Research Hypothesis – Definition, Types, How to Write & Examples

A research hypothesis is a clear, testable prediction about an expected relationship, difference, or outcome in a study. It draws on existing evidence or theory and matches the research question and design. Not every study needs one: exploratory qualitative work often runs on open questions instead.

Take a student who wonders whether sleep helps memory. The research question is: Does sleep after studying improve recall? The research hypothesis is: Undergraduates who sleep at least seven hours after studying will recall more material one week later than undergraduates who sleep five hours or less. One asks. The other predicts, says who is being compared, and can turn out wrong.

That gap is what makes a research hypothesis useful, and it’s also where most students get stuck. This guide explains the hypothesis in research in plain terms, then goes further than most guides do – 27 worked examples with variables, design, and a reason each one is testable; a clear split between a research hypothesis and the statistical H₀ and H₁; an honest answer on qualitative research; and advice on where the hypothesis sits in a paper or thesis. Every example here is illustrative. None is a published finding.

What Is a Research Hypothesis?

A research hypothesis is a testable prediction about what a study will find. It states an expected relationship between variables, an expected difference between groups, or an expected effect of an intervention, and it rests on earlier studies or theory instead of a hunch. Barroga and Matanguihan (2022) describe it as a prediction grounded in existing knowledge and a tentative answer to the research question.

Three points keep the definition honest. A hypothesis isn’t a fact; the study exists to find out whether data support it. It isn’t a random guess either, because you should be able to explain why you expect that result. And it isn’t a question. “Does sleep affect memory?” asks. “Sleep-deprived students will recall less,” predicts.

The hypothesis meaning in research gets stretched in everyday talk to cover any assumption. In methods writing, it’s narrower: a statement that a dataset could count against.

Research Hypothesis vs. Statistical Hypothesis

The research hypothesis states the prediction in the language of your topic. Statistical hypotheses (H₀ and H₁) restate it as claims about a population so a test can evaluate it. They describe the same study at two levels.

Level	What it says	Example
Research hypothesis	The substantive prediction, in plain words	Year 9 students who take weekly retrieval quizzes will score higher on an end-of-term vocabulary test than students who restudy the same word lists.
Alternative hypothesis (H₁)	The same prediction as a claim about population means	μ(quiz) > μ(restudy)
Null hypothesis (H₀)	The claim the test examines directly	μ(quiz) ≤ μ(restudy)

Many papers state only the research hypothesis and leave H₀ implied. Others, especially in statistics-heavy fields, write both. The section on null and alternative hypotheses below explains how to read the result of the test.

Notation also varies. For a one-sided prediction, H₀ and H₁ should together cover every possibility, so H₀ is written μ(quiz) ≤ μ(restudy) against H₁: μ(quiz) > μ(restudy). Some textbooks write H₀ as an equality, μ(quiz) = μ(restudy), and keep the one-sided H₁; the test behaves the same at that boundary. Follow the convention of your course or journal, and keep H₀ and H₁ consistent throughout the paper.

What a Research Hypothesis Does in a Study

A hypothesis isn’t decoration. It does four jobs:

• It tells you what to measure, so you don’t collect data that answers nothing.
• It shapes the design and the sample size, because a prediction about a small difference needs more participants than one about a large difference.
• It points to the analysis. A prediction about group differences leads to one family of tests, a prediction about association to another.
• It gives the results something to be judged against. Stating the prediction before you see the data guards against hypothesising after the results are known, a habit sometimes called HARKing. Preregistering the hypothesis on a public registry makes that protection visible.

What Makes a Good Research Hypothesis?

A good research hypothesis has five characteristics: it is specific, falsifiable, grounded in evidence, built on variables you can measure, and matched to the study design. Check a draft against those five points before you commit to it.

Specific and Clear

Name the population, the variables, and the expected direction. “Social media affects students” fails because nobody can say what would count as affecting, or which students. “Undergraduates who spend more than three hours a day on social media will report higher loneliness scores than undergraduates who spend less than one hour” can be checked.

Testable and Falsifiable

Testable means you can collect data that bears on the prediction. Falsifiable, a principle associated with Karl Popper, means some possible result would count against it. “Some students may benefit from tutoring” can never be contradicted, because “some” and “may” survive any result. “Students who attend weekly tutoring will score higher on the final exam than comparable students who don’t” can be contradicted by a lower or equal mean.

Grounded in Research or Theory

Your reason for expecting a result should come from earlier studies, a theory, or a clearly described pattern in pilot data. That reason is what separates a hypothesis from a guess, and reviewers look for it in the introduction.

Identifies Relevant Variables

A variable is anything that can take different values. Say what each one is and how you’ll measure it. “Stress” is a topic; “score on the Perceived Stress Scale” is a variable.

Variable role	What it is	Example
Independent (predictor)	The factor expected to produce or relate to change	Hours of sleep
Dependent (outcome)	The factor you measure to see the change	Recall score
Moderator	Changes the strength or direction of a relationship	Age group
Mediator	Carries part of the effect from predictor to outcome	Attention level
Control (covariate)	Held constant or adjusted for	Prior grade average

Fits the Study Design

The wording of your hypothesis has to match what the design can support. Random assignment to conditions allows causal wording such as “will increase.” A survey or records study supports associative wording such as “is associated with.” Writing a causal hypothesis for a design that can’t isolate cause sets you up for conclusions the data can’t carry. Even with random assignment, a causal reading depends on how the study is run: confounding, non-compliance, attrition, and measurement problems can all weaken it.

Weak vs. Strong: Three Rewrites

Weak version	Stronger version	What changed
Technology affects learning.	Secondary students who use a spaced-repetition app for ten minutes daily over eight weeks will score higher on an end-of-term science test than students who don’t use it.	Name the population, the exposure, the time frame, the outcome, and the direction.
Exercise is good for mental health.	Adults who complete 150 minutes of moderate exercise per week for 12 weeks will report lower depressive symptom scores on the PHQ-9 than adults in a waiting-list group.	“Good for” becomes a measured outcome and a comparison group.
Customers like discounts.	Shoppers shown a 10% discount banner will complete more purchases per session than shoppers shown no banner.	“Like” becomes an observable behaviour.

Research Question vs. Research Hypothesis

A research question asks what a study will find out. A research hypothesis predicts the answer. Most quantitative studies have a question and one or more hypotheses derived from it, and the two should say the same thing in different forms.

#	Research question	Research hypothesis
Form	An open question	A declarative prediction
Job	Defines what the study sets out to learn	States what you expect the data to show
Resolved by	An answer developed in the discussion	Evidence that supports or fails to support it
Typical use	Always	When earlier work or theory justifies a prediction
Example	Is lecture attendance related to exam performance?	Higher lecture attendance is associated with higher exam scores among first-year undergraduates.

The University of Northern Iowa’s research methods text gives a practical rule: write a question when a topic is new and nothing in the literature lets you predict the outcome, and write a hypothesis when previous research or theory supports a specific expectation. A single research question can also produce several hypotheses, one for each prediction it contains.

If you’re still shaping the question itself, our research question examples guide covers formats and wording in depth, so this article doesn’t repeat it.

From a Research Question to a Prediction

Here is one topic moving through the stages.

1. Topic: remote work and productivity.
2. Research question: Do employees who work remotely two days a week complete more tasks than employees who work on site five days a week?
3. Prediction in plain words: Hybrid workers will get more done.
4. Research hypothesis: Employees on a two-day remote schedule will complete more tasks per week, as logged in the company project system, than employees on a five-day on-site schedule in the same department.

The last version adds the people, the measure, the comparison, and the direction. It also makes a risky claim, which is the point. A hypothesis that can’t lose can’t teach you anything.

Types of Research Hypotheses

The common types of hypotheses in research are null, alternative, directional, non-directional, simple, complex, associative, and causal. They aren’t eight competing boxes. Each type answers a different question about the same statement, so one hypothesis can carry several labels at once.

Question you’re asking	Types	Example label
What role does it play in a statistical test?	Null, alternative	Alternative (H₁)
Does it predict a direction?	Directional, non-directional	Directional
How many variables does it involve?	Simple, complex	Simple
What kind of link does it claim?	Associative, causal	Causal

A well-formed prediction such as “Students who take weekly retrieval quizzes will score higher than students who restudy” is, at once, an alternative, directional, simple, and causal hypothesis. Classifications also vary by discipline and by textbook, so use the labels your supervisor, journal, or field expects.

Null Hypothesis (H₀)

The null hypothesis is the claim a statistical test examines directly. It usually states that there is no relationship, difference, or effect in the population; in a one-sided test, it states that any effect is not in the predicted direction. It works as the default position the test challenges. The National University statistics guide calls it a presumption of status quo or no change.

A null hypothesis in research example: There is no difference in mean vocabulary test scores between students who take weekly retrieval quizzes and students who restudy word lists. Read it as a position for testing, not a belief you hold.

Alternative Hypothesis (H₁ or Hₐ)

The alternative hypothesis states what you expect the data to support: an effect, a difference, or a relationship. It’s the counterpart of H₀, and rejecting H₀ is what leads you to favour it. Both H₁ and Hₐ are in use; choose one and keep it consistent. This article uses H₁.

For the alternative hypothesis in research, the example is: Mean vocabulary test scores differ between students who take weekly retrieval quizzes and students who restudy word lists.

Directional Hypothesis

A directional hypothesis predicts which way the effect will go: higher, lower, faster, more likely. It suits cases where theory or earlier studies justify a direction, and it can be tested directionally (with a one-tailed test) when the research question justifies it and the analysis plan specifies it in advance. Choose the direction and the test before you see the data. The right test also depends on your design and its assumptions, not on direction alone.

Example: Employees on a two-day remote schedule will complete more tasks per week than employees on a five-day on-site schedule.

Non-Directional Hypothesis

A non-directional hypothesis predicts a difference or relationship without saying which way. It fits situations where no theory exists or earlier findings conflict, and it’s commonly tested with a two-tailed test, though the choice of test should follow your design and prespecified analysis plan, as Barroga and Matanguihan note.

Example: Task completion differs between employees on a two-day remote schedule and employees on a five-day on-site schedule.

Simple Hypothesis

A simple hypothesis links one independent variable to one dependent variable.

Example: Higher daily screen time is associated with shorter sleep duration among teenagers.

Complex Hypothesis

A complex hypothesis involves three or more variables: several predictors, several outcomes, or a moderator or mediator.

Example: Workload is positively associated with burnout among nurses, and this association is weaker for nurses who report high supervisor support.

Associative Hypothesis

An associative hypothesis says variables change together and makes no claim that one produces the other. It’s the right wording for correlational and many observational designs.

Example: Higher household income is associated with higher school attendance among children aged 6 to 12.

Causal Hypothesis

A causal hypothesis claims that changing one variable produces a change in another. A hypothesis can be worded causally, but the data can only support that reading if the design allows it, usually through random assignment or a strong quasi-experimental setup with controls, and only if the study is carried out well.

Example: Participants randomly assigned to a five-hour sleep condition will make more errors on a sustained attention task than participants assigned to eight hours.

Other Labels You Will See

Some sources add statistical hypotheses (claims about parameters), empirical hypotheses (claims checked against observation), logical hypotheses, and working hypotheses (provisional statements that guide early work). This is why the answer to “how many types of hypotheses are there?” depends on the source: two, three, seven and eleven all appear in print. The labels overlap, so understanding the four questions in the table above is more useful than memorising a number.

Research Hypothesis Examples

The first 24 hypothesis examples below cover eight fields, and three quantitative examples (25 to 27) follow in their own section. Each one shows the research question, the hypothesis, the variables, the type and design, and why it can be tested. They’re written for teaching, so the populations and measures are plausible but invented. Treat each one as an example of a hypothesis in research to adapt, not to copy: keep the structure, change the content, and check that your version fits your own design and literature. Where an example is labelled causal, that describes the claim the hypothesis makes. Whether data can support a causal reading depends on how the study is run: random assignment, compliance, attrition and measurement all matter.

Education Hypothesis Examples

Example 1: Retrieval practice and vocabulary
Research question	Does retrieval practice improve vocabulary retention in secondary school students?
Research hypothesis	Year 9 students who complete weekly retrieval quizzes will score higher on an end-of-term vocabulary test than students who restudy the same word lists.
Variables	Independent: study method (quizzing or restudy). Dependent: vocabulary test score.
Type and design	Directional, causal. Randomised classroom experiment.
Why it is testable	Both groups sit the same test, and a lower or equal quiz-group mean would count against the prediction.

Example 2: Lecture attendance and exam scores
Research question	Is lecture attendance related to exam performance among first-year university students?
Research hypothesis	Among first-year undergraduates, higher lecture attendance rates are associated with higher final exam scores.
Variables	Predictor: attendance rate (percentage of sessions attended). Outcome: final exam score.
Type and design	Directional, associative. Cross-sectional study using course records.
Why it is testable	The sign and size of the correlation can be computed from existing records, and a zero or negative coefficient would contradict it.

Example 3: Flipped classroom and engagement
Research question	Does a flipped classroom change student engagement compared with traditional lecturing?
Research hypothesis	Mean engagement scores will differ between students in flipped and traditional sections of the same course.
Variables	Independent: teaching format. Dependent: score on a validated engagement scale.
Type and design	Non-directional, comparative. Quasi-experimental comparison of two intact sections.
Why it is testable	Findings on flipped formats are mixed, so a non-directional prediction is honest, and the difference in means can be tested.

Healthcare and Nursing Hypothesis Examples

Example 4: Hourly rounding and patient falls
Research question	Does a structured hourly rounding protocol reduce patient falls on medical wards?
Research hypothesis	Wards that adopt hourly rounding will record fewer falls per 1,000 patient days than wards that continue usual care.
Variables	Independent: rounding protocol. Dependent: falls per 1,000 patient days.
Type and design	Directional, causal. Quasi-experimental comparison of wards, ideally with a pre-implementation baseline.
Why it is testable	Fall rates come from incident reports, and a higher or equal rate on intervention wards would count against it.

Example 5: Staffing ratios and patient satisfaction
Research question	Is the patient-to-nurse ratio related to patient satisfaction?
Research hypothesis	Higher patient-to-nurse ratios are associated with lower patient satisfaction scores across hospital units.
Variables	Predictor: average patient-to-nurse ratio per shift. Outcome: unit-level satisfaction score.
Type and design	Directional, associative. Observational, cross-sectional.
Why it is testable	Both measures already exist in hospital records. Because the design is observational, the wording says “associated with,” not “reduces.”

Example 6: Supervised walking and blood pressure
Research question	Does a supervised walking programme lower resting systolic blood pressure in adults with mild hypertension?
Research hypothesis	Adults with mild hypertension who complete a 12-week supervised walking programme will show a larger drop in resting systolic blood pressure than adults who receive health education only.
Variables	Independent: programme (walking or education only). Dependent: change in resting systolic blood pressure (mmHg).
Type and design	Directional, causal. Randomised controlled trial.
Why it is testable	Blood pressure is measured with a calibrated device before and after, and random assignment supports a causal reading.

Psychology Hypothesis Examples

Example 7: Sleep restriction and attention
Research question	Does sleep restriction impair sustained attention the next day?
Research hypothesis	Participants restricted to five hours of sleep will make more errors on a sustained attention task than participants who sleep eight hours.
Variables	Independent: sleep condition. Dependent: error count on the task.
Type and design	Directional, causal. Laboratory experiment with random assignment.
Why it is testable	Errors are counted objectively, and equal or fewer errors in the restricted group would contradict it.

Example 8: Social media use and loneliness
Research question	Is daily social media use related to loneliness among university students?
Research hypothesis	Daily social media use is related to loneliness scores among undergraduates.
Variables	Predictor: self-reported daily minutes on social media. Outcome: score on a loneliness scale.
Type and design	Non-directional, associative. Cross-sectional survey.
Why it is testable	Earlier studies report positive, negative, and near-zero links, so committing to a direction isn’t justified. A correlation of zero would count against it.

Example 9: Breathing exercise and test anxiety
Research question	Does a brief breathing exercise reduce test anxiety?
Research hypothesis	Students who complete a ten-minute guided breathing exercise before a test will report lower state anxiety than students who sit quietly for ten minutes.
Variables	Independent: pre-test activity. Dependent: state anxiety score.
Type and design	Directional, causal. Randomised experiment with an active control.
Why it is testable	Anxiety is measured with a standard self-report scale at the same point for both groups.

Business and Management Hypothesis Examples

Example 10: Flexible start times and turnover
Research question	Is flexible scheduling linked to employee retention?
Research hypothesis	Employees offered flexible start times will have lower 12-month turnover than employees on fixed schedules in the same company.
Variables	Independent: scheduling arrangement. Dependent: whether the employee left within 12 months.
Type and design	Directional, comparative. Observational comparison of HR records.
Why it is testable	Turnover is recorded. Employees may choose their arrangement, so the result describes a difference, not a proven effect.

Example 11: Leadership style and team performance
Research question	Is transformational leadership related to team performance?
Research hypothesis	Higher team ratings of their manager’s transformational leadership are associated with higher supervisor-rated team performance.
Variables	Predictor: transformational leadership score. Outcome: performance rating.
Type and design	Directional, associative. Survey with ratings collected from separate sources.
Why it is testable	Two separate sources reduce shared-method bias, and a weak or negative correlation would count against the prediction.

Example 12: Workload, burnout and managerial support
Research question	Does managerial support change the link between workload and burnout?
Research hypothesis	Workload is positively associated with burnout, and this association is weaker among employees who report high managerial support.
Variables	Predictor: workload. Outcome: burnout. Moderator: managerial support.
Type and design	Complex, directional, associative. Survey analysed with moderated regression.
Why it is testable	The interaction term in the regression tests whether the slope really differs across support levels.

Marketing Hypothesis Examples

Example 13: Personalised subject lines
Research question	Do personalised email subject lines increase open rates?
Research hypothesis	Emails whose subject line includes the recipient’s first name will have a higher open rate than otherwise identical emails without it.
Variables	Independent: subject line version. Dependent: open rate.
Type and design	Directional, causal. A/B test with random assignment of recipients.
Why it is testable	Opens are logged for both versions during the same send window.

Example 14: Brand trust and repurchase intention
Research question	Is brand trust related to repurchase intention among online shoppers?
Research hypothesis	Higher brand trust is associated with higher repurchase intention among online shoppers.
Variables	Predictor: brand trust score. Outcome: repurchase intention score.
Type and design	Directional, associative. Online survey.
Why it is testable	Both constructs are measured with multi-item scales, and the correlation can be estimated with confidence intervals.

Example 15: Age and response to influencer endorsements
Research question	Do younger and older consumers respond differently to influencer endorsements?
Research hypothesis	Purchase intention after viewing an influencer endorsement will differ between consumers aged 18 to 24 and consumers aged 45 to 60.
Variables	Independent: age group. Dependent: purchase intention score.
Type and design	Non-directional, comparative. Survey experiment with a shared stimulus.
Why it is testable	Both groups see the same endorsement and answer the same items, so the group means can be compared.

Computer Science and Technology Hypothesis Examples

Example 16: Transformer model vs. baseline for phishing detection
Research question	Does a transformer-based classifier detect phishing emails more accurately than a logistic regression baseline?
Research hypothesis	On the same held-out test set, the transformer model will achieve a higher F1 score than the logistic regression baseline.
Variables	Independent: model type. Dependent: F1 score.
Type and design	Directional, comparative. Benchmark experiment with repeated runs and fixed data splits.
Why it is testable	Both models are scored on identical data, and repeated runs show whether the gap exceeds run-to-run variation.

Example 17: Load time and app abandonment
Research question	Is page load time related to mobile app abandonment?
Research hypothesis	Longer screen load times are associated with higher abandonment rates in the app’s session logs.
Variables	Predictor: median load time per session. Outcome: abandonment (session ended before task completion).
Type and design	Directional, associative. Observational analysis of log data.
Why it is testable	Logs record both variables. Device type and network quality may confound the link, so they should be included as controls.

Example 18: Redesigned onboarding flow
Research question	Does a redesigned onboarding flow help new users finish setup faster?
Research hypothesis	Users who see the redesigned onboarding flow will complete setup in less time than users who see the original flow.
Variables	Independent: onboarding version. Dependent: time to complete setup (seconds).
Type and design	Directional, causal. Between-subjects usability experiment.
Why it is testable	Completion time is timestamped, and random assignment keeps user type similar across versions.

Economics and Social Science Hypothesis Examples

Example 19: Household income and school attendance
Research question	Is household income related to school attendance in rural districts?
Research hypothesis	Higher household income is associated with higher school attendance rates among children aged 6 to 12 in rural districts.
Variables	Predictor: monthly household income. Outcome: attendance rate.
Type and design	Directional, associative. Household survey or secondary data.
Why it is testable	Both variables can be measured, and a flat or negative association would count against it.

Example 20: Cash transfers and school enrolment
Research question	Does a conditional cash transfer programme increase school enrolment?
Research hypothesis	Children in households that receive the transfer will have higher enrolment rates than comparable children in households that don’t.
Variables	Independent: programme participation. Dependent: school enrolment.
Type and design	Directional, causal. Randomised rollout or a quasi-experimental design such as difference-in-differences.
Why it is testable	Enrolment is recorded by schools, and the comparison group stands in for what would have happened without the programme.

Example 21: Social trust and voter turnout
Research question	Is neighbourhood social trust related to voter turnout?
Research hypothesis	Neighbourhoods with higher average social trust scores will have higher voter turnout.
Variables	Predictor: neighbourhood trust score. Outcome: turnout rate.
Type and design	Directional, associative. Ecological (area-level) analysis.
Why it is testable	Both measures exist at neighbourhood level. The result can’t be applied to individual voters, which is a known limit of area-level data.

Environmental Research Hypothesis Examples

Example 22: Tree canopy and surface temperature
Research question	Do city blocks with more tree canopy have cooler surfaces in summer?
Research hypothesis	City blocks with higher tree canopy cover will have lower mean summer afternoon surface temperatures than blocks with lower canopy cover.
Variables	Predictor: canopy cover (percentage). Outcome: surface temperature from satellite imagery.
Type and design	Directional, associative. Observational, spatial analysis controlling for building density.
Why it is testable	Both measures come from remote sensing, and building density is included so canopy isn’t credited with an effect that belongs to something else.

Example 23: Nutrient load and algal growth
Research question	Does nitrogen input raise algal biomass in lakes?
Research hypothesis	Lake sites with higher nitrogen loads will show higher chlorophyll-a concentrations during the growing season.
Variables	Predictor: nitrogen load. Outcome: chlorophyll-a concentration.
Type and design	Directional, associative in the field; causal if tested in a controlled mesocosm experiment.
Why it is testable	Chlorophyll-a is a standard measure of algal biomass, and the same hypothesis can be tested under field or controlled conditions.

Example 24: Bag levy and plastic bag use
Research question	Does a plastic bag levy reduce plastic bag use at supermarkets?
Research hypothesis	The number of plastic bags issued per transaction will fall after the levy takes effect.
Variables	Independent: levy (before or after). Dependent: bags issued per transaction.
Type and design	Directional, causal in intent. Interrupted time series using till data.
Why it is testable	Till data give many time points before and after, which helps separate the levy from a pre-existing trend. A simple before-and-after comparison would be weaker.

Quantitative Research Hypothesis Examples

A hypothesis in quantitative research is a prediction that numerical data can test, usually through a statistical test. The three examples below follow the three common aims: relating variables, comparing groups, and testing an intervention. They stay on the hypothesis itself. For full study designs, data collection, and test selection, see our quantitative research examples guide.

Correlational Example

Example 25: Physical activity and well-being
Research question	Is weekly physical activity related to well-being among office workers?
Research hypothesis	Weekly minutes of moderate to vigorous activity are positively correlated with well-being scores among full-time office workers.
Statistical form	H₀: ρ ≤ 0. H₁: ρ > 0.
Test	Pearson correlation if the relationship looks roughly linear and the data meet its assumptions. Spearman rank correlation if the relationship is monotonic but not linear, the scales are ordinal, or outliers distort the data. Check a scatterplot first.
How to read it	A positive r shows the variables move together. It doesn’t show that activity raises well-being; people who feel well may simply exercise more.

Comparative Example

Example 26: Commute mode and commute stress
Research question	Do cyclists and drivers differ in commute stress?
Research hypothesis	Mean commute stress scores differ between adults who cycle to work and adults who drive.
Statistical form	H₀: μ(cycling) = μ(driving). H₁: μ(cycling) ≠ μ(driving).
Test	Independent-samples t-test, using Welch’s version if group variances differ.
How to read it	Report the difference in means with a confidence interval and an effect size, not only the p-value. Commuters chose their mode, so confounders such as distance and job type need attention.

Experimental or Intervention Example

Example 27: Spaced vs. massed practice
Research question	Does spaced practice produce better delayed recall of statistics terms than massed practice?
Research hypothesis	Participants randomly assigned to study over four short sessions across one week will score higher on a delayed test than participants assigned to study the same material in one session of equal total length.
Statistical form	H₀: μ(spaced) ≤ μ(massed). H₁: μ(spaced) > μ(massed).
Test	One-tailed independent-samples t-test on delayed-test scores, as set in the analysis plan before data collection. If you also collect a prior-knowledge pretest before the conditions begin, ANCOVA with that pretest as a covariate is a common alternative; choose it in the plan, not after seeing the data.
How to read it	Random assignment lets you read a difference causally if the study ran cleanly, and a pretest taken before assignment can sharpen precision without changing the hypothesis. A non-significant result still wouldn’t show spacing has no effect.

Null and Alternative Hypothesis Examples

Null and alternative hypotheses are paired claims about a population. H₀ specifies the no-effect, no-difference, or other null claim being tested. H₁ is the competing claim, and it’s what your research hypothesis predicts. The test asks how surprising your data would be if H₀ and the model assumptions were true.

H₀ and H₁ in Plain English

Research hypothesis	H₀ in words	H₁ in words
Weekly retrieval quizzes raise vocabulary scores	Quizzing makes no difference to mean scores	Quizzing raises mean scores
Commute mode relates to commute stress	Mean stress is the same for cyclists and drivers	Mean stress differs between cyclists and drivers
Activity relates to well-being	There is no correlation	There is a positive correlation
Hourly rounding lowers fall rates	Fall rates are the same or higher with rounding	Fall rates are lower with rounding

Directional H₁ Example

Using Example 1, where μ is the population mean end-of-term vocabulary score for each study method: H₁: μ(quiz) > μ(restudy). H₀: μ(quiz) ≤ μ(restudy). With two independent groups, the matching test is a one-tailed independent-samples t-test (Welch’s version if the group variances differ). A directional H₁ concentrates the test on one tail, so it can detect an effect in the predicted direction with less data. The cost is that an effect in the opposite direction can’t count as a finding. Specify the direction and the test in your analysis plan before data collection or analysis, and only when earlier evidence justifies it. Picking a tail after seeing which way the data point inflates false positives. The choice of test also rests on your design and assumptions, not on direction alone.

Non-Directional H₁ Example

Using Example 3, where μ is the population mean score on the engagement scale in each course format: H₁: μ(flipped) ≠ μ(traditional). H₀: μ(flipped) = μ(traditional). The matching test is a two-tailed independent-samples t-test (Welch’s if variances differ). Because the design compares two intact sections instead of randomised groups, baseline differences between the sections need attention. A two-tailed test treats a difference in either direction as evidence against H₀. It’s the safer default when findings in the literature disagree.

Research Hypothesis vs. Statistical Hypotheses

The research hypothesis is what you believe and why. The statistical hypotheses are the formal pair the test evaluates. Reading the result correctly is where many write-ups go wrong, so here are the rules that matter:

• Set the significance level (α) before analysis. 0.05 is a convention, not a law of nature, as the StatPearls overview of hypothesis testing notes.
• If p ≤ α, the data would be unusual if H₀ and the model’s assumptions held. You reject H₀, and the results are consistent with H₁. That isn’t proof that H₁ is true.
• If p > α, you fail to reject H₀. This doesn’t show H₀ is true. A small sample or noisy measure can hide a real effect, and Greenland and colleagues (2016) list the belief that a non-significant result shows no effect among the most common misreadings of p-values.
• Know the two errors. A Type I error rejects a true H₀. A Type II error fails to reject a false H₀. Larger samples reduce Type II errors. A stricter α reduces Type I errors but, with the sample held fixed, raises the risk of Type II errors.
• Don’t equate significance with validity. A significant result doesn’t show that the hypothesis is correct, that the effect is large enough to matter, or that the study was sound. Read it alongside the effect size, the confidence interval, the sample limits, and the design.
• Report more than p. Give the effect size and a confidence interval so readers can judge how large and how precise the result is, and state the study’s limitations.

How to Write a Research Hypothesis Step by Step

To write a research hypothesis, move from a focused question to a specific, testable prediction in seven steps:

1. Start with a focused research question.
2. Review existing evidence and theory.
3. Identify and define the variables.
4. Specify the population or context.
5. Predict a relationship, difference, or effect.
6. Check that the prediction is testable.
7. Write the hypothesis, and H₀ and H₁ where needed.

The sections below run one example through every step: Does a short daily walk improve afternoon concentration in office workers?

Start With a Focused Research Question

A manageable question names a population, an exposure, and an outcome. “Is exercise good for work?” is too wide. “Does a ten-minute daily walk improve afternoon concentration in office workers?” isn’t. If your question can be answered yes or no without evidence, narrow it until it needs data.

Review Existing Evidence and Theory

This is the step that turns a guess into a hypothesis. Hypotheses come from several places: earlier studies, an established theory, patterns in pilot data, and the “future research” sections of papers in your field. Read enough to explain why you expect the result. Our guide to the review of related literature shows how to run that search and record what you find.

In the walking example, you might find studies linking brief physical activity with short-term attention. That gives you a reason and a direction.

Identify and Define the Variables

Name the predictor and the outcome, then say how you’ll measure each. Predictor: a ten-minute walk on five weekdays. Outcome: score on a standard attention task given at 3 p.m. Add any variable that might distort the result, such as caffeine intake, and decide how to control it.

Specify the Population or Context

Say who the claim applies to and where. “Full-time office workers in one company, aged 25 to 55” is a claim you can test. “Everyone” isn’t. A narrow population is easier to defend, and you can widen the claim in later studies.

Predict a Relationship, Difference, or Effect

Choose the verb that matches your design. Use “will score higher” or “will reduce” for an experiment. Use “is associated with” for an observational study. Use “differs” when you can’t justify a direction. Many fields write hypotheses in the future tense (“will”), others in the present (“is associated with”). Pick the convention your discipline uses and stay consistent.

Check Whether the Prediction Is Testable

Ask three questions. Can I measure both variables as defined? Is there a result that would contradict my prediction? Does my design match the claim? If any answer is no, revise before collecting data. This is the point where “how to formulate hypothesis in research” stops being abstract: the check forces you to see what the study can and can’t show.

Write the Research Hypothesis and, Where Needed, H₀/H₁

Now put it together in one sentence. Here’s the walking example: Office workers who take a ten-minute walk on each weekday will score higher on an afternoon attention task over four weeks than office workers who take no walk. In statistical form: H₀: μ(walk) ≤ μ(no walk); H₁: μ(walk) > μ(no walk).

These templates give you a reliable research hypothesis format. Replace the bracketed parts and keep the structure.

Hypothesis type	Template
Directional, causal	[Population] who [receive intervention] will show [higher/lower] [outcome] than [population] who [comparison condition].
Associative	Higher [predictor] is associated with [higher/lower] [outcome] among [population].
Comparative, non-directional	[Outcome] differs between [group A] and [group B].
Moderation	The association between [predictor] and [outcome] is [stronger/weaker] for [moderator level].
Null	There is no difference in [outcome] between [group A] and [group B].

Notice what the templates have in common: a named group, a measurable outcome, and a comparison. If one is missing, the statement isn’t ready.

Does Qualitative Research Have a Hypothesis?

Most qualitative studies don’t test a hypothesis. They begin with open-ended research questions and build understanding from the data, a process called inductive reasoning. Practice varies across traditions, however, so “never” is too strong.

Barroga and colleagues (2023) contrast the inductive approach of qualitative research with the deductive logic of quantitative hypothesis testing, and note that qualitative work can generate hypotheses from data instead of testing them in advance. Barroga and Matanguihan (2022) describe qualitative research questions as continually reviewed and reformulated as a study develops. Both are overview papers, so read their patterns as general guidance, not rules for every qualitative tradition.

In practice:

Approach	Typical guiding statement	Example
Phenomenology, ethnography, narrative inquiry	Open research question	How do first-generation students experience their first year?
Case study	Research question, sometimes with propositions	What factors shaped the hospital’s adoption of telehealth?
Grounded theory	Open question; tentative ideas may emerge during analysis	How do nurses manage conflicting priorities on night shifts?
Mixed methods	Questions for the qualitative strand, hypotheses for the quantitative strand	Qualitative: how do employees describe remote work? Quantitative: remote work days are associated with job satisfaction.

Two cautions. First, a statistical null hypothesis doesn’t apply to qualitative analysis, because there is no significance test to run. Second, expectations differ between departments and journals. If you’re writing a proposal, ask your supervisor what the department expects, and explain your reasoning in the methodology section if you work without a hypothesis.

Where Does the Hypothesis Go in a Research Paper?

There is no single place. In most empirical papers, the hypothesis appears at the end of the introduction, after the background and the gap, is applied in the methods and analysis plan, and is revisited in the results and discussion. In theses, institutions differ: it might sit in the introduction chapter, at the end of the literature review, or in the methodology chapter, so follow your institution’s template.

Proposal or Introduction Context

State the hypothesis after you’ve shown the gap and the reasoning that supports your prediction. Readers should reach it already convinced it’s a sensible bet. In a proposal, it often sits next to the research question and objectives. If your abstract needs a one-sentence statement of the aim or hypothesis, see how to write an abstract for a research paper.

Methodology and Testing Context

The methods section connects each hypothesis to a variable, a measure, and a test. A short table linking “Hypothesis → Variables → Measure → Test” shows reviewers that nothing is untested. Your wider research methodology explains the design choices that make those links possible.

Results and Discussion

Report each hypothesis in the order you stated it, with the statistic, effect size, and confidence interval. Then say plainly whether the data supported it, and state the sample and design limits that bear on that conclusion. A hypothesis that isn’t supported is still a result. Explain what it might mean (a measure that was too noisy, a population that differed, a theory that needs adjustment) without rewriting the hypothesis to match the data.

Research Hypothesis vs. Research Objective vs. Research Question

These three do different jobs, and thesis committees often check whether they line up.

#	Research question	Research objective	Research hypothesis
What it is	What you want to find out	What the study will do	What you predict you’ll find
Form	Question	Action statement (to examine, to compare)	Testable prediction
Example	Does a daily walk improve afternoon concentration?	To compare attention scores between walkers and non-walkers over four weeks	Walkers will score higher on an afternoon attention task than non-walkers

A hypothesis isn’t a theory, either. A theory is a broad, well-supported explanation built from many studies; a hypothesis is a specific, testable prediction that may be derived from one. Testing a hypothesis can strengthen or weaken a theory, but a single hypothesis doesn’t become a theory by being supported once.

Common Mistakes When Writing a Research Hypothesis

Making a Vague Prediction

“Stress affects performance” can survive almost any data. Name the population, the measure, and the direction: “Nursing students with higher perceived stress scores will earn lower clinical assessment marks.”

Using Variables That Cannot Be Defined or Measured

“Happiness” and “quality of life” aren’t measurable until you pick an instrument. State the scale, the observation, or the record that will stand in for the concept.

Claiming Causation From a Non-Causal Design

If the design is a survey or a records review, write “is associated with.” A hypothesis that says “causes” invites the criticism that confounders explain the result. Reserve causal wording for designs with random assignment or strong controls.

Writing a Question Instead of a Hypothesis

“Does remote work improve productivity?” is a research question. Convert it: “Employees on a two-day remote schedule will complete more tasks per week than employees on a five-day on-site schedule.”

Ignoring Existing Evidence

A hypothesis with no basis in prior work is a guess with formatting. Cite what leads you to the prediction, including studies that point the other way, and explain why you still expect your result.

Treating Every Study as Hypothesis-Driven

Descriptive and exploratory studies, many qualitative studies and some data-mining projects don’t predict an outcome in advance. Forcing a hypothesis onto them leads to invented predictions that the design can’t test. Use a research question instead.

Mismatching the Hypothesis and Research Question

If your question asks about differences between groups and your hypothesis predicts a correlation, a reviewer will see two different studies. Reread both side by side: same variables, same population, same kind of claim.

Frequently Asked Questions

Key Takeaway

A research hypothesis is a testable prediction, grounded in evidence, tied to variables you can measure and matched to your design. Write it after your research question and your review of the literature. Use H₀ and H₁ when a statistical test needs them, and read the results with care: failing to reject H₀ isn’t proof of no effect. Not every study needs a hypothesis, and a good research question is better than a forced one.

If you’re turning a hypothesis into a study design or a manuscript, our research consultancy team can review the question, variables, and analysis plan with you. When the paper is ready for submission, the journal publication service covers the route from manuscript to journal.