Common mistakes

The 5 Thesis Statistics Mistakes That Cost Students Their Grade (And How to Catch Them Before Your Defense)

Q: Is it acceptable to use a t-test on Likert scale data?

It depends on the data. Strictly speaking, single Likert items are ordinal - use Mann-Whitney U. Composite scores averaged across 5+ Likert items with Cronbach's alpha ≥ .70 are often treated as approximately metric, and t-tests are defensible with a note in the methods section. When in doubt, use the non-parametric alternative.

Q: My Shapiro-Wilk test is significant - what should I do?

A significant Shapiro-Wilk (p < .05) means your data significantly deviates from a normal distribution. Switch to the non-parametric equivalent: Mann-Whitney U instead of independent t-test, Kruskal-Wallis instead of ANOVA, Spearman instead of Pearson. Report the Shapiro-Wilk result and your decision in the methods section.

Q: How do I know if my effect size is large enough to matter?

Cohen's benchmarks are a starting point: for Cohen's d, small = 0.2, medium = 0.5, large = 0.8. For eta-squared, small = .01, medium = .06, large = .14. Always interpret effect size in the context of your field - what counts as meaningful varies by discipline and practical application.

Q: What is the most important assumption to check before running a t-test?

Normality is the most critical assumption for a t-test. Run Shapiro-Wilk on each group separately. If either group is significantly non-normal (p < .05), switch to the Mann-Whitney U test. Also run Levene's test for equal variances - if significant, use Welch's corrected t-test. Both checks must be reported in your thesis methods section.

Q: How do I write up a non-significant result in my thesis?

Report the result exactly as you would a significant one. Example: "An independent samples t-test found no significant difference between groups, t(48) = 0.87, p = .389, d = 0.25." Then interpret: "These results fail to support the hypothesis that..." Never write "the hypothesis was rejected" - you failed to reject the null hypothesis, which is different.

March 20265 min read

Most thesis statistics mistakes are not random - the same five errors appear in thesis after thesis. They are not caught because they look correct in SPSS output. This guide identifies each one, explains why it happens, and gives you a pre-submission checklist to catch them before your supervisor does.

Free sample chapter

Data Analysis From Survey to Results

Step-by-step guidance for choosing the right test, running it, and writing up APA results - in plain language, not theory. Get the free sample chapter when you join the waitlist.

Statistical Analysis Step by Step book cover

Key takeaways

Treating Likert items as metric is the #1 error - single items are ordinal; use Mann-Whitney U, not t-test.
Skipping assumption checks is the #2 error - Shapiro-Wilk and Levene's must be run and reported before every parametric test.
Statistical significance ≠ practical importance - always report Cohen's d, η², or r alongside every p-value.
Never report only significant hypotheses - selective reporting is a methodological flaw, not a writing choice.
p = .000 is impossible - always write p < .001 when SPSS shows zero.

The Most Common Data Analysis Mistake Students MakeWatch on YouTube

Mistake 1: Treating Ordinal Likert Scale Data as Metric (CRITICAL)

Single Likert items (e.g., 1–5 satisfaction rating) are ordinal data. The distance between 1 and 2 is not guaranteed to equal the distance between 4 and 5. Running a t-test or ANOVA on single ordinal items violates the interval scale assumption.

SPSS will not warn you. It will produce a t-statistic and p-value even though the test is not valid for ordinal data.

Data Type	Correct Test for 2 Groups	Correct Test for 3+ Groups
Single Likert item (ordinal)	Mann-Whitney U	Kruskal-Wallis
Composite score (5+ items, α ≥ .70)	t-test (defensible)	ANOVA (defensible)

⚠️

CRITICAL: Check your SPSS Variable View. If any Likert item is set to 'Scale' measurement level, fix it to 'Ordinal'. This one setting error can invalidate your entire analysis.

Mistake 2: Skipping Assumption Checks Before Parametric Tests (CRITICAL)

Running a t-test or ANOVA without checking normality and equal variances is a fundamental methodological error. Your supervisor will ask for these results. 'I ran the test' is not sufficient - you must show the tests are valid.

Fix: run Shapiro-Wilk (normality) and Levene's (equal variances) first. Report both. State which result justified your choice of test.

⚠️

CRITICAL: No assumption checks = no valid interpretation. If you cannot show your data met the parametric assumptions, your test results cannot be trusted.

Mistake 3: Confusing Statistical Significance With Practical Importance (HIGH)

p < .05 means your result is unlikely by chance. It says nothing about whether the effect matters.

With N = 500, almost any difference becomes statistically significant - including differences too small to matter in practice. With N = 20, a large real effect may not reach significance.

Effect size is the measure of practical importance. Always report it.

Test	Effect Size Measure	Small / Medium / Large
t-test	Cohen's d	0.2 / 0.5 / 0.8
ANOVA	η² (eta-squared)	.01 / .06 / .14
Correlation	r	.10 / .30 / .50
Regression	R²	% variance explained

Mistake 4: Reporting Only Significant Results - Selective Reporting (HIGH)

Reporting only the hypotheses that turned out significant and omitting non-significant ones is a form of selective reporting. It is not just a writing style choice - it is a methodological flaw that distorts the picture of what your data actually showed.

All hypotheses must be reported with their full statistics:
t(48) = 0.87, p = .389, d = 0.25

Write: 'No significant difference was found between groups, t(48) = 0.87, p = .389, d = 0.25.'
Do NOT write: 'This hypothesis was not supported' and omit the statistics.

⚠️

Selective reporting can be flagged during your defense as a methodological weakness. Report every hypothesis you tested with its full statistics - significant or not.

Mistake 5: Writing p = .000 in Your Thesis (MEDIUM)

When SPSS shows p = .000, it means the p-value is smaller than .0005 and has been rounded to three decimal places. A true p-value of zero is mathematically impossible.

Always write p < .001 - never p = .000.

This is an easy fix but appears in almost every first thesis draft.

Pre-Submission Checklist: Catch All 5 Mistakes

Run through this checklist before submitting to your supervisor.

Check	What to Verify	Status
Likert measurement level	SPSS Variable View: Likert items = Ordinal (not Scale)	☐
Assumption checks reported	Shapiro-Wilk W and p, Levene's F and p in Methods section	☐
Effect size for every test	Cohen's d, η², or r reported next to every p-value	☐
All hypotheses reported	Non-significant results include full t/F/r, p, and effect size	☐
No p = .000	Search thesis for '.000' - replace all with '< .001'	☐

Frequently asked questions

Is it acceptable to use a t-test on Likert scale data?

▾

It depends on the data. Strictly speaking, single Likert items are ordinal - use Mann-Whitney U. Composite scores averaged across 5+ Likert items with Cronbach's alpha ≥ .70 are often treated as approximately metric, and t-tests are defensible with a note in the methods section. When in doubt, use the non-parametric alternative.

My Shapiro-Wilk test is significant - what should I do?

▾

A significant Shapiro-Wilk (p < .05) means your data significantly deviates from a normal distribution. Switch to the non-parametric equivalent: Mann-Whitney U instead of independent t-test, Kruskal-Wallis instead of ANOVA, Spearman instead of Pearson. Report the Shapiro-Wilk result and your decision in the methods section.

How do I know if my effect size is large enough to matter?

▾

Cohen's benchmarks are a starting point: for Cohen's d, small = 0.2, medium = 0.5, large = 0.8. For eta-squared, small = .01, medium = .06, large = .14. Always interpret effect size in the context of your field - what counts as meaningful varies by discipline and practical application.

What is the most important assumption to check before running a t-test?

▾

Normality is the most critical assumption for a t-test. Run Shapiro-Wilk on each group separately. If either group is significantly non-normal (p < .05), switch to the Mann-Whitney U test. Also run Levene's test for equal variances - if significant, use Welch's corrected t-test. Both checks must be reported in your thesis methods section.

How do I write up a non-significant result in my thesis?

▾

Report the result exactly as you would a significant one. Example: "An independent samples t-test found no significant difference between groups, t(48) = 0.87, p = .389, d = 0.25." Then interpret: "These results fail to support the hypothesis that..." Never write "the hypothesis was rejected" - you failed to reject the null hypothesis, which is different.

Not sure which statistical test to use?

Answer 5 quick questions about your research design and get the right test - with an explanation of why - in under two minutes.

Statoria Team

Statistics educators & software developers

We build Statoria to help bachelor and master students get through their thesis data analysis without stress. Our guides are written by researchers with experience in social science statistics and student supervision.

Related guides

Data analysis