03 · What You Need to Know
When Several Hypotheses Can Belong to One Research Question
A Research Question Can Be Broader Than a Single Prediction
A research question identifies what the study seeks to determine. A hypothesis states an expected answer or empirical pattern. Because a question can contain several components, it does not always translate into exactly one prediction.
Consider:
How does structured generative AI tutoring affect undergraduate students' learning outcomes?
If "learning outcomes" includes conceptual understanding and problem-solving performance, the researcher might formulate two hypotheses:
H1: Students receiving structured generative AI tutoring will achieve higher conceptual-understanding scores than students receiving the comparison condition.
H2: Students receiving structured generative AI tutoring will achieve higher problem-solving scores than students receiving the comparison condition.
Both hypotheses address the same broad research question, but each makes a distinguishable prediction about a different outcome.
One-to-One Correspondence Is Not a Universal Rule
Research questions, objectives, hypotheses, outcomes, and statistical tests should align, but alignment does not require a rigid numerical correspondence.
A primary research question may lead to one principal hypothesis. It may also imply several related hypotheses. Conversely, some research questions, particularly descriptive or exploratory ones, may require no formal hypothesis at all.
The more useful principle is conceptual alignment: every hypothesis should help answer an identified research question, and every central research question should be addressed by the design and evidence collected. Methodological guidance similarly emphasizes that research questions and hypotheses should be established during study planning and aligned with study objectives.
When Multiple Hypotheses Make Sense
Several hypotheses may be appropriate when one research question contains multiple distinct predictions. This can occur when the study examines several outcomes, several theoretically important predictors, different mechanisms, interactions, or prespecified subgroup expectations.
| Research structure |
Why multiple hypotheses may be useful |
Example |
| Several outcomes |
Each outcome represents a distinguishable prediction |
An intervention is predicted to affect achievement and self-efficacy |
| Several predictors |
The theory predicts different relationships with the same outcome |
Self-efficacy and instructor support are each predicted to relate to engagement |
| Different mechanisms |
Separate hypotheses represent different parts of a theoretical explanation |
An intervention is predicted to influence engagement directly and indirectly through self-efficacy |
| Moderation |
The expected effect depends on another variable |
The intervention effect is predicted to be stronger among novice learners |
| Several prespecified comparisons |
Different comparisons answer distinct parts of the same question |
Two intervention conditions are each compared with a control condition |
Separate Hypotheses Can Be Clearer Than One Complex Hypothesis
Suppose a researcher writes:
Students receiving AI-supported tutoring will demonstrate higher achievement, greater self-efficacy, stronger engagement, and lower cognitive load than students receiving conventional tutoring.
This is a legitimate complex prediction, but four empirically distinguishable claims are embedded within it.
What happens if achievement and self-efficacy improve but engagement and cognitive load do not? Saying simply that "the hypothesis was partially supported" provides less information than reporting the evidence for each prediction separately.
Dividing the statement into several hypotheses can make the study easier to interpret. This illustrates why a complex hypothesis may sometimes be better represented as several simpler hypotheses.
Multiple Hypotheses Should Still Form a Coherent Family
The fact that one research question can support several hypotheses does not mean every possible relationship among the measured variables deserves one.
If your question concerns whether an intervention improves learning, hypotheses about achievement, retention, and transfer may form a coherent family when each follows from the theoretical rationale. Adding hypotheses about every demographic variable simply because those data were collected would be harder to justify.
Each hypothesis should have a substantive reason to exist before statistical significance enters the conversation.
Primary and Secondary Hypotheses Can Help Establish Priority
When several hypotheses are included, it can be useful to distinguish the prediction that addresses the central study aim from additional predictions.
A primary hypothesis usually corresponds to the most important research question, objective, or outcome. Secondary hypotheses address additional prespecified questions that remain scientifically relevant but are not the principal basis of the study.
This distinction can influence study design, sample-size planning, statistical analysis, and interpretation. In experimental and analytical research, primary outcomes should align directly with the primary aim, while secondary outcomes should have corresponding secondary aims and a clear justification.
Primary hypothesis
The principal prediction associated with the study's central research objective or primary outcome.
Secondary hypothesis
An additional prespecified prediction that addresses a secondary objective, outcome, mechanism, or related question.
Multiple Hypotheses Create a Multiplicity Problem
There is an important statistical consequence to multiplying hypotheses.
If researchers perform many statistical tests and treat each one independently using the same conventional significance threshold, the probability of obtaining at least one false-positive result across the collection can increase. The issue is known as multiplicity or multiple testing.
This does not mean that every study with several hypotheses requires the same statistical correction. The appropriate strategy depends on how the hypotheses are organized, which claims are primary, the inferential framework, and whether conclusions depend on one or several tests.
Still, multiplicity should be considered during study planning rather than discovered after a collection of p-values has appeared. Guidance from the U.S. Food and Drug Administration, for example, emphasizes that multiple endpoints can increase the risk of false conclusions unless multiplicity is handled appropriately.
Watch Out
More hypotheses mean more opportunities to find an apparently interesting result by chance. If your study involves many formal tests, decide in advance which hypotheses are primary and how multiplicity will be addressed rather than treating every p-value as an independent verdict.
More Hypotheses Can Affect Statistical Power and Sample-Size Planning
A study designed around one primary outcome may not be adequately powered to provide precise evidence for every secondary outcome or subgroup analysis. When multiplicity adjustments are required, the threshold for individual claims may also become more demanding.
Adding hypotheses therefore has consequences beyond manuscript length. Each important prediction may create requirements for measurement quality, sample size, analytical planning, and interpretation.
This is one reason researchers should establish priorities before data collection rather than treating all possible hypotheses as equally important.
One Broad Question Should Not Become an Excuse for Unlimited Hypotheses
Suppose your dataset contains 15 variables. There are many possible pairwise relationships among them. You could write dozens of hypotheses and claim that they all belong beneath the broad question "What factors are associated with student success?"
Technically, perhaps. Methodologically, the study would become difficult to justify and interpret.
The number of possible analyses is not the same as the number of scientifically motivated hypotheses. Predictions should emerge from the research problem, theory, prior evidence, or another defensible rationale rather than from the mere availability of variables.
Once the list begins expanding substantially, it is worth asking whether the study has accumulated too many hypotheses.
Do Multiple Hypotheses Need Separate Numbers?
Numbering hypotheses as H1, H2, H3, and so forth is a reporting convention rather than a universal methodological requirement. It can nevertheless be useful when a study contains several predictions because it allows the methods, results, tables, and discussion to refer to each hypothesis consistently.
More complex studies may use hierarchical labels such as H1a and H1b when several predictions belong to one conceptual family. Use such structures only when they clarify the conceptual organization. A taxonomy impressive enough to require a legend is usually trying to tell you something about the study's complexity.
Each Hypothesis Should Be Testable on Its Own Terms
If several hypotheses correspond to one research question, each should still identify a sufficiently clear prediction and be evaluable using the study's design and data.
A weak hypothesis does not become stronger merely because it belongs to a coherent group. Each prediction should meet the requirements for being empirically testable.
Hypotheses Should Be Specified Before Their Results Are Known
Multiple hypotheses are particularly vulnerable to retrospective reconstruction. Researchers may test numerous relationships, retain those producing interesting results, and then present only those relationships as though they had been predicted.
Exploratory analysis is legitimate. The problem is concealing its exploratory origin.
If additional hypotheses emerge after examining the data, they can be reported as newly generated hypotheses and investigated further. They should not silently join the list of supposedly prespecified predictions.
