Pooled vs Unpooled Variance Approaches
You'll learn: When to combine variance estimates from two groups versus treating them separately when building confidence intervals for the difference between two means.
The Core Idea
When comparing two groups (say, average test scores for Group A and Group B), you need to estimate variability. You have two choices:
Pooled variance assumes both groups have the same underlying population variance. You combine (pool) the variance from both samples into one shared estimate. Think of it like averaging recipes from two cooks who use the same technique—you trust they vary food portions similarly.
Unpooled variance treats each group's variance as unique. You keep them separate. This is like recognizing that one cook measures carefully while the other eyeballs everything—they have different levels of consistency.
When to Use Each
Use pooled variance when:
- You believe both populations have equal variances (homogeneity of variance)
- Sample sizes are similar
- You want simpler calculations and slightly more power
Use unpooled variance when:
- The two groups clearly have different spreads
- Sample sizes differ substantially (one group much larger than the other)
- You're unsure—unpooled is the safer default
The unpooled approach is also called the Welch's t-test method and doesn't assume equal variances. It's more conservative but protects you from incorrect conclusions when variances differ.
Practical Example
Comparing average recovery times: Hospital A (n=30, consistent care) vs Hospital B (n=15, highly variable patient cases). The different sample sizes and different variability suggest unpooled variance is appropriate.
Key Takeaway: Use pooled variance when groups have similar spread and size; default to unpooled variance when in doubt or when variances appear unequal—it's the safer, more flexible choice.