A market research company succeeds when its findings accurately describe the people studied and honestly state how far those findings can be used. A dashboard containing hundreds of survey answers does not prove that everyone in Singapore agrees with the result. The company must protect participants, identify the target population, choose suitable evidence and show clients the limitations alongside useful insights.
This guide follows Harbourline Insights, a wholly fictional Singapore research firm hired by a café group to understand demand for a new weekday lunch option. The client’s marketing director wants a headline saying ‘Singapore prefers our new menu’. The firm sends a survey invitation to 1,000 existing loyalty members, obtains 400 responses and finds 248 respondents say they would consider the option. The issue is not the arithmetic. It is whether a self-selected loyalty-member response can be represented as the preference of an entire nation.
The Gold Standard master hub examines reliable enterprise decisions. The Advertising Companies guide covers commercial persuasion and advertising evidence, while Laboratory Testing Companies covers different forms of defensible measurement. Here the subject is the research company commissioning, sampling, interviewing, analysing and publishing evidence about actual people.
Quick decision routes for research buyers and researchers
Commissioning research? Start with who the sample represents and whether the questions answer your business decision. Running the research firm? Follow participant care, sampling, analysis, disclosure and defensible correction.
Chapter index
Research purpose, population and lawful participation · Chapters 1–11
- A research company is not a factory for persuasive statistics
- Separate market research from disguised selling or lead collection
- Identify the true research sponsor and responsible contractor
- Specify the population before collecting the first answer
- Know whether the sample is probability-based or self-selected
- Write questions that measure the intended concept
- Pilot the entire respondent journey, not just spelling
- Respect the respondent’s right to decline or stop
- Keep research and consent lawful under Singapore’s PDPA
- Protect children and other vulnerable participants specially
- Avoid collecting more identifiers than the project needs
Sampling integrity, uncertainty and appropriate analysis · Chapters 12–18
- Track invitations, starts, completes and exclusions separately
- Check interviewers and panels without manufacturing respondents
- Distinguish observed answers from population estimates
- Use weighting only where there is a defensible basis
- Do not attach a conventional margin of error to a convenience sample
- Use qualitative interviews for depth, not fake percentages
- Handle AI-generated summaries as draft interpretations
Reporting, AI oversight and client accountability · Chapters 19–23
1. A research company is not a factory for persuasive statistics
Market, opinion and social research can support product design, public understanding, service evaluation or planning. The researcher’s contribution is systematic evidence, not a predetermined conclusion. A client may ask a question for commercial reasons, but should not be sold a claim that the research cannot support.
Harbourline asks the café group what decision it faces: whether to trial a new lunch item at particular outlets, how much to prepare or whether customers understand the offer. Those are different from proving that Singaporeans prefer one brand. A research company should explain what an appropriately scoped study can answer and refuse to change its conclusions because the sales team wants a stronger poster.
Back to chapter index · Next chapter
2. Separate market research from disguised selling or lead collection
A research invitation may look similar to a promotional message, yet respondents have different expectations. Ethical research should not secretly turn a questionnaire into a sales appointment or disclose individual answers to marketing teams without a legitimate basis. A reward or customer-interest question does not remove these responsibilities.
The ICC/ESOMAR International Code, whose current page was updated 16 October 2026, stresses a clear distinction between research and non-research activity. Harbourline defines which data are for aggregate insights and whether any separate marketing contact is proposed. The company should not promise confidential research while giving the café a list of named people who selected ‘interested’.
Back to chapter index · Next chapter
3. Identify the true research sponsor and responsible contractor
A project may involve a brand commissioning the study, a research company designing it, a panel provider recruiting participants and an analytics vendor processing responses. These organisations have different roles and access rights. The sponsor should be identified accurately where required and responsibilities must not disappear into subcontracting.
Harbourline documents who designed the questionnaire, who invited the 1,000 loyalty members, who holds their contact details and who will review the final findings. It should not imply that every respondent independently volunteered from the general public when all invitations were supplied from a client’s customer database. Provenance is part of the method and a foundation of trustworthy interpretation.
Back to chapter index · Next chapter
4. Specify the population before collecting the first answer
A population might be current customers, prospective customers within a neighbourhood, Singapore adults, commuters or decision-makers in certain industries. The answer to one cannot automatically represent another. Research design needs a clear population definition and an account of who is excluded.
The café’s immediate operational decision concerns its own weekday customers near selected locations. Surveying current loyalty members may be useful for that decision, but it does not establish national consumer preference. AAPOR’s survey best practices emphasise identifying the target group and whether the available sampling frame covers it. Harbourline states that boundary before anyone buys the results.
Back to chapter index · Next chapter
5. Know whether the sample is probability-based or self-selected
Probability samples give eligible members of a defined population a known non-zero chance of selection under a specified frame. Opt-in and convenience samples use different recruitment methods. Both can be useful for certain questions, but their statistical interpretation and limitations are not identical.
Harbourline’s response group comes from a customer loyalty mailing and voluntary replies. It must not describe this as a random representative sample of Singapore. AAPOR’s disclosure standards require researchers to distinguish probability and nonprobability recruitment and explain the sample frame and its coverage. Large response count alone cannot correct a biased recruitment method.
Back to chapter index · Next chapter
6. Write questions that measure the intended concept
Wording, response choices and order can change the meaning of answers. Asking ‘How excited are you about our delicious new lunch?’ is different from asking whether a person would consider an accurately described new option. A company should avoid steering respondents while preserving clarity.
Harbourline tests alternative wording with small pilot interviews before sending its full survey. Its team checks that ‘consider trying’ is not reported as an actual purchase, and that a person can say neither or don’t know. The company’s obligation is to measure the question it claims to measure. A flattering question design may help marketing temporarily, but it weakens the decision the business was paying to improve.
Back to chapter index · Next chapter
7. Pilot the entire respondent journey, not just spelling
A questionnaire may work on a researcher’s laptop and fail on a small phone, omit a response choice or display one question after an unintended skip. Participants may misunderstand terminology or leave when a survey becomes burdensome. Quality work includes real pilot testing and appropriate accessibility review.
Harbourline tests the survey with representative devices and people before distributing all invitations. It checks consent information, question logic, completion, accessibility and the integrity of captured values. A pilot finding can justify revising the instrument, but responses gathered under incompatible versions should not be silently combined without competent methodological review. A sound company’s workflow records the version each participant saw.
Back to chapter index · Next chapter
8. Respect the respondent’s right to decline or stop
Participation in research should not rely on misleading pressure. The current ICC/ESOMAR framework highlights the right to decline, duty of care and additional safeguards for vulnerable people. Privacy and legal rules still depend on the actual activity and context.
Harbourline tells invitees the nature of the research, who is responsible, how information will be used and what their participation entails. A loyalty customer should not have to complete the study to obtain an already promised commercial service. The company designs a reasonable way to refuse and avoids implying that an unresponsive customer has expressed a particular opinion. Useful data begins with a credible relationship.
Back to chapter index · Next chapter
9. Keep research and consent lawful under Singapore’s PDPA
Research companies may process contact details, purchasing patterns and responses that constitute personal data. The correct legal basis depends on the purpose and actual statutory exceptions; consent is not the only possible legal basis for all analytics and research processing. Companies need competent assessment rather than a standard checkbox copied without thought.
The PDPC organisational guidance and the Selected Topics guidelines address analytics, research and anonymisation. Harbourline distinguishes permission to contact loyalty members about research from reuse of responses in unrelated marketing. It protects personal data and explains any applicable research exception accurately instead of claiming a universal exemption.
Back to chapter index · Next chapter
10. Protect children and other vulnerable participants specially
Research with children, young people or vulnerable adults may call for guardian permission, age-appropriate information and additional safeguards, depending on law and circumstances. Asking a minor to click an adult consent statement is not adequate evidence that the study is appropriate.
The ESOMAR/GRBN guideline addresses research involving children and vulnerable individuals. Harbourline’s lunch survey targets eligible adult respondents; a proposed school-student study would require a fresh ethical and legal assessment. Marketing does not gain the right to question children about personal matters simply because a research firm operates the online form.
Back to chapter index · Next chapter
11. Avoid collecting more identifiers than the project needs
A study may require eligibility screening but not a person’s full address, identity number or private medical history. Extra fields add risk and can discourage candid participation. The company should know who can see raw responses and when information can be aggregated or anonymised appropriately.
Harbourline needs to identify which locations respondents use, not automatically obtain their full identification documents. The firm uses proportionate account matching where necessary and limits access to raw details. If a report will contain only aggregate answers, the source files should still be protected according to their actual nature. A dataset is not genuinely anonymous merely because names were removed while rare combinations still reveal individuals.
Back to chapter index · Next chapter
12. Track invitations, starts, completes and exclusions separately
Participation can be measured at several stages: people invited, eligible recipients, survey starts, complete answers and cases excluded for valid reasons. Those counts need different denominators. An undefined response rate can make a survey appear stronger than its actual recruitment.
Harbourline sends 1,000 invitations and receives 400 usable responses in the fictional example. The simple response-to-invitation ratio is 40%, but not necessarily an AAPOR standard response rate, which can require different classification of eligible cases and nonresponse. The other 600 invitees are not proven to dislike the menu. A responsible company records what happened to invitations and which questions people answered rather than treating missing values as negative votes.
Back to chapter index · Next chapter
13. Check interviewers and panels without manufacturing respondents
Fieldwork can be damaged by duplicate accounts, fabricated interviews, inattentive answers, interviewer influence or unsuitable recruitment. Quality checks should be proportionate and preserve legitimate participants. An aggressive automated fraud filter may also discard genuine responses with atypical patterns.
Harbourline reviews fieldwork timing, duplicate signals, survey logic and recruitment conditions without inventing a perfect fraud test. The AAPOR best practices recommend quality monitoring across sample, questionnaire, interviewers and data processing. The research company should preserve the reason for excluding cases and avoid secretly discarding inconvenient but valid opinions to make its results more persuasive.
Back to chapter index · Next chapter
14. Distinguish observed answers from population estimates
If 248 of 400 respondents select an option, that is 62% of the observed response group. Whether it estimates another population depends on recruitment, coverage, measurement, weighting and additional assumptions. A percentage can be mathematically correct while the headline drawn from it is false.
Harbourline tells its client, ‘62% of 400 responding loyalty members would consider this lunch option in this survey.’ It does not say 62% of Singaporeans prefer it. The difference is not overly cautious wording; it protects the usefulness of the research. A manager deciding how many lunches to prepare needs a credible response group and realistic demand context, not an exaggerated national headline.
Back to chapter index · Next chapter
15. Use weighting only where there is a defensible basis
Weights can adjust the contribution of respondents to better match known population characteristics under an appropriate analytical approach. They cannot conjure opinions from people outside the sampling frame or guarantee that hidden selection differences have disappeared. Methods and sensitivity require qualified review.
Harbourline may compare its respondent profile with actual customer records when studying current loyalty members. If the survey has too few responses from a relevant group, its team assesses what can validly be adjusted and reports limitations. The company should not weight an online convenience sample until its headline agrees with marketing. AAPOR’s disclosure checklist requires transparency about weighting and processing.
Back to chapter index · Next chapter
16. Do not attach a conventional margin of error to a convenience sample
Sampling error estimates rely on assumptions about the design and how observations were generated. Reporting ‘±3%’ for a self-selected online sample as if it were a simple random draw can create a false impression of precision. Other uncertainty, including nonresponse and measurement error, can dominate.
Harbourline avoids using the familiar margin-of-error shorthand to decorate its loyalty-member survey. It explains the sample, actual response count and reasonable limitations. Where statistical modelling or alternative uncertainty intervals are appropriate, competent analysts must describe the method. The buyer deserves an explanation of what the data can and cannot settle, not a reassuring symbol copied from a political poll.
Back to chapter index · Next chapter
17. Use qualitative interviews for depth, not fake percentages
Interviews and focus groups can reveal why people choose products, which words confuse them and what needs further investigation. A qualitative interview study does not need to pretend it estimates national proportions. Its strength often lies in detailed examples and tested themes.
Harbourline conducts a small number of interviews with current café customers and reports recurring concerns about waiting time and menu clarity. It explains recruitment and participants’ context rather than announcing ‘seven out of ten Singaporeans dislike queues’. A competent business may combine qualitative discovery with subsequent quantitative work, with each stage’s purposes and limitations documented clearly.
Back to chapter index · Next chapter
18. Handle AI-generated summaries as draft interpretations
Software can transcribe interviews, group text responses and propose themes. It may omit context, hallucinate quotations or misclassify a person’s meaning. A research company needs human methodological oversight and clear rules for protecting identifiable participant data sent to service providers.
Harbourline checks AI-generated summaries against the original authorised data and avoids putting invented quotes in the client report. Its contract and privacy controls identify whether external processing is permitted. The 2025 ICC/ESOMAR Code emphasises accountability, transparency and human oversight in modern data analytics. A fluent conclusion is not evidence that the underlying interviews support it.
Back to chapter index · Next chapter
19. Make the report readable without erasing uncertainty
A useful report can show the objective, population, dates, recruitment, key questions, actual denominators, limitations and practical implications. Figures should be labelled so a client can distinguish confirmed observations from hypotheses. A polished chart does not excuse missing fieldwork information.
Harbourline creates a decision page about whether to pilot the new menu at selected outlets and what remains uncertain. It includes the questionnaire wording and a concise methods statement. The client may choose to launch a trial, but the firm’s report must not present future sales as a guaranteed outcome. Research should support thoughtful decisions, not replace managerial responsibility for risk.
Back to chapter index · Next chapter
20. Keep client confidentiality separate from research honesty
Clients may have commercially sensitive plans or internal sales data, while respondents may expect their answers to remain confidential. The research company needs clear ownership, controlled access, retention and a method for reporting findings without exposing protected information unnecessarily.
Harbourline gives the café group aggregate analysis rather than exporting named responses to every branch manager. Its client agreement addresses publication rights and necessary methodological disclosure, and does not grant permission to misrepresent how data were collected. A company can preserve legitimate trade secrets while still being honest about sample limitations in any public headline.
Back to chapter index · Next chapter
21. Track corrections when a published finding is challenged
Researchers can find coding errors, unsuitable eligibility rules or misinterpreted survey questions after delivery. The firm should preserve the original report, assess effects and issue traceable corrections where warranted. Silently replacing the numbers on a client portal can leave decisions or advertisements using the old finding.
Harbourline later discovers one question label was incorrectly mapped in a graph. It records the actual error, which results and recommendations are affected, and which clients or public audiences require a correction. A good research company’s reputation depends on correcting problems fairly, not claiming that careful analysts are incapable of error. The final deliverable is a defensible decision record.
Back to chapter index · Next chapter
22. Price research to protect participant care and quality
Research costs include design, recruitment, fieldwork, incentives, translation, quality controls, statistical analysis, privacy compliance and client support. A low fee that omits actual quality work can encourage shortcuts such as using an inappropriate opt-in panel while selling the results as nationally representative.
Harbourline quotes alternative projects suited to the café’s real needs: a narrower loyalty-member study versus a separately designed broader market study. It explains how recruitment choices change cost and inference, and rejects a contractual clause requiring a favourable result. Sustainable economics fund impartial analysis and honest limitations. Delivering 400 cheap answers is not automatically better than 200 carefully recruited ones for the right decision.
Back to chapter index · Next chapter
23. The gold standard is an insight that survives scrutiny
Harbourline’s fictional study produced a legitimate observation about responding loyalty members, but its marketing headline would have invented the preferences of a broader population. The company corrects the population label, preserves the survey method and helps the client plan an appropriately scoped small menu trial.
A gold-standard market research company protects participants, defines the intended population, measures honestly, distinguishes observations from assumptions, and makes useful recommendations without hiding uncertainty. The value is not that every number looks impressive. It is that another informed person can understand what happened and decide how much confidence the finding deserves.
Worked survey: the correct percentage with the wrong public claim
The figures are fictional and do not estimate national opinion. They illustrate a key issue in interpreting voluntary participation.
| Fictional study metric | Value | Defensible meaning |
|---|---|---|
| Invitations to existing loyalty members | 1,000 | Not invitations to a probability sample of all Singaporeans |
| Usable voluntary responses | 400 | 40% of invitations, not automatically standard response-rate calculation |
| Responses saying they would consider the new option | 248 | 248 ÷ 400 = 62% of responding loyalty members |
| Invited members who did not provide usable responses | 600 | Opinions not known from this survey |
| Estimated preference of all Singapore residents | Not established | Needs a separate suitable research design |
Fast answers to common market-research questions
Does a survey with 400 answers represent the country? Not automatically. It depends on who could be selected, how recruitment worked, nonresponse, weighting assumptions and what the questions actually measured.
Can a researcher use existing customer data without requesting consent for every analytic operation? The correct legal basis depends on the actual circumstances and PDPA provisions. An applicable research exception cannot be assumed for unrelated marketing uses.
Is an online poll always nonprobability? No. Survey mode and sampling method are different. Probability-recruited samples may be answered online, while volunteer samples generally lack known inclusion probabilities.
Does 62% support a slogan that Singapore prefers the product? Not if the 62% came only from responding loyalty members asked whether they would consider trying the item. Population and question meaning must not be silently changed.
Can AI draft the conclusions? It can assist analysis under suitable governance, but responsible humans must verify source evidence, privacy arrangements and whether the conclusions reflect actual data.
Client and research-company quality audit
- Does the proposal identify a specific decision and population rather than a desired sales headline?
- Are the sponsor, data processor and actual survey recruiter identified?
- Can the firm show the exact questions, response choices and pilot findings?
- Are research participants treated respectfully with appropriate lawful data handling?
- Are probability and opt-in sampling methods distinguished accurately?
- Do reports retain fieldwork dates, denominators, exclusions and limitations?
- Are weighting, margins of error and model-based inferences used only under valid assumptions?
- Can qualitative themes and AI summaries be checked against the original data?
- Is the company free to report inconvenient findings and issue transparent corrections?
Current ethical and Singapore reference guidance
- ICC/ESOMAR International Code (2025; updated October 2026) — ethical research and human oversight
- AAPOR: Best Practices for Survey Research — questionnaire, samples, response and analysis
- AAPOR: Disclosure Standards — research methods and limitations for public claims
- ESOMAR/GRBN: Research with Children — vulnerable respondent safeguards
- PDPC: Data Protection Guidance — personal data responsibilities
- PDPC: Analytics and Research Selected Topics — research and anonymisation context
- eduKate: Advertising Companies — truthful use of research claims
- eduKate: Laboratory Testing Companies — parallels in evidence integrity
This educational company-management article is not a claim about actual public opinion, a research ethics approval, a statistical estimation of Singapore’s population or legal advice on personal-data processing. Harbourline and its survey results are fictional. Valid inference depends on the true sampling, data and professional methodology.