Confidence Intervals for Difference in Proportions
What you'll learn: How to construct confidence intervals to compare proportions from two independent groups.
The Core Idea
Sometimes you want to compare two groups: Does treatment A work better than treatment B? Do men and women differ in their voting preferences? Instead of looking at a single proportion, you're interested in the difference between two proportions (p₁ - p₂).
A confidence interval for this difference tells you the plausible range of how much these two population proportions truly differ, accounting for sampling variability in both groups.
How It Works
You have two independent samples:
- Group 1 with sample proportion p̂₁
- Group 2 with sample proportion p̂₂
The point estimate of the difference is simply p̂₁ - p̂₂.
To build the confidence interval around this difference, you need the standard error of the difference:
SE(p̂₁ - p̂₂) = √[p̂₁(1-p̂₁)/n₁ + p̂₂(1-p̂₂)/n₂]
Then construct the interval just like before:
(p̂₁ - p̂₂) ± z* × SE(p̂₁ - p̂₂)
where z* is your critical value (1.96 for 95% confidence).
Interpreting the Interval
If the interval contains zero, the data are consistent with no difference between the groups. If the entire interval is positive, group 1 likely has a higher proportion. If entirely negative, group 2 is higher.
Think of it as building a confidence interval for each group separately, then combining the uncertainty from both.
Key Takeaway: Confidence intervals for p₁ - p₂ let you quantify the plausible difference between two independent proportions, helping you determine whether groups truly differ or if observed differences might just be sampling variation.