How to Write a Research Hypothesis: Definition, Types, and Examples

A research hypothesis is a testable prediction about what a study expects to find. It states a likely relationship, difference, or effect involving defined variables and a relevant population. For example, “Among first-year students, more hours of weekly retrieval practice will be associated with higher biology quiz scores” predicts a measurable relationship. A hypothesis is provisional rather than a fact: the data may support it, fail to support it, or reveal a more complicated pattern. A good hypothesis grows from a research question, theory, and prior evidence. It should be specific enough that another reader can identify what is being measured and what outcome would contradict the prediction. It should also match the study design. If a survey records two variables at one time point, the hypothesis normally describes an association rather than claiming that one variable causes the other. Not every study needs a hypothesis; exploratory and many qualitative studies may be better organized around research questions.

Begin by defining a hypothesis as a testable prediction, then choose the type that fits your evidence and research design. Write the research question first, review the theoretical and empirical rationale, and identify the population, predictor or independent variable, outcome or dependent variable, and comparison. Decide whether evidence justifies a directional prediction. If it does, state higher, lower, positive, or negative; if it does not, predict a difference or relationship without direction. Define how each variable will be measured and use causal wording only when the design can support causal inference. A practical form is: “Among [population], [exposure or group] will be associated with [direction] in [outcome] compared with [comparison].” For example, “Among first-year nursing students, weekly retrieval quizzes will be associated with higher final examination scores than standard review alone.” Finally, confirm that possible data could contradict the statement and that the planned analysis directly evaluates it.

A research question asks what the study seeks to discover, while a hypothesis predicts the answer expected from the study. “Is bedtime screen use related to sleep duration among adolescents?” is a research question. “Greater screen use during the hour before bedtime will be associated with shorter sleep duration among adolescents” is a hypothesis. The question can remain open when evidence is limited, the work is exploratory, or the methodology aims to understand experiences and processes. A hypothesis is appropriate when theory or previous research supports a specific, testable expectation. Both should align with the problem, objectives, variables, sample, method, and analysis. Do not convert every question into a hypothesis mechanically. Some qualitative designs use open questions because a fixed prediction could constrain inquiry, and some descriptive studies estimate prevalence without testing a predicted relationship. Follow disciplinary, supervisor, protocol, and journal expectations. When both appear, present them consistently so the hypothesis answers the question rather than introducing a different population, variable, or outcome.

The main labels describe the number of variables, expected direction, statistical role, or nature of the claim. A simple hypothesis links one predictor or independent variable with one outcome or dependent variable. A complex hypothesis includes multiple predictors, outcomes, mediators, moderators, or groups. A directional hypothesis predicts the direction of a difference or association; a non-directional hypothesis predicts that a difference or association exists without saying which way it will go. In statistical testing, the null hypothesis specifies no stated effect, difference, or association under the model, while the alternative hypothesis represents the competing prediction. An associative hypothesis describes covariation without claiming causality. A causal hypothesis predicts that changing one factor changes another and therefore requires a design and assumptions capable of supporting that interpretation. These categories can overlap. A hypothesis may be simple, directional, alternative, and associative at the same time. Choose labels for methodological clarity, not merely to fill a proposal template.

The null hypothesis, usually written H0, represents no specified population difference, effect, or association under the statistical model. The alternative hypothesis, written H1 or Ha, represents the effect or relationship against which the null is evaluated. If a study compares two teaching methods, the null might state that their population mean scores are equal. A two-sided alternative states that the means differ; a one-sided alternative states that one mean is higher or lower. Statistical testing does not normally “prove” either statement. Researchers assess how compatible the observed data are with the null model under defined assumptions, then report the estimate, uncertainty, and test result. “Fail to reject H0” does not prove that no meaningful effect exists; the study may have limited precision or power. Similarly, rejecting H0 does not establish practical importance or causality. Define the effect, direction, significance threshold, and analysis plan before examining outcomes, and interpret results alongside confidence intervals, study limitations, and subject-matter relevance.

Use a directional hypothesis when credible theory or prior evidence supports a specific expectation such as higher, lower, positive, or negative. Use a non-directional hypothesis when you expect a relationship or difference but the direction is genuinely uncertain, evidence is sparse, or previous findings conflict. The decision should be made before inspecting outcome data. A directional statement can correspond to a one-sided statistical test, but the two choices are not automatically identical; the analysis must follow disciplinary practice and be justified in the protocol. One-sided testing has an important consequence: a substantial effect in the unanticipated direction is not treated as support for the pre-specified alternative. Therefore, direction should never be chosen simply because it makes statistical significance easier to obtain. If both directions would matter scientifically or practically, a two-sided approach is often more suitable. Explain the rationale in the literature review or methods, and report unexpected results transparently rather than rewriting the original prediction.

Qualitative research often does not require a formal predictive hypothesis. Designs intended to understand lived experience, meaning, context, process, or theory development commonly use open research questions. Forcing a fixed prediction into such a study can narrow attention prematurely or conflict with an inductive methodology. However, qualitative projects are diverse. Theory-informed qualitative work may use sensitizing concepts, propositions, or expectations, and mixed-methods research may include quantitative hypotheses alongside qualitative questions. The correct choice depends on the research paradigm, design, discipline, university requirements, and purpose of the study. State what guides the inquiry without pretending to perform statistical hypothesis testing when that is not the method. If a proposal form requests a hypothesis, discuss the issue with the supervisor or methods adviser rather than inventing one that does not fit. The essential requirement is alignment: the problem, questions, conceptual framework, sampling, data collection, analysis, and claims should form a coherent methodological plan.

There is no universal correct number. Use as many hypotheses as the research questions, design, sample size, and analysis can justify, but no more than the study can evaluate clearly. A focused project may have one primary hypothesis. A larger thesis may include one primary hypothesis plus several secondary hypotheses linked to separate outcomes, mechanisms, or subgroups. Too many hypotheses increase analytical complexity and the risk of selective emphasis or chance findings, particularly when multiple tests are performed. Label primary, secondary, and exploratory hypotheses and ensure each has a rationale, operational definitions, relevant data, and a planned analysis. Avoid splitting one idea into many nearly identical statements only to make the project appear more substantial. Conversely, do not compress unrelated outcomes into one long hypothesis that becomes impossible to interpret. Consult a methods or statistics adviser early when multiplicity, mediation, moderation, repeated measures, or subgroup analysis is involved.

You can refine ideas after data collection, but you must report the timing and status transparently. A hypothesis specified before outcome analysis is confirmatory. A pattern noticed after examining the data is exploratory or post hoc and should be labeled accordingly. Changing a hypothesis to match an observed result and presenting it as pre-specified can mislead readers about the strength of the evidence. If an operational detail must change because a measure failed or a protocol assumption was wrong, document the reason, timing, and analytical consequence. Preserve the original hypothesis and analysis plan where possible, report deviations, and consider testing the new prediction in an independent dataset or future study. Unexpected findings can generate valuable research questions; transparency does not make them unimportant. It simply helps readers distinguish prediction from discovery and evaluate uncertainty appropriately. Requirements vary across fields, registries, funders, ethics boards, and journals, so consult the relevant protocol and reporting guidance.

Seek expert help when the difficulty involves more than sentence-level wording. A supervisor or subject specialist should review whether the prediction follows from the literature and theory. A methods or statistics adviser should be involved when you are uncertain about operational definitions, sample size, causal assumptions, repeated measures, clustering, mediation, moderation, multiple outcomes, or the analysis needed to test the hypothesis. An academic editor is useful when the underlying reasoning is established but the statement is ambiguous, overly long, grammatically unclear, inconsistent with the abstract or methods, or too strong for the design. Ethical editing should preserve your ideas and explain significant revisions; it should not invent a rationale, fabricate references, design the study without author involvement, or guarantee approval or publication. Share the research question, brief rationale, variable definitions, design, and intended analysis so the reviewer can assess alignment in context. You remain responsible for every claim and final submission.