Hypothesis testing is the statistical way to check whether your research claim is supported by the data, by comparing a null hypothesis (H0) with an alternative hypothesis (H1). For a thesis, you set a significance level (usually 0.05), pick a test that fits your data type, then read the p-value in the output to reach a decision.
- H0 and H1 are two complementary statements about the population
- The p-value answers how surprising your data would be if H0 were true
- Every decision is to reject or fail to reject H0, and both are valid for a thesis
- Your research questions and hypotheses from Chapter 1
- A clean dataset that is ready to analyse
- Statistical software (SPSS, JASP, or Jamovi)
Key numbers in thesis hypothesis testing
Understanding the logic of H0 and H1 in a thesis
Every thesis hypothesis test starts from two statements about the population. The null hypothesis (H0) says there is no difference or no relationship, for example that the mean scores of two groups are equal, or that motivation has no link to achievement. The alternative hypothesis (H1) says the opposite, that a difference or a relationship exists, which is what you proposed in Chapter 1. The logic mirrors the presumption of innocence in a courtroom. H0 is treated as true until the data provide strong enough evidence to doubt it. You never prove H1 with certainty. Instead you judge whether the data would be too strange if H0 were actually true. When the data are strange enough, you reject H0 and support H1. Many students mistakenly write their research hunch as H0. For a thesis, your claim almost always becomes H1, while H0 is the neutral statement you set out to challenge with data. Writing the H0 and H1 pair correctly at the start makes every interpretation in Chapter 4 far easier.
Steps to understand and run a hypothesis test
These six steps follow the flow supervisors usually expect, from stating the hypotheses through to reporting your results in Chapter 4.
- 1
State the null and alternative hypotheses
Translate your research question into a clear H0 and H1 pair. An example for a management thesis: H0 states that promotion has no effect on purchase decisions, while H1 states that promotion does affect purchase decisions. Write each hypothesis as a research sentence and as a statistical symbol, for example H0: r = 0 and H1: r is not equal to 0 for a correlation test. Make sure every hypothesis in Chapter 1 has a matching test in Chapter 3, so no claim is left hanging without a tool to test it.
Tips- One research question usually yields one hypothesis pair
- Use measurable verbs such as effect, difference, or relationship
- 2
Set the significance level and the test direction
The significance level (alpha) is the risk you accept of wrongly rejecting a true H0. Social science and economics theses commonly use alpha 0.05, while studies that demand high precision sometimes use 0.01. Decide the test direction too. A two-sided test is used when you only suspect a difference without naming its direction. A one-sided test is used when your theory names a specific direction, for example a new method is expected to produce higher results. Choose the direction in advance, before seeing the data, so the decision stays honest.
Tips- Alpha 0.05 means you accept a 5 percent chance of wrongly rejecting H0
- Unsure about direction? A two-sided test is safer and more often requested by examiners
Avoid changing alpha or the test direction after seeing the results to make the p-value look significant. This practice weakens the credibility of your thesis. - 3
Check the assumptions and choose the right test
Each test has requirements. Parametric tests such as the t-test and ANOVA assume interval or ratio data, a roughly normal distribution, and homogeneous variances. Check normality with the Kolmogorov-Smirnov or Shapiro-Wilk test, and check homogeneity with the Levene test. When assumptions are not met, move to an equivalent non-parametric test, for example Mann-Whitney to replace the two-group t-test, or Kruskal-Wallis to replace ANOVA. Matching the data type with the right test is the part examiners ask about most often during a defence.
Tips- Ordinal data from questionnaires often fit non-parametric methods better
- Record your assumption test results in the appendix as proof you checked them
- 4
Run the test and read the p-value with the test statistic
After the test runs in SPSS or JASP, the output shows a test statistic (such as a t, F, or chi-square value) and a p-value often labelled Sig. The p-value is the probability of getting data this extreme, or more extreme, if H0 were true. A small p-value signals that the data are hard to explain under H0. Read the degrees of freedom and the test statistic too, because examiners often ask you to report them fully, for example t(58) = 2.41 with p = 0.019.
Tips- The Sig. column in SPSS is the p-value you are looking for
- Report the p-value as it is, writing out the actual number in place of a table asterisk
- 5
Decide to reject or fail to reject H0
Compare the p-value with the alpha you set. When the p-value is smaller than alpha, you reject H0 and conclude the result is statistically significant. When the p-value equals or exceeds alpha, you fail to reject H0. Mind the wording. You do not accept H0, you simply lack enough evidence to reject it. A non-significant result is still a valid finding worth discussing, because it shows that the relationship you proposed has yet to appear in your data.
Tips- p < 0.05 means reject H0, p >= 0.05 means fail to reject H0
- A non-significant result is an honest piece of information for your thesis
- 6
Interpret in research language and report the effect size
A statistical decision needs to be translated back into your research question. Move beyond the sentence reject H0 and write its meaning, for example there is a significant effect of learning motivation on student achievement. Add an effect size such as Cohen d, eta squared, or a correlation coefficient, so readers know how large the effect is alongside whether an effect exists. The p-value shows the strength of evidence, while the effect size shows its practical magnitude. A mature thesis reports both together.
Tips- Statistically significant does not always mean practically large
- Cohen d near 0.2 is small, 0.5 is medium, and 0.8 is large as a rough guide
Types of hypothesis test and when to use them in a thesis
| Test | When to use | Sample thesis question | Key output |
|---|---|---|---|
| t-test | Compare the means of two groups | Is there a score gap between class A and class B | t value, Sig. |
| ANOVA | Compare the means of three or more groups | Do three service types differ in satisfaction | F value, Sig. |
| Chi-square | Test the link between two categorical variables | Is gender linked to major choice | chi-square value, Sig. |
| Correlation | Measure how closely two numeric variables move | How closely tied are motivation and achievement | r coefficient, Sig. |
| Regression | Predict an outcome from one or more predictors | How strongly does promotion drive sales | beta coefficient, Sig. |
Match the test to your data scale and the number of variables. Choosing the wrong test is the first issue examiners usually flag.
Key terms you must master
Null hypothesis (H0)
The neutral statement that no difference or relationship exists. It is the starting point you set out to challenge with data.
Alternative hypothesis (H1)
Your research claim that a difference or relationship exists. It is supported once H0 is successfully rejected.
Significance level (alpha)
The risk you accept of wrongly rejecting H0. The usual value is 0.05 in a quantitative thesis.
P-value
The probability of data this extreme if H0 were true. The smaller it is, the harder H0 is to keep.
Type I error
Rejecting an H0 that is actually true. Its size is controlled directly by the alpha you set.
Type II error
Failing to reject an H0 that is actually false. It shrinks as you raise the sample size and statistical power.
Parametric tests for a thesis: when they fit, when to be careful
- Interval or ratio data, such as test scores or sales figures
- The distribution is roughly normal and variances across groups are homogeneous
- The sample is large enough for stable, easy-to-interpret results
- You want to report a standard effect size such as the widely known Cohen d
- Ordinal data from a questionnaire scale whose intervals may not be equal
- A distribution far from normal even after a Shapiro-Wilk check
- A small sample that makes assumptions hard to satisfy convincingly
- Extreme outliers that pull the mean and blur the conclusion
“Students who understand hypothesis testing tend to stay calm in a defence, because they can explain why they chose a test, what their p-value means, and how large the effect they found is. That understanding can be built up step by step through practice on their own data.”
Checklist before writing Chapter 4
- Every research hypothesis has a clear H0 and H1 pair
- The significance level and test direction are set before analysing the data
- Normality and homogeneity assumption checks are run and documented
- The test type is chosen to fit the data scale and number of variables
- The p-value, test statistic, and degrees of freedom are reported in full
- The decision to reject or fail to reject H0 is worded correctly
- An effect size is reported to show the practical magnitude of the finding
- Hypothesis testing judges whether the data are strong enough to reject the null hypothesis, and stops short of proving a claim with certainty
- Set the significance level and test direction before seeing the data so the decision stays honest
- Match the test type to the data scale and check the assumptions before running it
- Report the p-value together with an effect size so the practical magnitude is also visible