5 Statistical Mistakes That Fail Thesis Defenses (And How to Avoid Them)

Examiners don't fail students for having imperfect data — they fail them for analysis errors that reveal a misunderstanding of the methods. The good news: the mistakes that sink defenses are predictable, and almost all of them are preventable. Here are the five that come up most often, why examiners care, and how to avoid each.

Mistake 1: Running the Wrong Test for the Design

This is the most damaging error because it invalidates everything that follows. The classic examples:

• Using an independent t-test on paired/repeated data (pre-test vs post-test on the same people).
• Running multiple t-tests instead of an ANOVA when comparing three or more groups — which inflates your Type I error rate.
• Treating ordinal Likert items as continuous without justification, or using a parametric test when the data clearly violates its assumptions.

Why examiners care: the wrong test means your p-values are meaningless. An examiner who spots this will pull the entire results chapter into question.

How to avoid it: match the test to your design before you touch the data — number of groups, paired vs independent, type of outcome variable. Our guide to choosing the right statistical test walks through the decision tree.

Mistake 2: Never Checking (or Reporting) Assumptions

Every parametric test has assumptions — normality, homogeneity of variance, independence, linearity, no severe multicollinearity. Students routinely run the test and report the result without ever checking whether the assumptions held.

Why examiners care: this is the question examiners love to ask because they know most students skip it. "Did you check for normality? How? What did you do about the violation?" If you have no answer, it signals you ran the analysis mechanically without understanding it.

How to avoid it:

• Check normality with Shapiro-Wilk and Q-Q plots (here's how in SPSS).
• Check homogeneity of variance with Levene's test.
• Check multicollinearity with VIF/tolerance in regression.
• And critically — report that you checked, state the results, and explain what you did if an assumption was violated (e.g., switched to a non-parametric test, used a robust correction). Examiners reward this; it shows competence.

Mistake 3: Misinterpreting the P-Value

The single most common verbal slip in defenses. Students say things like:

• "p = .03 means there's a 97% chance my hypothesis is true." ❌
• "p < .05 proves the effect exists." ❌
• "A non-significant result means there is no effect." ❌

Why examiners care: misreading the p-value reveals a shaky grasp of inference itself. It's a fast way to lose credibility even if your analysis was technically correct.

How to avoid it: internalize what the p-value actually is — the probability of observing your data (or more extreme) if the null hypothesis were true. Significance means the result is unlikely under the null; it does not prove your hypothesis, measure effect size, or confirm a null when non-significant. Pair every significant result with an effect size (Cohen's d, η², r) so you can speak to how big the effect is, not just whether it exists.

Mistake 4: Ignoring Effect Size and Practical Significance

With a large sample, almost anything becomes "statistically significant." A correlation of r = .08 can have p < .001 with enough participants — and mean almost nothing in practice.

Why examiners care: reporting only p-values suggests you don't understand the difference between statistical and practical significance. Examiners will ask, "Yes, it's significant — but is it meaningful?"

How to avoid it: always report and interpret effect sizes alongside significance. State whether the effect is small, medium, or large, and what that means for your research question. This single habit makes a results chapter look markedly more sophisticated.

Mistake 5: Inconsistent or Non-APA Reporting

Numbers that don't match between your tables, text, and abstract. Degrees of freedom that don't add up to your sample size. Missing test statistics. p-values reported as "p = .000" (it's never exactly zero — report as p < .001).

Why examiners care: sloppy reporting makes them wonder what else is sloppy. Inconsistent numbers across sections is a red flag that the analysis was rushed or partially redone without updating everything.

How to avoid it:

• Follow APA 7th conventions precisely — F(2, 147) = 5.34, p = .006, η² = .07.
• Make sure your degrees of freedom reconcile with your sample size.
• Cross-check that every number matches across tables, text, and abstract.
• Report exact p-values (except p < .001).

Our APA reporting guide has the templates for each common test.

The Pattern Behind All Five

Notice what connects these: every one is an error of understanding, not effort. Examiners aren't hunting for a tiny rounding error — they're probing whether you understand the methods you used. That's why the defense is oral. You can't memorize your way past "why did you choose this test, and how do you know its assumptions held?"

The fix, therefore, isn't just to get the analysis right — it's to understand it well enough to defend it. That's the difference between help that gets you through the viva and "help" that leaves you exposed the moment an examiner asks a follow-up.

A Pre-Defense Checklist

Before you walk in, confirm you can answer:

• Why is this the correct test for my design?
• Which assumptions did I check, how, and what were the results?
• What does my p-value actually mean — in plain words?
• What is the effect size, and is the effect practically meaningful?
• Do all my numbers match across tables, text, and abstract?

If you can answer all five confidently, you've cleared the bar that fails most candidates.

---

We don't just run your analysis — we make sure you can defend it. Every Insighter Digital project includes the right test for your design, full assumption checks, effect sizes, APA-formatted reporting, and a walkthrough so the examiner's questions hold no surprises. Get defense-ready or view our packages.