03 · What You Need to Know
The Right Number of Hypotheses Depends on What the Study Is Built to Answer
There Is No Universal Maximum
Research methodology does not provide a general rule such as "a study should have no more than five hypotheses." Different research designs legitimately require different structures.
A tightly focused experiment may have one primary hypothesis. A factorial experiment may contain several theoretically important predictions, including main effects and interactions. A longitudinal study may evaluate several prespecified outcomes or time-dependent relationships. A complex theoretical model may imply numerous distinct paths.
Counting hypotheses without considering the design therefore tells you very little.
The Primary Research Question Should Still Be Visible
One of the clearest signs that a study has accumulated too many hypotheses is that its central question becomes difficult to identify.
Methodological guidance recommends establishing a focused primary research question because increasing the number of research questions can increase study-design and statistical complexity and can threaten the feasibility of answering all of them adequately.
The same principle applies to hypotheses. If a reader reaches H14 and can no longer tell what the study is fundamentally trying to establish, the problem is probably conceptual before it is numerical.
Think in Terms of Primary, Secondary, and Exploratory Claims
Not every interesting prediction needs equal status.
A useful hierarchy distinguishes the hypothesis most closely tied to the central study objective from additional prespecified hypotheses and from analyses intended primarily for exploration.
| Level |
Purpose |
Typical treatment |
| Primary hypothesis |
Addresses the central research objective |
Receives the strongest design, sample-size, and inferential priority |
| Secondary hypothesis |
Addresses additional prespecified questions |
Interpreted in relation to the study's secondary aims and multiplicity structure |
| Exploratory hypothesis or analysis |
Investigates additional patterns or generates future predictions |
Reported transparently with appropriately cautious interpretation |
This hierarchy prevents a common problem: treating fifteen hypotheses as though all fifteen were equally central to why the study exists.
Every Additional Hypothesis Has a Cost
Adding a hypothesis is not merely adding another sentence to the introduction. A serious hypothesis creates obligations.
You need a rationale for the prediction. You need valid measurements for the variables involved. The design must allow the relationship or effect to be evaluated. The analysis must address the hypothesis appropriately. The sample must provide enough information for the intended inference. The result must then be interpreted in relation to the other tests being performed.
If you cannot meet those obligations, the problem is not solved by numbering the prediction H9.
Too Many Hypotheses Can Create a Multiple-Testing Problem
Suppose a researcher performs one hypothesis test using α =.05. Under the relevant null model and assumptions, that test has a 5% Type I error rate.
Now suppose the researcher conducts many independent tests, each at.05, and treats any statistically significant result as a successful finding. The chance of obtaining at least one false-positive result across the family increases as the number of tests increases.
This calculation is an illustration, not a universal formula for every multiple-testing situation. Real tests may be correlated, hypotheses may be organized hierarchically, and different inferential procedures handle multiplicity in different ways.
The underlying lesson remains important: repeatedly applying the same nominal threshold across many opportunities for a positive finding can increase the chance of erroneous conclusions. NIH methodological resources and FDA guidance therefore emphasize advance planning for multiplicity when studies contain multiple outcomes or hypothesis tests.
Multiplicity Does Not Mean “Use Bonferroni for Everything”
Researchers sometimes learn about multiple testing and immediately divide.05 by the number of p-values in the manuscript. That is not a universal solution.
The appropriate multiplicity strategy depends on the family of hypotheses, their priority, dependence structure, and what claims the study intends to make. Options may include procedures such as Bonferroni, Holm, Hochberg, hierarchical or gatekeeping strategies, simultaneous modelling approaches, or other methods suited to the design.
In some inferential structures, adjustment may not be necessary in the same way. For example, if success requires all prespecified primary outcomes to meet their criteria rather than any one of them, the multiplicity problem differs from a design in which any successful endpoint can establish the study's main claim.
Multiplicity should therefore be planned with statistical reasoning, not treated as a formatting correction applied to a spreadsheet at the end.
Watch Out
Do not create a large hypothesis family first and ask how to repair the multiplicity problem later. Decide which claims are primary, secondary, and exploratory while designing the study, because those priorities can affect sample size, analysis, and interpretation.
Too Many Primary Hypotheses Are Particularly Difficult
Several secondary questions may be manageable, but declaring many outcomes or hypotheses equally primary creates additional problems.
Primary outcomes ordinarily represent the results most central to determining whether a study met its main objective. CONSORT guidance notes that most randomized trials have a single primary outcome and that multiple primary outcomes can create interpretive problems associated with multiplicity.
Imagine a study with five supposedly primary outcomes. What happens if three favor the intervention and two do not? Is the study successful? Does one significant outcome suffice? Must all five succeed? Those rules need to be defined in advance.
Without such prioritization, "primary" begins to mean little more than "something we measured."
Sample Size May Be Adequate for H1 but Not H8
Studies are commonly designed around a primary outcome or effect. Secondary hypotheses may involve less frequent outcomes, smaller subgroups, interactions, or noisier measures for which the study contains considerably less information.
A sample large enough to detect a plausible primary effect may be poorly suited to a moderator analysis or subgroup comparison. Nonsignificant secondary results can therefore be difficult to interpret when the study was never designed to estimate those effects precisely.
Before adding a hypothesis, ask not only whether you can run the analysis but whether the study can provide useful evidence about it.
Interactions and Subgroups Can Multiply Hypotheses Quickly
Suppose an intervention is evaluated across sex, year level, prior achievement, AI literacy, and learning modality. Testing whether the intervention effect differs across every one of those characteristics creates several additional hypotheses.
If each subgroup is then analyzed separately, the number of comparisons expands further.
Such analyses can be scientifically important when they follow from a clear rationale. They can also become unstable when subgroup sizes are small and the analyses were not planned. A large menu of subgroup hypotheses therefore deserves particular scrutiny.
Several Outcomes May Be Justified, but They Need Priorities
Complex phenomena rarely have only one meaningful outcome. An educational intervention, for example, might affect achievement, retention, engagement, self-efficacy, satisfaction, cognitive load, and persistence.
Measuring several of these outcomes can be sensible. Declaring every one of them the central hypothesis is another matter.
Identify which outcome most directly represents the principal research objective. Other outcomes can provide supporting, mechanistic, safety-related, or exploratory evidence. Methodological discussions of primary and secondary outcomes consistently emphasize that the outcomes should align with corresponding study aims and that excessive primary outcomes can create interpretive difficulties.
More Hypotheses Can Make a Study Less Coherent
Imagine a thesis containing hypotheses about achievement, motivation, self-efficacy, satisfaction, anxiety, technology acceptance, intention to continue, demographics, learning style, academic rank, and platform usage.
Perhaps there is a coherent theory connecting them. If there is, the conceptual model should make that structure visible.
If there is not, the study may actually contain several smaller studies sharing one questionnaire. The solution may be to narrow the research problem, separate primary from exploratory analyses, or reserve some questions for future work.
Do Not Split One Prediction Into Artificially Many Hypotheses
The opposite problem also occurs. Researchers sometimes produce a long hypothesis list by splitting a coherent prediction into tiny fragments.
For example, a theoretical model may predict one overall interaction, but the researcher writes separate hypotheses for every possible pairwise comparison simply to make the proposal look more detailed.
The number of hypotheses should reflect genuinely distinguishable scientific predictions, not the number of sentences you can derive from them.
When one question legitimately implies several predictions, the principles for using multiple hypotheses for the same research question can help determine whether splitting improves clarity.
Complex Hypotheses Can Sometimes Reduce a Long List, but Be Careful
You might combine several related predictions into a complex hypothesis rather than listing them separately. This can improve conceptual organization when the predictions form one coherent theoretical statement.
It does not make the underlying analytical multiplicity disappear.
If a complex hypothesis predicts effects on four separate outcomes and you test all four separately, you still have several inferential claims regardless of whether they were printed beneath one hypothesis number. The distinction between simple and complex hypotheses is therefore conceptual, not a loophole around multiple testing.
Exploratory Questions Do Not Need to Be Disguised as Formal Hypotheses
A researcher may have several additional relationships worth examining but insufficient prior evidence to justify strong predictions. These can be reported honestly as exploratory questions or analyses.
There is no methodological prize for converting every curiosity into an H-number.
Exploratory analyses can generate valuable hypotheses for subsequent work. Their interpretation should simply reflect their exploratory role, particularly when many analyses were conducted.
Preregistration Can Force Useful Prioritization
Specifying hypotheses and analysis plans before examining the relevant results can reveal whether a study has become unwieldy. Writing down the primary hypothesis, secondary hypotheses, outcomes, and analytical decisions forces researchers to decide what matters most before the p-values begin negotiating for promotion.
Preregistration is not mandatory for every research design, nor does it make a large hypothesis set automatically defensible. It can, however, make the distinction between planned and exploratory analyses more transparent.
