A hypothesis is a testable claim about a defined population quantity, relationship or difference. Specify what would be measured and which parameter or comparison is being tested before choosing the statistical procedure. "Staff training is important" is too vague without a defined outcome and claim.
A statistical hypothesis needs precise variables, a parameter or comparison, and a design that can address it. This guide shows how to write one, how it connects to your objectives and your statistical test, and how to report the decision in Chapter 4 without bluffing. It sits between writing research questions and objectives and choosing the right statistical test, and it makes both easier.
A clear hypothesis should stay connected to the method and results throughout the report. UniDraft helps organise the objectives, analysis output and instructions you confirm into an editable project document. Start your project documentation with UniDraft, then check that the same variables and claims appear throughout. It does not choose or execute your statistical tests.
The short answer
- Write hypotheses only if your project is the kind that tests something and your department expects them. Otherwise use research questions.
- Derive each hypothesis from one objective or research question, and name the same variables.
- State the null hypothesis (H0): no significant relationship, difference or effect. State the alternative (H1) if your department requires it.
- Decide the test, the significance level and the population before you collect data.
- In Chapter 4, report the test, the p-value and the decision. In Chapter 5, interpret it honestly.
What a hypothesis is, and what it is not
Statistical sources describe a hypothesis test as a rule for deciding whether to accept or reject a claim about a population using the evidence in a sample. Minitab's support documentation puts it this way: the null hypothesis is the statement being tested, usually "no effect" or "no difference", and the alternative is what you aim to show using the sample. See Minitab's explanation of a hypothesis test.
So a hypothesis is not:
- A prediction of the whole study. "This study will show that X is good" is a hope, not a testable claim.
- An undefined assertion. "Most students use phones in class" needs a defined population, period and use measure. A descriptive proportion can be tested when a precise population claim and suitable design are specified; descriptive work is not automatically excluded from hypothesis testing.
- A predetermined conclusion. Do not declare that respondents agree before collecting evidence. A properly defined claim about a response proportion can be tested; deciding the answer in advance is the problem.
- A research objective. The objective says what you want to do; the hypothesis states what you will test.
Do you need hypotheses?
Not every project does. Think about the type of study you are running.
- Quantitative studies testing relationships, differences or effects. Hypotheses are usual. Examples: effect of a teaching method on scores, relationship between inflation and stock returns, difference in yield between two fertiliser rates.
- Descriptive surveys. Studies that describe opinions, practices or levels often use research questions only. Adding forced hypotheses can feel artificial.
- Qualitative studies. Exploring experiences or meanings uses research questions. Hypotheses are not normally part of the design.
- Mixed designs. You may use questions for the qualitative strand and hypotheses for the quantitative strand.
Your department's handbook or your supervisor decides. The Chapter 1 guide shows where hypotheses sit among the other sections if your template includes them.
The chain: objective, hypothesis, test, decision
For objectives that involve a statistical test, connect the question, defined parameter, null and alternative, data, suitable procedure and reporting decision. Not every objective needs a hypothesis, and a hypothesis alone does not determine the test without the design and assumptions.
Here is the chain in practice, for a made-up project, with illustrative wording only:
- Objective. To examine the association between defined staff-training and productivity measures in selected firms.
- Research question. What association exists between the specified training and productivity measures?
- Null hypothesis (H0). The population correlation between the specified training and productivity measures is zero.
- Alternative hypothesis (H1). The population correlation between the specified training and productivity measures is not zero.
- Test. A correlation procedure suitable for the measures, sampling design and assumptions, agreed before analysis. Correlation alone does not establish a causal effect.
- Decision. Reject H0 if the p-value is below the stated significance level, otherwise fail to reject it.
If any link is missing, the chain breaks, which is exactly where students lose marks. A hypothesis that uses "productivity" while the questionnaire measured "job satisfaction" has no data to test it. The questionnaire design guide helps you make sure every item serves a variable that a hypothesis needs.
How to write a good hypothesis
Use this checklist for each hypothesis.
- Name the variables. Say what you are measuring. "Staff training" and "employee productivity" are variables; "performance generally" is vague.
- State the relationship or difference. Relationship, effect, difference, association.
- Name the population or setting when it matters. "Among nurses in the teaching hospital" narrows the claim.
- Make it testable with your data. You must be able to calculate something from your questionnaire, measurements or records.
- Use neutral, precise language. Avoid "very", "greatly" and "obviously".
- One hypothesis, one claim. Do not bundle several relationships into one sentence.
- Keep wording consistent. The same variable names in Chapter 1, Chapter 3 and Chapter 4.
Null and alternative
For a simple two-sided correlation test, the null can specify a population correlation of zero and the alternative a nonzero correlation. For a two-group mean comparison, they can specify a zero versus nonzero population mean difference. These are claims about parameters; statistical significance describes the later test decision, not the underlying parameter itself. Follow required local wording while keeping that distinction clear.
- H0: The specified population correlation between X and Y equals zero.
- H1: The specified population correlation between X and Y does not equal zero.
Directional and non-directional
A non-directional alternative says only that there is a difference or relationship. A directional alternative says in which direction, for example that one group scores higher or that a relationship is positive.
Use a directional hypothesis only if theory or earlier research gives a reason, and decide before you see your data. Using a direction because the results happened to point that way is a quiet form of cheating. When in doubt, use the non-directional form.
Examples by field
These examples show the form. Do not copy them; adapt the variables to your own study, population and objectives.
Education.
- H0: There is no significant difference in the mean scores of students taught with the activity-based method and those taught with the lecture method.
- H1: There is a significant difference in the mean scores of students taught with the activity-based method and those taught with the lecture method.
Accounting and finance.
- H0: There is no significant relationship between firm size and the quality of financial reporting among listed firms.
- H1: There is a significant relationship between firm size and the quality of financial reporting among listed firms.
Nursing and public health.
- H0: There is no significant association between knowledge of hand hygiene and compliance among nurses in the selected hospital.
- H1: There is a significant association between knowledge of hand hygiene and compliance among nurses in the selected hospital.
Agriculture.
- H0: There is no significant difference in the mean yield of maize across the fertiliser rates used.
- H1: There is a significant difference in the mean yield of maize across at least two of the fertiliser rates used.
Mass communication.
- H0: There is no significant relationship between exposure to a radio campaign and awareness of the campaign message among residents of the selected community.
- H1: There is a significant relationship between exposure to a radio campaign and awareness of the campaign message among residents of the selected community.
Computer science and engineering.
- H0: There is no significant difference in response time between the proposed system and the existing system.
- H1: There is a significant difference in response time between the proposed system and the existing system.
Each pairs variables that can be measured and compared with a recognisable test.
Matching the hypothesis to the test
The test follows from what you are comparing and the type of data. The detailed logic is in the guide on which statistical test to use, but the principle is simple:
- Difference between two groups. Usually a test comparing two means or two distributions.
- Difference among three or more groups. A test comparing several means.
- Relationship between two variables. A correlation or association test.
- Modelled association between predictors and an outcome. A suitable regression may apply. Its coefficients are not automatically causal effects.
- Association between categories. A test suited to counts in categories.
Your sample must be adequate for the test, which is why the sample size guide belongs in the same planning session. Instruments should be reliable and valid for the variables, as explained in the guide on validity and reliability of a questionnaire.
The significance level and the decision
Before analysis, state the significance level (alpha), commonly 0.05. Specify the decision rule in advance, including how equality to alpha is handled. Use the unrounded p-value for the decision; a displayed rounded value at the boundary may be insufficient. Minitab's explanation adds the important caution: failing to reject does not prove the null is true, it means you lack sufficient evidence against it, and that asymmetry is why statisticians say "fail to reject" instead of "accept".
Three habits keep you honest:
- Set alpha in advance and state it in Chapter 3.
- Do not run many tests until one is significant. Test the hypotheses you planned.
- Report the actual p-value or the test statistic and degrees of freedom as your department's style requires, not only "significant".
Writing the hypotheses into each chapter
Chapter 1. List the hypotheses after the objectives and research questions, in the same numbering, if your template includes them.
Chapter 2. The literature review should explain why you expect the relationship you test. If a directional hypothesis is used, justify its direction before analysis. Label later exploratory changes honestly. Linking the hypotheses to a conceptual or theoretical framework helps here.
Chapter 3. State the test for each hypothesis, the significance level and the software. The methodology guide shows how to describe it.
Chapter 4. Report each test with the table or output, the p-value and the decision. The results chapter guide explains how to present tables without dumping raw output.
Chapter 5. Interpret the decision and compare with earlier studies. Explain unexpected results honestly. See the discussion and conclusion guide.
A reporting sentence in Chapter 4 can look like this, with placeholders to be replaced by your own values:
"For Hypothesis 1, the test returned a p-value of [value], which is [below / above] the 0.05 level of significance. The null hypothesis is therefore [rejected / not rejected]: there is [sufficient / insufficient] evidence of a significant [relationship / difference / effect] between [variable A] and [variable B]."
Write only the numbers your analysis produced.
Common mistakes supervisors flag
- Untestable statements. "Training is important" cannot be rejected by any data.
- Hypotheses with no matching evidence. Variables in the hypothesis that the questionnaire never measured.
- Mismatch with the test. A relationship hypothesis analysed with a test of group differences.
- Too many hypotheses. Ten hypotheses for a small study, each tested thinly.
- Hypotheses written after the results. Presenting post-hoc changes as prespecified.
- Confusing the null and alternative. Stating the null as a positive claim.
- "Accept the alternative" without evidence. Reporting significance without the p-value.
- Inconsistent numbering and wording across chapters.
- Copying hypotheses from a past project with the variables swapped. A panel can ask how you chose the variables and what they mean in your context.
- Formatting slips. The free formatting checker reads a .docx and flags mismatches with your institution's rules.
Defence questions to prepare for
Use these original practice questions to check whether you can explain your design; they are not a prediction of a panel’s questions. Prepare honest answers to questions such as:
- What is your null hypothesis and why did you state it that way?
- Why did you choose this test for this hypothesis?
- What does a p-value of this size tell us, and what does it not tell us?
- What does "fail to reject" mean, and why not "accept"?
- What would a different result have meant for your conclusion?
- How did your sample size affect the test?
- What are the limits of this result?
The guide to common project defence questions shows how to structure answers.
A drafting routine
- List your objectives. For each, ask whether it involves a relationship, difference or effect.
- Write the null hypothesis for those that do, naming the variables exactly.
- Write the alternative if your department asks.
- Check the data. Which questionnaire items or measurements will feed each hypothesis?
- Choose the test and the significance level and write them into Chapter 3.
- Show your supervisor before you collect data. If your supervisor sends changes, managing final year project supervisor corrections explains how to track them.
Pre-submission checklist
- Each hypothesis maps to one objective or research question.
- Variables are named identically in Chapters 1, 3 and 4.
- Every hypothesis can be tested with data you actually collected.
- The test, significance level and software are stated in Chapter 3.
- Chapter 4 reports each decision with its p-value or equivalent.
- Chapter 5 interprets the decisions honestly, including non-significant ones.
- Prespecified hypotheses and any later exploratory changes are clearly distinguished.
- Numbering, wording and formatting match your department's guideline.
Where UniDraft fits in
UniDraft builds the written project as an editable Word (.docx) document from the objectives, data and results you confirm, applying your supervisor's instructions, your department's format and your university's format, in that order of priority. It does not invent data, test results, citations or approvals, and it does not produce slides. It cannot choose your hypotheses or run your analysis for you; you supply the real analysis output, and your supervisor decides what the department expects.
Sources
- Minitab support: What is a hypothesis test?: null and alternative hypotheses, significance level, p-value and the meaning of failing to reject.