A research framework should not begin with a diagram of arrows or a familiar list of variables. Its starting point should be the question the study genuinely needs to answer. When the question is unclear, a model is often assembled from concepts that merely seem related. When the question is clear, selecting variables, arguing for relationships and designing measurement become much more consistent.
In theses, dissertations and even journal articles, one common mistake is choosing the model too early. A researcher sees a model used in an earlier paper, changes the context, adds several variables and then tries to make the question fit. This may save time at the beginning, but it creates difficulties later: the literature review lacks focus, hypotheses are weak, measures do not fit the phenomenon and the discussion merely repeats statistical results. A useful study should move in the opposite direction: from a practical problem or knowledge gap, to a clear question, then to concepts, variables and the model.
The research question sets the boundary of the model
A research question is not a rewritten version of a topic title. It needs to specify the phenomenon to be explained, the relevant actors or context and the kind of understanding the study seeks to create. The question “which factors influence young customers’ intention to continue using a digital banking application in Vietnam?” differs from “how does user experience influence continuance intention, and does trust mediate that relationship?” The second question already signals the outcome variable, the relationship to be tested and the possible mechanism that requires explanation.
Writing a sufficiently specific question also limits the model. If the study concerns continuance intention, not every concept related to digital banking needs to be included. The researcher should select factors with theoretical grounding and direct explanatory value for that behavior. This matters more than making the diagram contain many arrows. A compact model that answers the right question is usually more credible than a large model without a clear logic of priority.
Applied example
A retailer wants to understand why customers who bought once do not return to its app. If the team simply asks, “which factors influence loyalty?”, it may collect too many variables. If it asks, “which parts of the post-purchase experience reduce the intention of new customers to return?”, the study can focus on delivery transparency, refund handling, customer support and perceived trust. The question has turned a broad issue into a specific direction for measurement and action.
Distinguish concepts, variables and indicators
These three terms are often used interchangeably, but they have different roles. A concept is the abstract idea that matters to the study, such as trust, service quality or innovation capability. A variable is the way the concept is placed in an observable or testable relationship: trust may be a mediator, service quality an independent variable and continuance intention a dependent variable. Indicators are the specific questions or signs used to measure that variable.
This distinction helps researchers avoid selecting a scale simply because the wording “looks similar”. Before using any scale, one should check how the original concept was defined, whether it captures perception, evaluation or behavior, and in what context it was measured previously. A scale for trust in an e-commerce platform does not automatically fit trust in digital financial advice, even when both are labelled “trust”.
Move from theoretical reasoning to relationships among variables
A good hypothesis is not the statement “X has a positive effect on Y” written after looking at many diagrams. It should answer at least three questions: why might X influence Y, what mechanism explains the influence, and what conditions in the present context make the relationship worth testing? Theory provides the logic for the answer; prior studies provide evidence that the logic has been observed, debated or remains unsettled.
For example, if an application’s information quality is expected to affect continuance intention, the researcher should not merely cite one study that found a positive coefficient. The argument needs to explain that clear and timely information reduces uncertainty during user decisions, which can improve perceived usefulness and willingness to return. If earlier findings differ, the researcher should consider whether the difference comes from the type of service, level of risk, sample characteristics or measurement approach. This process creates an argument rather than a collection of variables.
| Check step | Question to answer | Evidence of good practice |
|---|---|---|
| Clarify the problem | What phenomenon needs explanation, and why? | The problem is tied to a specific context and decision maker. |
| Write the question | What does the study seek to know, about whom and under what conditions? | The question sets a clear boundary rather than covering every factor. |
| Select concepts | Which concepts are genuinely needed to answer the question? | Each concept has a definition and a clear role in the model. |
| Argue relationships | Why should the variables be related? | Theory, mechanism and relevant evidence are present. |
| Design measurement | Which indicators accurately reflect the concept in this context? | Content validity is examined before the main survey. |
Mediation, moderation and controls should appear only when needed
A mediator answers the question “through what process does X affect Y?”. A moderator answers “when, or for whom, does the X–Y relationship change?”. A control variable is included to reduce the risk that another characteristic distorts the estimate. These variables are useful when they directly serve the question and argument, but they should not be added simply to make the model look more sophisticated.
For instance, digital literacy may moderate the effect of information quality on continuance intention if there is reason to believe that users with low digital capability find it harder to assess and use information well. It should not be included only because a paper in another field used it. Every additional element raises requirements for theory, sampling, measurement and interpretation.
Measurement is the final test of research logic
Many models look reasonable on paper but become problematic when translated into a questionnaire, interview data or operational data. This is the stage for checking content validity: does each indicator reflect the intended concept, do respondents understand it as expected, and can the data distinguish closely related concepts? In quantitative research, expert discussion, cognitive pretesting and a pilot survey often reveal issues before the main data collection.
Researchers also need consistency in the level of analysis. If the question concerns firms but the data consist of individual employee perceptions, the study must explain why those perceptions represent an organizational phenomenon, or narrow the scope of the conclusion. Likewise, a variable about actual behavior should not be inferred simply from intention when behavioral data are unavailable.
A workflow that can be used immediately
Before drawing the model, write one short page with four parts: the problem to explain; the research question; three to five concepts that are truly essential; and a one- or two-sentence explanation for every anticipated relationship. Then compare this page with the literature you have read. If a variable does not help answer the question, consider removing it. If a relationship cannot yet be explained by theory or evidence, continue reading rather than rushing to write a hypothesis.
This process does not slow research down. It saves time in more costly stages such as questionnaire design, sampling, data cleaning and post hoc model revision. More importantly, it creates a study that can be defended through logic: readers can see the decisions that connect the original problem to every variable.
Conclusion
A coherent research framework is the result of deliberate choices, not a collection of currently popular variables. Begin with the question that needs an answer, distinguish clearly among concepts, variables and indicators, argue for every relationship, and test the logic through measurement. When this logic holds, the model does not only make data analysis easier; it also produces clearer knowledge contributions and practical implications.
References
Maxwell, J. A. (2013). Qualitative research design: An interactive approach (3rd ed.). SAGE.
Podsakoff, P. M., MacKenzie, S. B., & Podsakoff, N. P. (2016). Recommendations for creating better concept definitions in the organizational, behavioral, and social sciences. Organizational Research Methods, 19(2), 159–203. https://doi.org/10.1177/1094428115624965
Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and quasi-experimental designs for generalized causal inference. Houghton Mifflin.
Venkatesh, V., Brown, S. A., & Bala, H. (2013). Bridging the qualitative-quantitative divide: Guidelines for conducting mixed methods research in information systems. MIS Quarterly, 37(1), 21–54. https://doi.org/10.25300/MISQ/2013/37.1.02
