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.