03 · What You Need to Know
A Research Question and a Prediction Are Related, but They Are Not the Same Thing
Research questions, hypotheses, and objectives are closely connected. Methodological guidance commonly describes a research question as identifying the uncertainty or problem to be investigated, while a hypothesis states a testable prediction about the expected relationship or outcome when such a prediction is appropriate.
Not every study requires a formal hypothesis. Exploratory, descriptive, qualitative, methodological, and other forms of research may be guided primarily by questions or objectives. When a hypothesis is appropriate, however, preserving the distinction between question and prediction can clarify what was uncertain before the study began.
A Question Identifies the Uncertainty
Suppose researchers do not know whether AI-generated formative feedback produces better writing outcomes than instructor feedback in a particular population and context.
The research question identifies that uncertainty:
“How does research-writing performance compare between students receiving AI-generated formative feedback and students receiving instructor feedback?”
There are several possible answers. One condition may perform better, the other may perform better, or any difference may be negligible or uncertain.
A Hypothesis States the Expected Answer
If theory and previous evidence provide a basis for expecting AI-generated feedback to produce higher performance, the researcher may state that prediction explicitly as a hypothesis.
The prediction is valuable precisely because it can be compared with what the evidence eventually shows. If the predicted answer is already embedded as fact inside the research question, the logical separation becomes blurred.
Research question
What relationship, difference, effect, process, experience, or phenomenon does the study seek to establish or understand?
Hypothesis or prediction
What result does theory, prior evidence, or substantive reasoning lead the researcher to expect?
Look for Verbs That Smuggle the Prediction Into the Question
Words such as “improve,” “increase,” “reduce,” “enhance,” “weaken,” “cause,” and “lead to” can embed a directional proposition.
Consider:
- How does gamification increase student engagement?
- Why does remote work improve employee productivity?
- How does social media reduce academic performance?
- Why does generative AI weaken critical-thinking skills?
Each formulation presupposes the directional relationship named in the verb. The question then asks about its magnitude, explanation, or mechanism.
If the directional relationship has not been established sufficiently for the relevant context, the question may be starting with an empirical assumption that should still be investigated.
A Directional Question Is Not Automatically a Disguised Prediction
Suppose a large and credible body of prior evidence establishes that a treatment reduces a particular symptom, and researchers now ask which mechanism accounts for that reduction. In such a case, building on the established effect may be entirely reasonable.
Similarly, a question may explicitly ask whether X increases Y rather than whether X affects Y. That directional formulation can be appropriate when the scientific target itself concerns an increase.
The diagnostic issue is whether the direction is being investigated or presumed. If “Does X increase Y?” permits “no” as a legitimate answer, it remains a genuine empirical question. “Why does X increase Y?” usually presupposes that the increase exists.
“Does X Improve Y?” and “How Does X Improve Y?” Are Not Equivalent
The grammatical difference is small, but the logical difference can be substantial.
| Formulation |
What remains uncertain? |
What is presupposed? |
| Does X improve Y? |
Whether improvement occurs |
Not necessarily that improvement occurs |
| How much does X improve Y? |
The magnitude of improvement |
That improvement occurs |
| Why does X improve Y? |
The explanation or mechanism |
That improvement occurs |
| How is X related to Y? |
The form, direction, or magnitude of association |
Usually less about direction, although the exact wording and design still matter |
The appropriate version depends on what previous evidence has already established and what the current study is intended to discover.
Do Not Replace Every Prediction With Vague Nondirectional Wording
The solution is not to make every research question maximally neutral. “Is there any relationship between anything and anything?” is open, but not useful.
A strong question can specify the population, constructs, comparison, timeframe, and type of relationship precisely while remaining open about the empirical feature that the study is supposed to determine.
The objective is not neutrality for its own sake. It is logical honesty about what is known, what is predicted, and what remains uncertain.
Prediction Becomes Especially Problematic When It Also Implies Causation
“How does frequent AI use reduce critical thinking?” contains at least two commitments: that a reduction exists and that AI use produces it.
The study may therefore begin with both a directional prediction and a causal conclusion embedded in the wording. If the proposed evidence is merely cross-sectional association, the mismatch becomes even larger.
Questions using “effect,” “impact,” “influence,” or other causal language should therefore be checked separately for a hidden causal assumption.
A Question Can Also Disguise a Prediction Through Its Comparison
Suppose a question asks, “Why are high-performing students more effective users of generative AI than low-performing students?”
The prediction is not located in a causal verb. It appears in the comparative structure: high-performing students are assumed to be more effective AI users.
Removing the presupposition might yield a question about whether and how patterns of AI use differ by prior academic performance. The exact reformulation depends on the study, but the empirical difference should not be declared before it is demonstrated.
Predictions Can Be Strong Without Being Hidden
A researcher may have an exceptionally strong theoretical expectation and state it clearly in the hypothesis. That is preferable to weakening the theory merely to sound neutral.
The advantage of explicit prediction is accountability. Readers can see what was expected before the result and compare that expectation with the evidence. This distinction becomes particularly useful in confirmatory research.
Preregistration Can Preserve the Chronology of Prediction
Preregistration allows researchers to record hypotheses, methods, and analytical plans before outcomes are known. One purpose is to distinguish predictions specified in advance from explanations or analyses developed after seeing the data.
This does not make exploratory research inferior. Exploration can reveal unexpected patterns and generate new hypotheses. The value lies in knowing which is which.
Preregistration is also not an infallible safeguard. Research on preregistration practice suggests that preregistrations vary in specificity, and empirical evaluations do not support treating preregistration as a guarantee against every form of HARKing or analytical flexibility. Its contribution is primarily to prospective specification and transparency when implemented well.
Post Hoc Hypotheses Are Not the Problem; Disguising Them Is
Kerr introduced the term HARKing for presenting a hypothesis developed after results were known as though it had been an a priori hypothesis.
Suppose an unexpected subgroup difference appears during analysis and suggests a plausible theoretical explanation. Developing a new hypothesis from that finding is legitimate scientific reasoning. The problem occurs when the manuscript is rewritten so that the introduction claims the researcher predicted the subgroup effect from the beginning.
A transparent account might instead say that the unexpected pattern motivated an exploratory analysis or a hypothesis for future testing.
Ask Whether the Question Could Honestly Receive the Opposite Answer
This is perhaps the simplest diagnostic test.
Take the result implied by the wording and reverse it. If the question becomes nonsensical when the opposite result occurs, ask whether the question was truly open.
“Does AI feedback improve writing?” can honestly be answered “no.”
“Why does AI feedback improve writing?” becomes difficult to answer if there is credible evidence that it does not improve writing. The question has assumed the phenomenon it asks you to explain.
This test connects directly to whether the research question can produce a meaningful answer regardless of result direction.
