Course contentsShow
← Data Science
Lesson 412 of 2,14511. Hypothesis Tests for Proportions and CategoriesFree lesson

Confidence Interval for Difference

Constructing intervals using unpooled standard errors for estimation.

Confidence Interval for Difference

What You'll Learn

You'll learn how to construct a confidence interval for the difference between two proportions using unpooled standard errors for estimation.

The Core Idea

When you've tested whether two proportions differ, the next natural question is: by how much? A confidence interval for the difference gives you a range of plausible values for the true difference between two population proportions.

Unlike hypothesis testing (where you used a pooled proportion to calculate the standard error), confidence intervals use unpooled standard errors. Why? Because you're no longer assuming the proportions are equal—you're trying to estimate their actual difference!

The Formula

For two independent samples with proportions p̂₁ and p̂₂:

Point estimate: p̂₁ - p̂₂

Standard error (unpooled):

SE = sqrt[(p̂₁(1-p̂₁)/n₁) + (p̂₂(1-p̂₂)/n₂)]

Confidence interval:

(p̂₁ - p̂₂) ± z* × SE

Where z* is the critical value (e.g., 1.96 for 95% confidence).

Why Unpooled?

Think of it like measuring two different ingredients. When testing if they're the same, you might blend them together (pooled). But when estimating their individual properties, you keep them separate (unpooled) to capture their true variability.

Each sample proportion has its own uncertainty, and the unpooled approach respects that by calculating separate variance contributions from each group.

Example Context

If 40% of Group A (n=200) and 55% of Group B (n=180) prefer a product, your 95% CI might be (-0.25, -0.05), suggesting Group B's preference rate is between 5 and 25 percentage points higher.

Key Takeaway: Use unpooled standard errors when constructing confidence intervals for the difference in proportions—each group's variability matters independently for estimation.