Which Statistical Test Should I Use? A Simple Guide for Your Project

Almost every student gets to Chapter 4 and asks the same question: which statistical test should I use?

It feels like there are hundreds of tests to choose from. But for most undergraduate and Masters projects, you only need five: chi-square, the t-test, ANOVA, correlation and regression. And choosing between them comes down to just two things:

1. What your research question is asking, and
2. The type of data you have.

Get those two right and the test almost picks itself. Let’s walk through it.

Short on time? Use the free Which test do I need? tool on the InsightsHub homepage. Answer two or three questions and it tells you the test.

Step 1: Know your type of data

Every variable in your study is one of two broad types.

Categorical data puts people into groups or labels. Gender, religion, level of study, marital status and “yes/no” answers are all categorical.

Numerical data is a measurement you can take an average of. Age, CGPA, income, test scores and a total score from a scale are numerical.

Here’s the trap that catches many students: coding a variable as numbers in SPSS doesn’t make it numerical. If you code male as 1 and female as 2, gender is still categorical. The numbers are just labels.

A quick test: can you sensibly take the average of it? The average age of your respondents makes sense. The average religion doesn’t. So age is numerical and religion is categorical.

Step 2: Know what your question is asking

Most research questions and hypotheses ask one of four things:

• Are two things linked? (“Is education associated with knowledge of food safety?”)
• Do groups differ? (“Do male and female students differ in exam anxiety?”)
• Does one thing predict another? (“Does study time predict CGPA?”)
• What does my data look like? (describing your respondents and their answers)

Now combine Step 1 and Step 2.

The five tests, in plain English

Test	Use it when	Example research question
Chi-square	Both variables are categorical and you want to know if they’re linked	Is educational qualification associated with knowledge of food-borne infection?
Independent t-test	You compare a numerical outcome between two groups	Do female and male undergraduates differ in computer-based test anxiety?
One-way ANOVA	You compare a numerical outcome across three or more groups	Does academic stress differ across 100, 200, 300 and 400 level?
Pearson correlation	Both variables are numerical and you want to know if they move together	Is hours spent on social media related to CGPA?
Linear regression	You want to know if one or more variables predict a numerical outcome	Do study hours and class attendance predict CGPA?

Chi-square test of independence

Use chi-square when both variables are categorical. It tells you whether the two are associated, for example whether knowledge level (good or poor) depends on educational qualification.

In SPSS: Analyze → Descriptive Statistics → Crosstabs. Put one variable in Rows and the other in Columns, click Statistics and tick Chi-square.

Check this: look at the footnote under the Chi-Square Tests table. If more than 20% of cells have an expected count below 5, the chi-square result isn’t reliable. Use Fisher’s exact test instead, and say so in your write-up. Supervisors notice this.

Independent samples t-test

Use a t-test when your outcome is numerical and you’re comparing two separate groups, like male and female, or public and private school students.

In SPSS: Analyze → Compare Means → Independent-Samples T Test. Your numerical outcome goes in Test Variable and your grouping variable in Grouping Variable (click Define Groups and enter the codes, e.g. 1 and 2).

If the same people were measured twice, say before and after a training, use the paired samples t-test instead.

One-way ANOVA

ANOVA is the t-test’s big brother. Use it when you compare a numerical outcome across three or more groups, like levels of study or age bands.

In SPSS: Analyze → Compare Means → One-Way ANOVA. If the result is significant, ANOVA only tells you that some groups differ, not which ones. Click Post Hoc and tick Tukey to see exactly where the differences are.

Pearson correlation

Use correlation when both variables are numerical and you want to know whether they move together. The correlation coefficient, r, runs from −1 to +1:

• close to +1: as one goes up, the other goes up
• close to −1: as one goes up, the other goes down
• close to 0: no linear relationship

In SPSS: Analyze → Correlate → Bivariate, and tick Pearson.

Remember: correlation is not causation. A link between social media hours and CGPA does not prove that social media causes lower grades.

Linear regression

Use regression when you want to know whether one or more variables predict a numerical outcome, and by how much. With more than one predictor, it’s called multiple regression.

In SPSS: Analyze → Regression → Linear. Your outcome goes in Dependent and your predictors in Independent(s).

Report the R² (how much of the outcome your predictors explain), the overall F test, and the coefficient and p-value for each predictor.

If your outcome is yes/no (passed or failed, adopted or not), use binary logistic regression instead.

When your data isn’t “normal”

The t-test, ANOVA and Pearson correlation assume your numerical data is roughly normally distributed. If it’s heavily skewed, or your sample is very small, use the non-parametric version:

Instead of…	Use…
Independent t-test	Mann-Whitney U test
Paired t-test	Wilcoxon signed-rank test
One-way ANOVA	Kruskal-Wallis test
Pearson correlation	Spearman’s rho

A note on Likert scales

Single Likert items (Strongly Agree to Strongly Disagree) are ordinal: they have an order, but the gaps between answers aren’t guaranteed to be equal. When you add several items into a total or average score for a construct like “study habits”, that score is usually treated as numerical. That’s why you’ll often see t-tests and correlations run on scale scores.

Whatever you do, check whether any items are negatively worded. Those need to be reverse-coded before you add them up, or your scale’s reliability (Cronbach’s alpha) will collapse.

How to report your result

Don’t paste the SPSS table and leave it. Every result needs three parts: the test, the numbers, and what it means.

An independent samples t-test showed no significant difference in CGPA between male and female students, t(241) = 1.24, p = .217.

There was a significant association between level of study and use of AI writing tools, χ²(3, N = 180) = 10.28, p = .016.

Then explain what it means for your study in one or two plain sentences. That’s the part your supervisor and your defence panel care about most.

Quick recap

1. Decide whether each variable is categorical or numerical.
2. Decide whether your question asks about a link, a difference or a prediction.
3. Two categorical variables → chi-square. Numerical outcome, two groups → t-test. Three or more groups → ANOVA. Two numerical variables → correlation. Predicting an outcome → regression.
4. Check the assumptions, and switch to the non-parametric version if they fail.
5. Report the test, the numbers, and what they mean.