When Chi-Squared Assumptions Fail
You'll learn when the Chi-Squared test breaks down and why Fisher's Exact Test becomes your go-to alternative.
The Problem with Small Samples
You've learned that Chi-Squared tests require minimum expected frequencies (usually at least 5 per cell) to work reliably. But what happens when you're analyzing a small dataset or a rare event? Your expected cell counts might drop below this threshold, making the Chi-Squared approximation inaccurate.
Think of it like using a map designed for driving cross-country to navigate your neighborhood streets—the scale just isn't right for the job.
Why Small Counts Break Chi-Squared
The Chi-Squared test relies on an approximation to a continuous distribution. When expected frequencies are low (say, under 5), this approximation becomes poor because:
- The discrete nature of count data matters more with small numbers
- The sampling distribution doesn't follow the Chi-Squared distribution well
- P-values become unreliable—you might reject when you shouldn't (or vice versa)
Enter Fisher's Exact Test
Fisher's Exact Test solves this by calculating the exact probability of observing your data pattern under the null hypothesis. Instead of approximating with a continuous distribution, it uses combinatorics to enumerate all possible tables with the same row and column totals.
It's like counting every single possible outcome by hand rather than using a mathematical shortcut. More laborious, but perfectly accurate even with tiny samples.
When to Make the Switch
Use Fisher's Exact Test when:
- Any expected cell frequency is less than 5
- Total sample size is very small (say, under 20-30)
- You have a 2×2 table and want the most accurate result possible
Key Takeaway: When sample sizes are small or expected frequencies drop below 5, Chi-Squared assumptions fail—switch to Fisher's Exact Test for accurate results based on exact probabilities rather than approximations.