Many studies begin with a long list of variables: service quality, trust, perceived value, satisfaction, repurchase intention and word of mouth. The list may be theoretically defensible, yet it can still take the research in the wrong direction. The important question is not how many variables a model contains. It is whether each one helps answer a specific decision.
A sound research model does not start by asking, “Which variables are popular?” It starts by asking, “Once the findings are available, who should make a different decision, and what decision is it?” Once that decision is clear, variable selection, measurement and analysis become far more focused.
Start with the decision the study must support
Write the brief in a practical form: who must decide what, in which context, and what do they need to know to decide better? Imagine a coffee chain deciding whether to reduce perceived waiting time or improve how staff recover from service failures in order to bring customers back. This is no longer a broad topic about “customer experience”; it is a resource-prioritisation decision.
For that decision, the outcome may be return intention. Relevant explanatory variables could be perceived waiting time and service-recovery quality. Adding ten more variables merely because they appeared in earlier papers makes the model heavier, but it does not necessarily make the manager’s choice clearer.
Use three questions to filter each variable
For every proposed variable, ask three questions. First, if this variable changed, could the user of the findings make a different decision? Second, is there a credible theoretical mechanism or evidence explaining why it relates to the outcome? Third, can it be measured for the right people, at the right time, with acceptable quality?
Passing the second question alone is not enough. For example, technology affinity may relate to intention to use a food-delivery app. But if the business must decide whether to redesign delivery operations or change its replacement policy, that variable does not identify an operational lever. Depending on the research question, it may be a control variable or may not belong in the model at all.
Separate outcomes, explanations and conditions
A common mistake is to call every model element an “influencing factor.” Before measurement begins, define the role of each variable. The outcome is what the study seeks to improve or predict. Explanatory variables are levers or mechanisms that may relate to that outcome. Conditions indicate when, for whom or in which situation a relationship differs.
In the coffee-chain example, return intention is the outcome. Perceived waiting time and service-recovery quality are explanations. Whether a customer buys takeaway or stays in the café may be a useful condition if there is a practical reason to expect the groups to experience waiting differently. This role discipline keeps the model from becoming a diagram with many arrows but no decision story.
Do not measure what cannot be acted on
A highly reliable scale may still have little managerial value if the organisation cannot act on what it measures. This does not mean studies should include only easy-to-change variables. Background characteristics such as income, experience or familiarity can be necessary for understanding context. They should, however, play the right role: description, segmentation or control rather than a proposed solution.
A simple test is to imagine the final report. If variable X is strongly related to outcome Y, what could a manager do next Monday? If the answer is only “understand customers better,” X has not yet delivered enough decision value. If the answer is “redesign peak-hour order-prioritisation rules” or “train a specific service-recovery script,” the study is much closer to action.
Lock the variable map before writing the questionnaire
Before drafting questions, create a short working table for the research team; it need not appear in the thesis. For each line, record the decision to support, the outcome, explanatory variables, theoretical rationale, data source, measurement time and the action that could follow from the result. The table reveals variables chosen simply because they sound familiar, scales that do not fit the context, and data being requested from the wrong respondent.
If the study asks whether late delivery reduces return intention, asking delivery staff about customers’ intentions is a weak choice. The data should come from customers or from a suitable behavioural record. Similarly, when actual delivery time is already captured by the system, a recall question should not wholly replace that record if the aim is to assess operational performance.
Choose less so you can explain more clearly
Reducing the number of variables does not make research poorer. It protects the research logic. A focused model lets readers follow the path from problem to evidence to decision. It also reduces respondent burden, lowers the risk of conceptual overlap and keeps the discussion from becoming a list of coefficients.
Before finalising the model, reread the decision question at the beginning. Every variable should be present because it helps answer that question, not because it appears in a famous framework. With that discipline, the questionnaire stops being the starting point; it becomes a tool deliberately designed to collect the evidence needed for a clear choice.
