Course contentsShow
← Data Science
Lesson 322 of 2,1459. Hypothesis Testing FoundationsFree lesson

Exact vs Asymptotic P-Values

Distinguish between p-values calculated from exact distributions versus large-sample approximations and when each applies.

Exact vs Asymptotic P-Values

You'll learn when to trust exact mathematical p-values versus large-sample approximations.

The Core Distinction

When you calculate a p-value, you're measuring how extreme your test statistic is under the null hypothesis. There are two ways to find this:

Exact p-values use the true, theoretical probability distribution without approximation. You calculate them directly from the exact mathematical formula for your test statistic's distribution. Think of this as measuring with a precision scale.

Asymptotic p-values rely on large-sample approximations (like the Central Limit Theorem). As your sample size grows, your test statistic's distribution approaches a normal or other standard distribution. These are like using a "good enough" ruler when precise measurement is impractical.

When Each Applies

Use exact p-values when:

  • You have small sample sizes (typically n < 30)
  • The exact distribution is computationally feasible
  • You need precise results for critical decisions
  • Example: Fisher's exact test for 2×2 tables with small counts

Use asymptotic p-values when:

  • Your sample size is large (CLT kicks in)
  • Computing exact distributions is too complex
  • The approximation error is negligible
  • Example: Z-tests and large-sample t-tests

The Practical Reality

Most standard hypothesis tests (Z-tests, t-tests, chi-square tests) give you asymptotic p-values based on mathematical approximations. They work beautifully when assumptions hold—especially with larger samples. Exact methods are your safety net when samples are tiny or distributions are tricky.

Key Takeaway: Exact p-values come from true distributions and suit small samples; asymptotic p-values use CLT-based approximations and work well for large samples. Know your sample size and distribution assumptions to choose wisely.