tests comprehension and confirms that both concepts render equivalently on common devices. The programme records randomisation, exposure completion and version number. Quotas are set for age, gender and region, while recruitment source is monitored rather than hidden.

Before launch, the team defines quality rules: eligibility must be consistent, duplicate indicators trigger review, open text must be meaningful where requested, and completion time is reviewed against the pilot distribution rather than an arbitrary universal threshold. A respondent is excluded only when multiple indicators support the decision. Original records remain available for audit.

The analysis plan specifies the primary outcome in advance: the difference in mean purchase-intention score between treatment groups, with confidence intervals and subgroup analysis treated as secondary. Balance on baseline familiarity and category involvement is checked. Any weighting is documented and sensitivity analyses compare weighted and unweighted results.

Outcome

Concept B produces a higher average purchase-intention score, but the difference is concentrated among consumers with high category involvement. The team therefore recommends Concept B for the initial campaign, while revising the execution for low-involvement consumers. The report states that the experiment supports a comparative effect within the sampled eligible population; it does not claim that all national consumers will respond identically.

Learning points

The case shows that a credible result depends on more than sample size. Clear eligibility, controlled randomisation, mobile usability, pre-specified quality rules, transparent exclusions and cautious generalisation jointly protect the decision. The operational design is part of the evidence, not separate from it.

Challenges and Limitations

Survey research is vulnerable to declining participation, professional respondents, coverage gaps, satisficing, social desirability, recall error, panel conditioning and mode effects. Quotas may conceal uncontrolled differences within cells. Weighting can improve descriptive alignment while increasing variance. Experiments embedded in surveys may have stronger causal identification but weaker realism than natural market behaviour. Open-text AI coding can improve scale but may obscure nuance or reproduce classification bias.

Researchers should therefore report the sample source, recruitment, field dates, mode, eligibility, achieved completes, exclusions, missingness, weighting, quality controls, questionnaire version and important deviations from plan. Transparency enables readers and managers to judge the strength of the evidence rather than treating a percentage as self-validating.

Best-Practice Checklist

Before fieldwork, confirm that:

• the target population and estimand are explicit;
• the mode matches the population and task;
• constructs, variables and response scales are justified;
• the questionnaire is accessible and mobile-usable where relevant;
• routing, randomisation and quotas have been tested;
• pilots have addressed comprehension and technical integrity;
• consent, privacy and retention procedures are documented;
• quality rules were defined before reviewing outcomes;
• the vendor statement of work specifies auditable deliverables;
• cleaning and missing-data decisions are reproducible;
• reporting will distinguish sampling uncertainty from nonsampling limitations.

Chapter Summary

Quantitative data collection is a complete system extending from population definition to fieldwork, quality control and documented analysis. Online, mobile, telephone and face-to-face modes each offer different combinations of coverage, cost, interaction and measurement risk. Instrument quality depends on construct definition, neutral wording, suitable scales, controlled order and careful programming. Panels require transparent recruitment, responsible incentives, deduplication and privacy protection. Pilots should test both respondent comprehension and technical execution.

Response quality and fraud detection should use multiple indicators, consistent rules and human review. Missing-data treatment must distinguish legitimate skips, refusals, technical failures and unknown values, while analysis should state its assumptions. Vendors should be governed through precise specifications, audit trails and quality-based deliverables. Ethical research protects autonomy, privacy, accessibility and data security. The central managerial lesson is that reliable survey evidence is produced by the integration of methodological design and disciplined operations.

Key Terms

Sampling frame: The operational list or access mechanism used to reach a target population.

Coverage error: Bias caused when the frame excludes or overrepresents parts of the target population.

Nonresponse error: Bias associated with eligible units failing to participate or complete measurement.

Construct: An abstract concept represented through observable measures.

Validity: The extent to which evidence supports the intended interpretation of a measure.

Reliability: The consistency of a measurement process.

Panel: A managed pool of potential research participants.

Quota: A fieldwork target controlling the number or composition of completed cases.

Weight: An analysis factor used to give observations differing influence in an estimate.

Paradata: Operational data about the process of responding, such as timestamps or device type.

Soft launch: A limited initial release used to identify problems before full fieldwork.

Randomisation: Allocation to conditions using a chance mechanism.

Satisficing: Providing an adequate-looking answer with insufficient cognitive effort.

Fraudulent response: Data submitted through deceptive or otherwise invalid participation.

Missing completely at random (MCAR): Missingness unrelated to observed and unobserved values.

Missing at random (MAR): Missingness conditionally unrelated to the missing value given observed data.

Missing not at random (MNAR): Missingness related to the unobserved value after accounting for observed data.

Disposition code: A standard classification of a contact or participation outcome.

Design effect: The variance inflation or reduction associated with a complex sample relative to a reference design.

Review Questions

Short Answer Questions

1. Distinguish a target population, sampling frame and sample.
2. Why does a larger sample not automatically eliminate nonsampling error?
3. State two design requirements for a mobile-first survey.
4. Explain the difference between a cognitive pilot and a technical pilot.
5. Why should fraud detection normally combine several indicators?

Long Answer Questions

1. Compare online, mobile, telephone and face-to-face surveys for a national brand-tracking study. Recommend a mode and justify the trade-offs.
2. Explain how a research objective is translated into constructs, variables, questionnaire items and programmed data fields.
3. Discuss the role of quotas, weighting and panel management in improving—without guaranteeing—sample quality.
4. Explain MCAR, MAR and MNAR, and evaluate the risks of complete-case analysis and mean substitution.
5. Design a governance framework for an external survey fieldwork vendor, including specifications, quality controls and deliverables.

Analytical / Application Questions

1. A concept test shows a large difference between two advertisements, but the treatment group contains many more extremely fast completions and duplicate-device flags. How should the researcher investigate and report the result?
2. A mobile survey has high break-off on a long matrix question, while desktop completion is acceptable. Diagnose possible causes and propose a redesign and pilot plan.
3. A brand manager asks for a nationally representative claim from an opt-in quota panel weighted only by age and gender. Evaluate the claim and propose a more defensible reporting strategy.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes sampling error from nonsampling error?

A. Sampling error results from studying a sample, whereas nonsampling error includes coverage, measurement and processing problems.

B. Sampling error is caused only by fraudulent respondents.

C. Nonsampling error disappears when the sample size is increased.

D. Sampling error occurs only in face-to-face surveys.

Correct Answer: A. Sampling error results from studying a sample, whereas nonsampling error includes coverage, measurement and processing problems.

Question 2

In the approximate margin-of-error formula E = z × √[p(1 − p) / n], what does n represent?

A. The population size

B. The number of completed observations

C. The response rate

D. The weighting factor

Correct Answer: B. The number of completed observations

Question 3

What is the primary purpose of a technical pilot?

A. To prove that the research hypothesis is correct

B. To replace the need for a sampling plan

C. To test routing, rendering, randomisation and data capture

D. To calculate the final confidence interval

Correct Answer: C. To test routing, rendering, randomisation and data capture

Question 4

Which practice is most appropriate for detecting low-quality responses?

A. Exclude every respondent with an unusually short completion time

B. Use multiple indicators and document consistent review rules

C. Exclude all respondents using shared networks

D. Remove responses that disagree with the expected brand result

Correct Answer: B. Use multiple indicators and document consistent review rules

Question 5

What does random assignment in a survey experiment primarily support?

A. Causal comparison between treatment conditions, subject to valid implementation

B. Automatic national representativeness

C. Elimination of all measurement error

D. A guarantee of complete response rates

Correct Answer: A. Causal comparison between treatment conditions, subject to valid implementation

Question 6

Which missingness mechanism assumes that missingness may depend on the unobserved value itself?

A. MCAR

B. MAR

C. MNAR

D. Randomised missingness

Correct Answer: C. MNAR

Question 7

Why can mean substitution for missing values be problematic?

A. It always increases sample size without changing analysis

B. It can reduce variability and distort relationships among variables

C. It converts a probability sample into a census

D. It prevents all forms of nonresponse

Correct Answer: B. It can reduce variability and distort relationships among variables

Question 8

Which statement about quotas is correct?

A. Quotas prove that a sample is probability-representative

B. Quotas control selected composition targets but do not remove all selection bias

C. Quotas make weighting unnecessary in every study

D. Quotas are relevant only to telephone surveys

Correct Answer: B. Quotas control selected composition targets but do not remove all selection bias

Question 9

A weighted mean is calculated as ȳ_w = ∑(wᵢyᵢ) / ∑wᵢ. What does wᵢ denote?

A. Respondent i’s analysis weight

B. The number of questions answered by respondent i

C. The population variance

D. The survey’s response rate

Correct Answer: A. Respondent i’s analysis weight

Question 10

Which vendor deliverable most directly supports an audit of fieldwork quality?

A. Only the final topline percentages

B. A marketing slogan for the research report

C. Anonymised IDs, disposition information, paradata and exclusion records

D. A promise that all respondents were sincere

Correct Answer: C. Anonymised IDs, disposition information, paradata and exclusion records