Match the test to the shape of your research question, not to what your friend used. “What is the level of X” needs only means and percentages. “Is there a significant relationship” needs Pearson’s r, Spearman’s rho or chi-square depending on your data type. “Is there a significant difference” needs a t-test or ANOVA. “Does X affect Y” needs regression.
How do you actually choose?
One question decides almost everything: what shape is my research question, and what kind of data did my instrument produce?
Nigerian undergraduate research questions come in five shapes, and each shape has a standard answer. Find yours in the left column.
| Your research question asks… | Data type | Test |
|---|---|---|
| What is the level / extent of X? | Likert items | Mean and SD against the criterion mean — no inferential test |
| What are the characteristics of respondents? | Categories | Frequency and percentage |
| Is there a significant relationship between X and Y? | Both scale / continuous | Pearson’s r |
| Is there a significant relationship between X and Y? | Both categorical | Chi-square test of independence |
| Is there a significant relationship between X and Y? | Ordinal, or not normally distributed | Spearman’s rho |
| Is there a significant difference between two groups? | Scale outcome | Independent-samples t-test |
| Is there a difference before and after? | Same respondents twice | Paired-samples t-test |
| Is there a significant difference among three or more groups? | Scale outcome | One-way ANOVA |
| Does X affect / predict Y? | Scale outcome | Simple or multiple linear regression |
Two things to notice. Most Nigerian undergraduate projects need only two or three of these, not all of them. And the first two rows are not tests at all — a great many research questions are answered completely by a mean table, which is why running a chi-square on them adds nothing.
Does every research question need a statistical test?
No, and this is the most useful thing on this page.
A research question of the form “What is the level of awareness of X among Y?” is descriptive. It is answered by a mean and a standard deviation for each item, compared against your criterion mean, with a decision column. There is no null hypothesis, nothing to reject, and no test statistic. Adding one does not make the project more rigorous; it makes it harder to defend, because you will be asked what the test was for.
Inferential tests belong to your hypotheses, not to your research questions. If your project has four research questions and two hypotheses, you likely need two inferential tests and four descriptive tables. Where each of those lands in the chapter is set out in the guide to writing Chapter Four.
Which test for a relationship?
All three of the common options answer “is there a relationship”, but they need different data.
- Pearson’s r when both variables are continuous or are scale scores built by averaging several Likert items. It gives a coefficient between −1 and +1, so it tells you direction and strength as well as significance.
- Spearman’s rho when your variables are ordinal, or when Pearson’s normality assumption fails. It is a tick-box in the same dialog as Pearson, so switching costs nothing.
- Chi-square test of independence when both variables are categories — sex against departmental choice, level against whether a respondent owns a laptop. Chi-square tells you whether an association exists, but not its direction or strength.
The most common error here is running chi-square on a mean. Chi-square works on counts in a contingency table. If your variable is a score, it is the wrong tool, and it is a question a panel asks precisely because it is easy to check.
Which test for a difference between groups?
Count the groups.
Two groups — male against female, urban against rural — is an independent-samples t-test. In SPSS the path is Analyze, then Compare Means and Proportions, then Independent-Samples T Test, and the dialog will not run until you click Define Groups and tell it which two codes are being compared.
Three or more groups — 300, 400 and 500 level; four departments — is a one-way ANOVA, under Analyze, then Compare Means, then One-Way ANOVA. Running three separate t-tests instead of one ANOVA is a real error rather than a stylistic one, because each additional comparison inflates the chance of a false positive.
The same respondents measured twice — before and after an intervention — is a paired-samples t-test, not an independent-samples one. Using the independent version on paired data throws away the pairing that makes the design worth running.
Why does everyone in my department use chi-square?
Because it is taught first, it is computable by hand, and it works on the categorical data that questionnaires produce most easily. For a genuinely categorical hypothesis it is the correct test.
It becomes a problem when it is applied by default. If your hypothesis concerns whether two scores are related, chi-square is not the tool, and forcing it means binning your scores into categories and discarding information you collected. If your department requires chi-square and your data is continuous, raise it with your supervisor early rather than at the defence, and be ready to say why the alternative fits better.
What assumptions do you have to check first?
Two, for most undergraduate designs, and both take one dialog box each.
Normality. In SPSS, Analyze, then Descriptive Statistics, then Explore; put your variable in the Dependent List, open Plots, and tick Normality plots with tests. That single checkbox produces a Kolmogorov-Smirnov test, a Shapiro-Wilk test, a normal Q-Q plot and a detrended Q-Q plot together.
Read the result carefully, because the logic runs backwards from what you are used to. The null hypothesis of a normality test is that the data came from a normal distribution, so p ≤ .05 means you reject normality. A significant result here is the inconvenient one. The two tests also carry separate p-values and can disagree, so say which one you used rather than reporting whichever was friendlier.
Homogeneity of variance. Levene’s test asks whether your groups have the same spread. One current detail matters: prior to SPSS Statistics version 31, Levene’s Test appeared automatically in the Independent-Samples T Test output; as of version 31 it is optional output that must be requested. If you were expecting a Levene’s row and cannot find one, that is why.
What if an assumption fails?
You substitute the test. You do not delete data and you do not proceed as though nothing happened.
| Planned test | Substitute when the assumption fails |
|---|---|
| Independent-samples t-test | Mann-Whitney U test |
| Paired-samples t-test | Wilcoxon signed-rank test |
| One-way ANOVA | Kruskal-Wallis test |
| Pearson’s r | Spearman’s rho |
If it is specifically the equal-variances assumption that failed on a t-test, you do not need to leave the parametric family at all: SPSS reports a Welch t-test — the unequal-variance version — on the second line of the same output block. Read that line and say that you did. For a one-way ANOVA with unequal variances, use a post hoc test that does not assume equal variances, such as Dunnett’s C.
None of these substitutions weakens your project. When assumptions hold, the parametric test is more powerful; when they do not, it is simply wrong, and the correct comparison is against a valid alternative rather than an invalid one.
What do you write in Chapter Three?
Name each tool and tie it to the specific question or hypothesis it answers. One sentence per tool is enough:
“Research questions one to three were answered using mean and standard deviation, with a criterion mean of 2.50. Hypothesis one was tested using the chi-square test of independence at the 0.05 level of significance, and hypothesis two using Pearson’s product-moment correlation coefficient. All analyses were carried out using IBM SPSS Statistics.”
Whatever appears there must be exactly what appears in Chapter Four — no extra tools, none quietly swapped. If a substitution happened because an assumption failed, amend Chapter Three to say so and why. “Why did you use this test?” is a standard question, and it is on the list in the guide to what is asked during a project defence. Where the tools sit in the overall structure is covered in how many chapters a final year project has, and the research questions they answer are framed back in Chapter One.
Keep the method and the results saying the same thing
Tesify holds your chapters in the structure your department expects, so the tool you name in Chapter Three and the table you build in Chapter Four cannot drift apart while you work. Over 9,000 students have used it to write more than 15,000 chapters, and every word is still written by you.
Frequently asked questions
What statistical test should I use for my project?
Match it to your research question. Level or extent questions need mean and standard deviation only. Relationship questions need Pearson’s r, Spearman’s rho or chi-square depending on data type. Difference questions need a t-test for two groups or ANOVA for three or more. Prediction questions need regression.
Do I need a statistical test for every research question?
No. Descriptive research questions are fully answered by means, standard deviations and percentages. Inferential tests belong to your hypotheses, and adding one where no hypothesis exists invites a question you cannot answer.
When do I use chi-square?
When both variables are categories and you are testing whether they are associated. It works on counts in a contingency table, so it is the wrong tool for a mean or a scale score.
What is the difference between Pearson and Spearman?
Pearson’s r is for continuous variables that meet the normality assumption; Spearman’s rho is the rank-based version for ordinal data or when normality fails. In SPSS both are tick-boxes in the same Bivariate dialog.
Can I use a t-test for three groups?
No. Use a one-way ANOVA. Running several t-tests across three groups inflates the chance of finding a difference that is not there, and a panel that spots it will ask about it.
How do I know if my data is normally distributed?
Run Analyze, Descriptive Statistics, Explore, then Plots, and tick “Normality plots with tests”. Read the Shapiro-Wilk and Kolmogorov-Smirnov p-values, remembering that a p-value at or below .05 means the data is not normal.
What do I do if my data is not normally distributed?
Substitute the test: Mann-Whitney U for an independent-samples t-test, Wilcoxon signed-rank for a paired t-test, Kruskal-Wallis for a one-way ANOVA, Spearman’s rho for Pearson’s r. Then record the substitution and its reason in Chapter Three.
Can I analyse a four-point Likert scale with a mean?
A single Likert item is strictly ordinal, and its proper measures are the median and mode. Most Nigerian departments nonetheless require a weighted mean per item against a criterion mean, and where your department requires it, you follow your department. Scale scores averaged across several items are more defensible as continuous.
What significance level should I use?
0.05 unless your department specifies otherwise, and state it explicitly in Chapter Three. Do not change it after seeing your results.
Do I need SPSS, or can I use Excel?
Excel handles means, standard deviations, percentages and simple correlation. Dedicated statistical software is more comfortable for ANOVA, non-parametric tests and normality checks. Name whichever you actually used in Chapter Three.
My supervisor says to use a test I do not understand. What do I do?
Ask which research question or hypothesis it answers, and write the answer down. If you cannot say what a test is for, you cannot defend it, and the question will come at the worst possible moment.
Can I change my test after seeing the results?
Only for a stated methodological reason such as a failed assumption, and you must record the change and its reason in Chapter Three. Changing tests until one returns significance is a different thing entirely, and it is visible to anyone who reads the chapter carefully.