The Assumption of Equal Variances (Homoscedasticity)
You'll learn: Why two-sample t-tests assume equal variances and when to use Welch's correction instead.
What Is Homoscedasticity?
When comparing two groups using a two-sample t-test, we often assume both populations have the same variance (spread). This is called homoscedasticity (homo = same, scedasticity = scatter). Think of it like comparing test scores from two classrooms: if one class has scores clustered tightly around the mean while the other has wildly varying scores, they have different variances.
Why Does This Matter?
The classic pooled variance t-test (from lesson 361) combines variance estimates from both groups into a single pooled estimate. This works beautifully when variances are truly equal—you gain statistical power and cleaner calculations. But if the variances are actually different, your test results can be misleading, especially if sample sizes differ between groups.
When Variances Aren't Equal
If your groups have unequal variances (heteroscedasticity), you should use Welch's t-test instead. Welch's correction doesn't pool the variances—it handles them separately and adjusts the degrees of freedom accordingly. It's more robust when the equal-variance assumption fails, with minimal downside even when variances are equal.
How to Check
You can test for equal variances using Levene's test or the F-test (covered in lesson 363), or simply inspect side-by-side boxplots. As a rule of thumb: if one group's variance is more than 2–3 times the other's, consider Welch's test.
Key Takeaway: The pooled t-test assumes equal variances; when this assumption is violated, Welch's t-test provides a safer, more reliable alternative without pooling.