Course contentsShow
← Data Science
Lesson 277 of 2,1458. Confidence IntervalsFree lesson

Standard Error for Proportions

Calculating the standard error of a proportion and when the normal approximation is valid (np ≥ 10, n(1-p) ≥ 10).

Standard Error for Proportions

You'll learn how to measure the variability of sample proportions and when you can trust the normal approximation.


What Is Standard Error for Proportions?

When you estimate a proportion from a sample (like the percentage of customers who'd recommend your product), that estimate has uncertainty. The standard error quantifies how much your sample proportion might vary from sample to sample.

The formula is:

SE=p(1−p)nSE = \sqrt{\frac{p(1-p)}{n}}

Where:

  • p is your sample proportion (e.g., 0.35 for 35%)
  • n is your sample size
  • (1-p) is the complement (proportion who don't have the trait)

Understanding the Formula

Think of p(1-p) as measuring "how split" your outcomes are. When p = 0.5 (50/50 split), variability is highest. When p is near 0 or 1 (almost everyone the same), variability is lower. Dividing by n means larger samples reduce uncertainty—you get more stable estimates.

The Normal Approximation Rule

We often use the normal distribution to build confidence intervals for proportions, but this only works reliably when both:

  • np ≥ 10
  • n(1-p) ≥ 10

These rules ensure you have "enough" successes and failures in your sample. With 200 people and 30% saying "yes," you have 60 successes and 140 failures—both well above 10, so the normal approximation is valid. But with only 15 people and 20% saying "yes" (3 successes), the approximation breaks down.

Key Takeaway: The standard error for proportions is √[p(1-p)/n], and you can safely use normal-based inference when both np ≥ 10 and n(1-p) ≥ 10.