A hypothesis is a proposed explanation or statement about a relationship or difference that can be evaluated against evidence. AnswerThis can help you compare candidate null and alternative hypotheses, with suggested variables, populations and measurements, before you check their rationale and fit with your study design.
A good hypothesis identifies what you expect and makes clear what observations could count against it. For example, “Among undergraduates, weekly study hours are associated with examination scores” becomes testable once you define the population, measurements and association of interest. The hypothesis examples below show how those choices connect to corresponding null statements.
TL;DR
Start with a research question and a reasoned expectation, then define the variables and measurements.
Match each null to the alternative’s population, outcome, comparison and target parameter.
Association does not establish causation, and directional expectations require justification before examining results.
Failure to reject a null does not establish that it is true or that an effect is absent.
What makes a good hypothesis?
A useful hypothesis follows from a question and reasoned basis, not an unsupported guess. It identifies variables and units, specifies a relationship or difference, and fits feasible measurements and design.
Testability means identifying evidence that could evaluate the statement. Falsifiability means allowing evidence to count against it. One unexpected score does not necessarily settle a probabilistic hypothesis about average performance: sampling variation, measurement error and assumptions matter.
Distinguish the substantive research hypothesis from its statistical formulation. “Retrieval practice supports later recall” is a substantive expectation. “The population mean recall score differs between the assigned practice conditions” specifies a statistical target. Testing that contrast does not, by itself, identify the psychological mechanism behind a difference.
Exploratory qualitative research can investigate experiences without predicting them. Interpretive work can develop arguments through textual analysis. Neither automatically requires a formal hypothesis.
AnswerThis’s free hypothesis generator drafts three candidate pairs of null and alternative hypotheses, with variables, a population and suggested measurements. Use its candidates to compare possible formulations during drafting; it does not search papers.
Hypothesis vs prediction vs research question
A question identifies the inquiry, a hypothesis proposes an answer, and a prediction specifies an expected observation. This hypothetical example keeps the topic constant.
Research question
Purpose: Defines the inquiry.
Example: Does retrieval practice rather than rereading influence delayed vocabulary recall among adult language learners?
Hypothesis
Purpose: Proposes a relationship or explanation.
Example: Practicing retrieval rather than rereading improves delayed vocabulary recall among adult language learners.
Prediction
Purpose: Specifies an expected observation under stated conditions.
Example: In a randomized comparison using the same vocabulary, learners assigned retrieval practice will have a higher mean recall score one week after practice than those assigned rereading.
A hypothesis can generate several predictions for different tasks or conditions; justify them rather than selecting afterward. “If–then” wording is optional. Here, the prediction specifies a comparison and measurement time.
Types of hypotheses
These classifications overlap. One hypothesis can be alternative, directional, simple and associative. They help specify what the study intends to evaluate.
Null and alternative
Meaning: Competing statistical statements about a population quantity.
Example: Null: the mean score difference is zero. Alternative: it is not zero.
Directional and non-directional
Meaning: A direction is specified, or departures in either direction are included.
Example: Directional: mean recall is higher under retrieval. Non-directional: mean recall differs between conditions.
Simple and complex
Meaning: Here, simple relates one predictor to one outcome; complex involves multiple predictors or outcomes.
Example: Simple: study hours are associated with examination score. Complex: at least one of study hours and sleep duration is associated with examination score in the specified multiple-regression model.
Associative and causal
Meaning: Co-variation is proposed, or an intervention or exposure is proposed to change an outcome.
Example: Associative: study hours correlate with score. Causal: assigning retrieval rather than rereading changes recall in a randomized experiment.
For the complex example, the joint null sets both predictor coefficients to zero in the specified model. This teaching use of simple and complex differs from the statistical terms simple and composite.
Causal wording requires suitable design and assumptions. Randomization helps, but missing outcomes, adherence and analysis still matter. Observational association alone establishes neither causal direction nor the absence of common causes.
Justify direction before examining results. A directional expectation does not automatically require a one-sided test: opposite effects and the inferential objective also matter. Prespecify sidedness rather than choosing it to suit results.
How to write a hypothesis in five steps
1. Start with a focused research question
Consider: “Among adult language learners, does retrieval practice rather than rereading change vocabulary recall one week later?” This identifies a comparison and outcome without deciding the answer. It focuses on recall rather than all aspects of language learning.
AnswerThis’s research question generator drafts candidates from a topic, method and optional population. Choose one that fits accessible evidence.
2. Review relevant evidence and develop a rationale
Read relevant theory and studies. Check whether their tasks, populations and follow-up justify your expectation. Record contrary evidence and uncertainty.
Here, a higher-recall expectation is a candidate to justify, not an established conclusion supplied by the example. If evidence does not support a direction, a non-directional hypothesis may better represent your rationale.
3. Define variables, population and measurements
Specify learner eligibility, vocabulary material and practice conditions. Define recall as a score on a stated assessment one week after the last practice session. Decide whether learners, classes or another unit receive the assignment.
Plan a randomized comparison. Equal practice time and comparable material do not substitute for randomization. Document outcome measurement procedures.
4. State the expected relationship or difference
A directional research hypothesis is: “Among adult language learners, assignment to retrieval practice rather than rereading produces a higher mean vocabulary recall score one week after practice.” A two-sided formulation would instead state that assignment changes mean recall without specifying direction.
Adapt these patterns to your inquiry:
Associative
Drafting template: Among a defined population, a specified predictor is associated with a specified outcome.
Comparative
Drafting template: Among a defined population, a specified outcome differs between group A and group B.
Directional comparative
Drafting template: Among a defined population, group A has a higher or lower specified outcome than group B.
Causal
Drafting template: Among a defined population, assigning an intervention rather than a comparison changes an outcome over a specified period.
Templates guide wording; they do not establish a rationale, suitable design or valid measurements. Replace general concepts with meaningful definitions rather than adding irrelevant detail.
5. Formulate the corresponding null and check design alignment
For the directional recall formulation, the alternative states that the population mean score under assignment to retrieval is greater than under assignment to rereading. Its complementary null states that the retrieval mean is less than or equal to the rereading mean. Both refer to the same learners, outcome and follow-up.
For a two-sided comparison, the alternative says the mean difference is not zero; the null says it is zero. Check alignment with assignment, outcome collection and analysis.
30 hypothesis examples by field
All examples are original illustrations requiring adaptation to actual evidence, measurements and design, not findings or claims of novelty. The scope in each row’s first column applies to both statements.
These null hypothesis examples target population quantities, not exact equality in a sample. A mean is an average, and a probability is the chance of a defined event. For the linear-association examples, the statistical target is the population Pearson correlation between the variables named in that row; “that population Pearson correlation” in a null refers to this quantity. A nonzero value indicates linear association, while zero does not exclude nonlinear relationships. The tables use words rather than statistical symbols.
Most pairs are two-sided. Directional pairs use complementary nulls: “higher” versus “equal or lower.” Introductory materials often show only the equality boundary used to calibrate the test, omitting the opposite direction included in the full null.
Education
Learning hypotheses need to distinguish study behaviors, measured outcomes and assigned teaching conditions.
1. Undergraduates: study hours and final examination score
Research or alternative hypothesis: Weekly study hours have a linear association with final examination scores among undergraduates.
Corresponding null hypothesis: The population Pearson correlation between weekly study hours and final examination score is zero.
2. Adult language learners randomized to retrieval or rereading
Research or alternative hypothesis: Assignment to retrieval rather than rereading changes mean vocabulary recall one week after practice.
Corresponding null hypothesis: Assignment to retrieval rather than rereading has zero average effect on vocabulary recall one week after practice.
3. First-year commuters and campus residents: belonging
Research or alternative hypothesis: Mean academic-belonging score is lower among commuters than campus residents.
Corresponding null hypothesis: Mean academic-belonging score is equal or higher among commuters than campus residents.
4. Secondary-school classes: size and participation
Research or alternative hypothesis: Class size has a linear association with mean contributions per pupil per lesson in secondary-school classes.
Corresponding null hypothesis: That population Pearson correlation is zero.
5. Master’s students receiving written or audio feedback
Research or alternative hypothesis: Mean revision-confidence score differs between feedback groups after receipt of feedback.
Corresponding null hypothesis: Mean revision-confidence score is equal between feedback groups after receipt of feedback.
Psychology
These examples separate associations between experiences from effects tested through experimental assignment.
6. Doctoral students: bedtime phone use and sleep onset
Research or alternative hypothesis: Bedtime phone-use minutes have a linear association with time taken to fall asleep among doctoral students.
Corresponding null hypothesis: That population Pearson correlation is zero.
7. Adults randomized to background speech or silence
Research or alternative hypothesis: Assignment to background speech rather than silence changes mean sustained-attention accuracy during the task.
Corresponding null hypothesis: Assignment to background speech rather than silence has zero average effect on sustained-attention accuracy during the task.
8. Adults tested on recognition of familiar and unfamiliar faces
Research or alternative hypothesis: Mean recognition accuracy is higher for familiar than unfamiliar faces.
Corresponding null hypothesis: Mean recognition accuracy is equal or lower for familiar than unfamiliar faces.
9. Postgraduates: social support and loneliness scores
Research or alternative hypothesis: Perceived social-support scores have a linear association with loneliness scores among postgraduates.
Corresponding null hypothesis: That population Pearson correlation is zero.
10. Adult participants assigned different questionnaire orders
Research or alternative hypothesis: Mean reported stress score differs between questionnaire-order groups at session completion.
Corresponding null hypothesis: Mean reported stress score is equal between questionnaire-order groups at session completion.
Public health and healthcare
Healthcare hypotheses often concern event probabilities, such as missed appointments, alongside continuous outcomes such as comprehension scores.
11. Rural adults: travel time and unmet appointment needs
Research or alternative hypothesis: Rural adults with longer and shorter travel times differ in their probability of reporting an unmet appointment need during the survey month.
Corresponding null hypothesis: That probability is equal between the same travel-time groups during the survey month.
12. Outpatients randomized to text reminders or usual scheduling
Research or alternative hypothesis: Reminder assignment changes the probability of missing the next scheduled appointment compared with usual scheduling.
Corresponding null hypothesis: Reminder assignment has zero effect on that probability compared with usual scheduling.
13. Adults receiving illustrated or text-only medication instructions
Research or alternative hypothesis: Mean comprehension score is higher in the illustrated-instruction group than the text-only group immediately after reading.
Corresponding null hypothesis: Mean comprehension score is equal or lower in the illustrated-instruction group than the text-only group immediately after reading.
14. Clinic attendees: waiting time and satisfaction
Research or alternative hypothesis: Waiting time has a linear association with visit-satisfaction scores among clinic attendees.
Corresponding null hypothesis: That population Pearson correlation is zero.
15. Weekday and weekend hospital discharges
Research or alternative hypothesis: The probability that a discharge summary documents medication changes differs between weekday and weekend discharges.
Corresponding null hypothesis: That documentation probability is equal between weekday and weekend discharges.
Business and management
Business hypotheses become clearer when broad aims such as retention are tied to defined employee or customer outcomes.
16. Graduate recruits: support and intention to leave
Research or alternative hypothesis: Supervisor-support scores have a linear association with intention-to-leave scores among graduate recruits.
Corresponding null hypothesis: That population Pearson correlation is zero.
17. Online shoppers randomized to discount or no discount
Research or alternative hypothesis: Offering a checkout discount changes the probability of completing that shopping session’s purchase compared with no discount.
Corresponding null hypothesis: Offering the discount has zero effect on that probability compared with no discount.
18. Remote and office-based support employees
Research or alternative hypothesis: Mean weekly self-reported task completion is higher among remote than office-based employees.
Corresponding null hypothesis: Mean weekly self-reported task completion is equal or lower among remote than office-based employees.
19. Customer accounts using monthly or annual billing
Research or alternative hypothesis: Mean service-satisfaction score differs between billing groups at renewal.
Corresponding null hypothesis: Mean service-satisfaction score is equal between billing groups at renewal.
20. Retail branches: training hours and complaint rate
Research or alternative hypothesis: Training hours per employee have a linear association with complaints per transaction across retail branches during the quarter.
Corresponding null hypothesis: That population Pearson correlation is zero.
Environmental science
Environmental hypotheses need clear sampling periods and conditions so site comparisons are not mistaken for experimental effects.
21. Urban survey sites: impervious cover and bird richness
Research or alternative hypothesis: Impervious-cover percentage has a linear association with recorded bird-species richness across urban survey sites.
Corresponding null hypothesis: That population Pearson correlation is zero.
22. Soil plots randomized to fertilizer or no fertilizer
Research or alternative hypothesis: Fertilizer assignment changes mean runoff nitrate concentration after a standardized rainfall simulation compared with no fertilizer.
Corresponding null hypothesis: Fertilizer assignment has zero average effect on that concentration compared with no fertilizer.
23. Shaded and unshaded pavements of matching material
Research or alternative hypothesis: Mean afternoon surface temperature is lower on shaded than unshaded pavements over the survey days.
Corresponding null hypothesis: Mean afternoon surface temperature is equal or higher on shaded than unshaded pavements over the survey days.
24. Upstream and downstream monitoring sites
Research or alternative hypothesis: Mean dissolved-oxygen concentration differs between site groups during summer sampling.
Corresponding null hypothesis: Mean dissolved-oxygen concentration is equal between site groups during summer sampling.
25. Coastal plots: vegetation cover and sediment loss
Research or alternative hypothesis: Vegetation cover has a linear association with sediment mass lost from coastal plots over the monitoring period.
Corresponding null hypothesis: That population Pearson correlation is zero.
Computer science and engineering
Technical comparisons require specified workloads or test conditions and a clear performance measure.
26. Support-ticket documents processed by classifiers A and B
Research or alternative hypothesis: Classifiers A and B differ in average per-document classification error on support-ticket documents.
Corresponding null hypothesis: The population mean difference in per-document classification error between A and B is zero.
27. Heat sinks randomized to wider or narrower fin spacing
Research or alternative hypothesis: Wider-spacing assignment changes mean steady-state temperature compared with narrower spacing under fixed heat input and airflow.
Corresponding null hypothesis: Wider-spacing assignment has zero average effect on that temperature compared with narrower spacing under those conditions.
28. Comparable server runs under cache configurations A and B
Research or alternative hypothesis: Mean request latency is lower under A than B for the specified workload.
Corresponding null hypothesis: Mean request latency is equal or higher under A than B for that workload.
29. Manufactured specimens: roughness and fatigue life
Research or alternative hypothesis: Surface roughness has a linear association with cycles to failure among manufactured specimens under the specified loading protocol.
Corresponding null hypothesis: That population Pearson correlation is zero.
30. Devices randomized to cooling designs A and B
Research or alternative hypothesis: Assignment to A rather than B changes at least one of two population means: steady-state temperature and fan power at fixed load.
Corresponding null hypothesis: Assignment to A rather than B has zero average effect on both steady-state temperature and fan power at fixed load.
Example 30 has two outcomes. Its alternative concerns at least one effect; its null requires both effects to be zero. Plan a joint or multiplicity-aware analysis.
Weak vs strong hypothesis examples
The improvement should change what evidence can evaluate, not merely lengthen the sentence.
Vague variables
Weak formulation: Studying more improves academic performance.
Stronger formulation and improvement: Among undergraduates, weekly study hours are associated with final examination score. This defines study amount and performance while avoiding an unsupported causal claim.
Undefined comparison or population
Weak formulation: Audio feedback is better.
Stronger formulation and improvement: Among master’s students, mean revision-confidence score differs after audio versus written feedback. This identifies whose outcome and which comparison without assuming superiority.
Unsupported causation
Weak formulation: Long commutes cause poor wellbeing.
Stronger formulation and improvement: Among employees in an observational survey, commute duration is associated with wellbeing score. This retains the topic while limiting the claim to the design.
Disconnected claims
Weak formulation: Flexible work improves retention and makes customers loyal.
Stronger formulation and improvement: Among employees, perceived schedule flexibility is associated with intention to remain. This selects one employee-level inquiry rather than combining unrelated units and outcomes.
Specify the association parameter and measurement plan rather than searching unspecified analyses for support.
Check your hypothesis before collecting data
Connect the hypothesis to a primary outcome, measurement time and planned analysis. Define the population quantity and independent unit, including treatment of repeated measurements or clustering. Prespecify confirmatory hypotheses and analysis choices before examining relevant results; label later exploratory analyses transparently so they can inform future work.
Sample-size planning follows the design, a justified target effect or desired precision, variability or event rates, and other assumptions. A well-written hypothesis does not supply those inputs. The guide to how to calculate sample size explains why assumptions must match the planned comparison.
The NIST guidance on hypothesis tests distinguishes population claims from sample estimates and statistical from practical significance. A p-value threshold is a decision rule within an analysis, not a measure of effect magnitude. Examine estimates, uncertainty and scientific context alongside the test.
The American Statistical Association’s statement on p-values also warns that a p-value does not give the probability that a hypothesis is true. A nonsignificant result can reflect limited information, a small effect or other features of the design and model; it does not demonstrate zero effect. Establishing equivalence requires a separate, appropriately planned analysis.
FAQs
What makes the retrieval-practice hypothesis testable?
It specifies adult language learners, retrieval practice versus rereading, and mean vocabulary recall one week later. Those details define what to compare and measure; the study still needs a reasoned rationale and valid measurements.
What null matches a hypothesis that retrieval practice produces higher recall?
If the alternative states that mean recall is higher with retrieval practice, the full null states that it is lower than or equal to recall with rereading. Keep the population, recall measure and one-week timing identical in both statements.
Does a directional hypothesis require a one-sided test?
No. A directional expectation does not automatically justify a one-sided test. Choose and justify the test’s sidedness before seeing the results, considering the research objective and whether an effect in the opposite direction matters.
How can AnswerThis help me draft matching null and alternative hypotheses?
The AnswerThis hypothesis generator drafts three candidate null–alternative pairs with variables, a population and suggested measurements. Use them to compare formulations, then check the evidence, measurements and study design yourself. The generator does not search papers.
Does failing to reject the null show that retrieval practice and rereading are equivalent?
No. A nonsignificant result does not establish equivalence or prove the null true. Examine effect estimates and uncertainty; demonstrating equivalence requires prespecified equivalence margins and an appropriate analysis.
Draft your hypothesis, name the measurements and comparison it requires, and write its matching null. Then check whether your proposed design can evaluate that pair before collecting the evidence.
