Course contentsShow
← Data Science
Lesson 376 of 2,14510. Hypothesis Tests for MeansFree lesson

The Assumption of Normality in t-Tests

Why t-tests assume normally distributed data or large samples, and when this assumption matters most.

The Assumption of Normality in t-Tests

You'll learn: Why t-tests require normally distributed data (or large samples), and when this assumption is most critical.

Why Normality Matters

When you perform a t-test, you're making decisions based on the sampling distribution of the mean. The mathematics behind t-tests assume this sampling distribution follows a t-distribution, which closely resembles a normal curve. For this to work properly, either:

  1. Your raw data should be approximately normally distributed, OR
  2. Your sample size should be large enough for the Central Limit Theorem to kick in

Think of it like a recipe: if your ingredients (data) have certain properties, the final dish (sampling distribution) comes out predictable and reliable.

When It Matters Most

Small samples (n < 30): The normality assumption is critical. With few observations, outliers or severe skewness can distort your t-statistic and lead to incorrect p-values. Always check your data visually (histograms, Q-Q plots) before proceeding.

Large samples (n ≥ 30): The assumption becomes less critical because the Central Limit Theorem protects you. Even if your raw data isn't normal, the distribution of sample means tends toward normality as sample size increases.

Paired t-tests: Remember, it's the differences that need to be normally distributed, not the original paired observations.

What to Do If Normality Fails

With small samples and non-normal data, consider:

  • Transforming your data (log, square root)
  • Using non-parametric alternatives you'll learn later
  • Collecting more data when possible

Key Takeaway: The normality assumption matters most for small samples; with large samples (n ≥ 30), t-tests are robust to violations thanks to the Central Limit Theorem, but extreme outliers can still cause problems.