Many survey projects begin by collecting questions: opening old questionnaires, translating a scale, and adding variables the team finds interesting. The approach feels fast because the questionnaire grows longer each day. Yet it can lead to a contradiction: a large amount of data is collected, but when results appear, the team still does not know which decision it needs to make, which variable actually answers the research question, and which answer is merely peripheral information. A good questionnaire does not begin with wording. It begins with a map connecting a decision, the evidence required, and the person or unit able to provide that evidence.
This map does not replace a literature review or scale-validation process. It has a simpler but important role: requiring the team to agree on what the study will help it understand, explain, or choose. Once the destination is clearer, selecting concepts, identifying respondents, designing indicators, and ordering questions become less arbitrary. The team can also see early what a short survey cannot measure, which information should come from secondary data, and which apparently attractive questions have no role in the model.
Start with a decision, not a list of variables
Before writing questions, write one sentence describing the decision or conclusion that the study should support. In academic research, this may be testing whether a theoretical mechanism explains behaviour in a specific context. In applied research, it may be prioritising which touchpoint to improve or identifying what prevents customers from continuing to use a service. The statement must be narrow enough to distinguish what the study truly needs to know from everything it could possibly collect.
The team can then list the minimum evidence required to answer it. If the aim is to understand why customers leave, a general satisfaction question is not enough. Evidence may be needed about experiences at particular moments, difficulty resolving a problem, intention to continue using the service, and context such as product type or length of relationship. If the aim is to test a hypothesis about trust, the team needs to distinguish trust from satisfaction, perceived risk, or behavioural intention rather than allowing these concepts to blend into a set of similar questions.
| Map component | Question to answer | Required output |
|---|---|---|
| Decision or conclusion | What will the study help choose, explain, or assess? | One statement defining scope and intended user of the result. |
| Evidence | What must be observed to answer that question? | Necessary concepts, control variables, and contextual data. |
| Response source | Who or which system knows this information most reliably? | Unit of analysis, sampling criteria, and data source. |
| Use of results | How will each group of indicators be analysed? | An analysis plan and preliminary interpretation criteria. |
Distinguish concepts to measure from information that is merely interesting
A common error is putting every question that interests the team into the questionnaire. Information about age, acquisition channel, purchase frequency, views of every feature, and many open comments may all be useful in discussion. However, every question increases response burden, the risk of non-completion, and data-cleaning cost. More importantly, a variable without a clear theoretical, analytical, or sampling role often produces many descriptive tables without moving the research story forward.
Attach a purpose label to every question: measure a concept in the model; classify respondents for comparison; control an alternative explanation; screen a respondent; or collect context for interpretation. If no label fits, set the question aside until a specific use is identified. This does not impoverish the study. It frees time and attention for indicators more likely to yield credible evidence.
Example: research on the post-purchase experience
A team wants to identify why customers do not return after purchasing electronics. Rather than immediately asking a long list about every store attribute, it maps the work: the conclusion needed is which post-purchase journey point lowers repurchase intention; the evidence required concerns clarity of guidance, ability to contact support, incident-resolution experience, and repurchase intention; the response source is customers who purchased in the previous three months; data on product type and purchase method are retained to check contextual differences. Questions about packaging colour or advertising appeal are removed because they do not yet serve the required conclusion.
Check the fit between the respondent and what must be measured
A well-worded concept still produces weak data when it is asked of the wrong person. Frontline employees may describe the process they perform but not necessarily know an organisation’s strategic decisions. New customers may evaluate registration but not yet have enough experience to assess loyalty. The measurement-decision map should therefore state who has direct experience of each topic, when memory is still sufficiently accurate, and which conditions make a response unsuitable.
This is also the time to revisit the unit of analysis. If the research question concerns firms but data are collected from individuals, there must be a clear reason to aggregate or the conclusion must be rephrased. If the model concerns the customer–brand relationship, questions should not ask respondents to guess at the intentions of internal employees. This fit should be checked before the questionnaire is released, not after the model produces difficult-to-explain findings.
Write questions after the analysis path is visible
Once the map is clear, moving to a questionnaire becomes more disciplined. For each concept, identify the working definition, a scale source if it is used or adapted, response format, reference period, and treatment of not-applicable cases. Then imagine the data file after collection: which variables will test hypotheses, which will describe the sample, which conditions will remove an observation, and which results could require the team to revisit the design.
Working backwards from the output helps uncover small but costly errors: inconsistent response scales, questions without a suitable answer option, screening criteria placed too late, or a missing variable needed to interpret group differences. It also prevents wording changes made only because a sentence sounds smoother. Readability matters, but a question must preserve the meaning of the concept and produce data that can be used for its stated purpose.
Finally, a measurement-decision map is a living document. After cognitive interviews or a pilot, the team may find that respondents understand a point differently, a variable has little variation, or another data source is needed. Revise both the map and the questionnaire so their logic remains connected. Done this way, a survey becomes more than a tool for collecting answers: it becomes an evidence path that can be explained from the research question to the conclusion.
