Course contentsShow
← Data Science
Lesson 434 of 2,14511. Hypothesis Tests for Proportions and CategoriesFree lesson

Fisher's Exact vs Chi-Squared: When to Use Each

Compare performance and appropriateness of Fisher's Exact and Chi-Squared tests under different conditions.

Fisher's Exact vs Chi-Squared: When to Use Each

You'll learn: How to decide whether Fisher's Exact Test or the Chi-Squared Test is more appropriate for your categorical data analysis.

The Decision Framework

Both tests evaluate relationships between categorical variables in contingency tables, but they work differently under the hood. The Chi-Squared test uses an approximation based on a mathematical distribution, while Fisher's Exact Test calculates the exact probability by considering all possible table arrangements.

Think of it like navigation: Chi-Squared is like using a GPS with slight rounding errors that works great on highways (large samples), while Fisher's Exact is like manually calculating every possible route—more precise but computationally intensive.

When to Use Chi-Squared

Use the Chi-Squared test when:

  • All expected cell frequencies are ≥ 5 (the critical rule of thumb)
  • You have a large sample size (typically n > 20–40)
  • Your table is larger than 2×2 (where Fisher's becomes computationally expensive)

The Chi-Squared approximation performs well under these conditions and is much faster to compute.

When to Use Fisher's Exact

Use Fisher's Exact Test when:

  • Any expected cell frequency is < 5
  • You have small sample sizes
  • You need exact p-values (not approximations)
  • You have a 2×2 table (most common scenario)

Fisher's Exact doesn't rely on the normal approximation, so it remains valid even with tiny cell counts like 1 or 2.

The Gray Zone

For moderately-sized 2×2 tables with all expected frequencies around 5–10, either test typically gives similar results. Fisher's is conservative (safer), while Chi-Squared is standard practice in many fields.

Key Takeaway: Use Fisher's Exact Test when expected cell frequencies are below 5 or sample sizes are small; use Chi-Squared Test for larger samples where the approximation is reliable and computationally efficient.