Two-Sample t-Test: Welch's Test for Unequal Variances
What You'll Learn
You'll learn how to compare two population means when you cannot assume the populations have equal variances, using Welch's approximation for the two-sample t-test.
The Problem with Unequal Variances
In the previous lesson, you learned the two-sample t-test for independent samples assuming equal population variances. But what if that assumption is unrealistic? When two populations have different variances (σ₁² ≠ σ₂²), the standard pooled t-test can give misleading results with incorrect Type I error rates.
Welch's t-test solves this by not pooling the sample variances. Instead, it uses each sample's variance separately and adjusts the degrees of freedom accordingly.
Welch's Test Statistic
The test statistic looks similar to the equal-variance version:
t = (x̄₁ - x̄₂) - (μ₁ - μ₂)
─────────────────────
√(s₁²/n₁ + s₂²/n₂)
Where x̄₁ and x̄₂ are sample means, s₁² and s₂² are sample variances, and n₁ and n₂ are sample sizes.
The Key Difference: Degrees of Freedom
The tricky part is calculating degrees of freedom using the Welch-Satterthwaite approximation:
df ≈ (s₁²/n₁ + s₂²/n₂)²
─────────────────────────────────────
(s₁²/n₁)²/(n₁-1) + (s₂²/n₂)²/(n₂-1)
This formula usually gives a non-integer value (round down to be conservative). When variances are equal, this approximates n₁ + n₂ - 2.
When to Use Welch's Test
Use Welch's test when:
- Sample variances differ substantially (rule of thumb: larger variance > 2× smaller variance)
- Sample sizes are unequal
- You're uncertain about the equal-variance assumption
Many statisticians recommend using Welch's test by default since it's robust and performs well even when variances are equal.
Key Takeaway: Welch's t-test extends the two-sample t-test to situations with unequal population variances by using separate variance estimates and an adjusted degrees of freedom calculation, making it a safer default choice for comparing two means.