03 · What You Need to Know
Build a Traceable Chain From Question to Evidence
Begin with the distinction between questions, objectives, and hypotheses
Research questions, objectives, and hypotheses are related, but they do not perform identical functions.
| Element |
Primary function |
Example |
| Research question |
States what the study seeks to find out |
Is frequency of generative AI use associated with academic writing self-efficacy among undergraduate students? |
| Objective |
States what the study will do to address the research problem |
To examine the association between frequency of generative AI use and academic writing self-efficacy among undergraduate students. |
| Hypothesis |
States a testable expectation about a relationship, difference, effect, or other parameter when hypothesis testing is appropriate |
Higher frequency of generative AI use is associated with academic writing self-efficacy. |
Not every study requires all three. Exploratory qualitative research, for example, may be organized around research questions and objectives without formal hypotheses. The protocol should use the conventions appropriate to the methodology rather than manufacturing a hypothesis simply because a template contains a heading for one.
WHO's recommended research protocol format similarly treats goals, specific objectives, and research questions as foundational elements and advises that objectives be specific and stated in advance.
Check that the wording of the question matches the design
Words such as describe, compare, associate, predict, explain, experience, and cause imply different evidentiary demands.
If a cross-sectional survey measures students' AI use and academic performance at one point in time, it may support a descriptive or associational question. By itself, it generally does not establish that AI use caused subsequent changes in academic performance.
If the question asks how participants experience a phenomenon, an interview-based qualitative design may be appropriate. If it asks whether an intervention causes a difference in an outcome, the design needs a defensible strategy for causal inference.
The study design should therefore follow the inferential ambition of the question, not merely the researcher's familiarity with a particular method.
Make sure the study population matches the population named in the question
A research question may concern "university students," while the actual sample consists entirely of first-year students from one program at one institution. That sample may still support a worthwhile study, but the question and claims should reflect the population the design can reasonably address.
Alignment requires consistency among the target population, sampling frame, eligibility criteria, recruitment strategy, and population to which the researcher intends to generalize or transfer the findings.
Sampling deserves separate scrutiny because even a well-aligned research question and measure can fail if the required participants or units never enter the study. The next step is therefore to ensure that sampling, measurement, data collection, and analysis fit together.
Translate abstract concepts into evidence you can actually collect
Many alignment problems arise between the research question and measurement. A question might refer to "learning," "engagement," "AI literacy," "well-being," or "research productivity," while the protocol collects a convenient variable that captures only one narrow aspect of that concept.
Ask what observable evidence would justify answering the question. Then determine whether the proposed instrument, observation, interview, record, test, device, or other data source actually produces that evidence.
WHO's protocol guidance places measurements, observations, instruments, and procedures within the methodology section because the credibility of the study depends substantially on how the design and methodology generate evidence relevant to the objectives.
Check whether the timing of measurement matches the question
Alignment has a temporal dimension. If the objective concerns change, the study generally needs evidence that can represent change. If it concerns persistence of an effect, the assessment schedule needs to extend far enough to examine persistence. If the hypothesis specifies an outcome after an intervention, measuring only before the intervention will not answer it.
This sounds obvious when stated plainly, yet temporal mismatches can hide behind general phrases such as "student performance will be assessed." A protocol should identify when measurements occur and why those time points are relevant to the question.
Each primary question should have an identifiable analytical route
Take each primary research question and ask: once the data exist, exactly how will this question be answered?
For a quantitative question about an association, identify the variables and analytical model or procedure that will estimate that association. For a group comparison, identify the groups, outcome, relevant time point, and analytical comparison. For a qualitative question, explain how the data-generation and analytical approach will produce an interpretation responsive to that question.
The objective is not to decorate every research question with a statistical test. It is to ensure that no primary question reaches the end of the protocol without an evidentiary and analytical pathway.
Hypotheses should correspond to variables the study actually defines
A hypothesis is difficult to test when its terms do not map clearly onto the protocol.
Suppose the hypothesis states that "responsible generative AI use improves academic achievement." The protocol would need to establish what counts as responsible AI use, how academic achievement is measured, what comparison or variation makes the hypothesis testable, and whether the design can support the word improves.
If the study merely measures current AI use and current grades, a hypothesis framed as an association may be more consistent with the evidence than one framed as an effect.
Primary and secondary elements should remain consistent across the protocol
A study may identify one outcome as primary in the objectives, another in the sample-size calculation, and a third in the analysis section. Such inconsistencies are more than editorial errors because different parts of the study may have been designed around different assumptions.
For randomized trials, SPIRIT 2025 specifically calls for primary and secondary objectives to be stated and for outcomes to be clearly defined. The guidance also links protocol transparency to the ability to identify undisclosed changes to primary outcomes or analyses after the study.
Whatever terminology your methodology uses, the hierarchy of the study should remain stable across the document. If an objective is primary, the sampling, measurement, and analysis plans should treat it accordingly.
The sample-size rationale should correspond to the primary analytical purpose
In studies requiring a formal sample-size calculation, the assumptions used in that calculation should correspond to the study's primary objective and planned analysis. A calculation based on detecting a correlation does not automatically justify a study whose primary analysis is a complex multivariable group comparison.
SPIRIT 2025 requires randomized trial protocols to explain how the sample size was determined, including assumptions supporting the calculation. The broader alignment principle applies wherever formal sample-size justification is used: calculate for the study you actually plan to conduct.
Do not let the available dataset silently redefine the research question
Secondary data can create a particular alignment problem. Researchers may begin with an important question, discover that the dataset does not measure the key construct adequately, and then substitute a convenient available variable while retaining the original wording of the question.
The defensible alternatives are to modify the question so that it matches what the data can support, obtain a more appropriate data source, or explicitly acknowledge the limitation of the available measure. Renaming a proxy does not turn it into the intended construct.
Use a question-to-method alignment matrix
One of the simplest ways to audit a protocol is to place the major components side by side. This does not need to appear in the final protocol unless useful, but it can expose gaps quickly during planning.
| Question or objective |
Required evidence |
Source or measure |
Design or procedure |
Analysis |
| Describe students' frequency of generative AI use |
Reported frequency of use |
Defined survey items |
Student survey |
Descriptive estimates |
| Examine association between AI use and writing self-efficacy |
AI-use measure and self-efficacy score for eligible participants |
AI-use items and specified self-efficacy instrument |
Same participants provide both measures |
Prespecified associational analysis |
| Explore how students explain their use of AI during writing |
Accounts of decisions, experiences, and reasoning |
Semistructured interview data |
Interview procedure appropriate to the qualitative approach |
Qualitative analytical method aligned with the research question |
If you cannot complete a row without writing "to be decided," you have found a point that needs either further planning or an explicit justification for remaining flexible.
Alignment should be checked in both directions
Most researchers check forward: question, then method, then analysis. It is equally useful to work backward.
Look at every major measure and ask which objective requires it. Look at every planned analysis and ask which research question it answers. Look at every subgroup, covariate, interview domain, or procedure and ask why it exists.
If an element cannot be traced to the research purpose, it may be unnecessary. If a research question has no corresponding evidence or analysis, something is missing.
Watch Out
Do not repair an alignment problem merely by changing wording. If the design cannot support a causal question, replacing "association" with "effect" in the objectives makes the mismatch worse, not better. Alignment requires changing the question, the design, or both until the evidentiary claim is defensible.
