Course contentsShow
← Data Science
Lesson 374 of 2,14510. Hypothesis Tests for MeansFree lesson

Assumptions of the Paired t-Test

Verify that differences are independent and approximately normally distributed.

Assumptions of the Paired t-Test

What you'll learn: You'll discover the two key conditions your data must satisfy before running a paired t-test—and why they matter.

Why Assumptions Matter

Just like a recipe works only with the right ingredients, the paired t-test produces reliable results only when your data meets certain conditions. Violating these assumptions can lead to incorrect p-values and faulty conclusions.

The Two Key Assumptions

1. Independence of Paired Differences

Each pair's difference must be independent of all other pairs' differences. Think of it like tasting cookies from different batches—one cookie's sweetness shouldn't affect another's measurement.

Example: If you measure blood pressure before/after treatment for 30 patients, each patient's change should be unrelated to others' changes. However, if you measured the same patient multiple times, those repeated measurements would violate independence.

2. Normality of Differences

The differences (not the original measurements) should follow an approximately normal distribution. This is less critical with larger sample sizes (n ≥ 30) thanks to the Central Limit Theorem.

How to check:

  • Create a histogram or Q-Q plot of the differences
  • Look for severe skewness or extreme outliers
  • For small samples (n < 30), normality is more important

What If Assumptions Are Violated?

  • Non-independent pairs: Results are unreliable; reconsider your study design
  • Non-normal differences (small n): Consider the Wilcoxon signed-rank test (a non-parametric alternative)
  • Non-normal differences (large n): Paired t-test is usually still robust

Key Takeaway: Before running a paired t-test, verify that your paired differences are independent observations and check their distribution for approximate normality—especially with smaller samples.