Course contentsShow
← Data Science
Lesson 296 of 2,1458. Confidence IntervalsFree lesson

Sample Size for Comparing Two Groups

Calculate required sample size for detecting differences between two means or two proportions.

Sample Size for Comparing Two Groups

What you'll learn: How to determine how many observations you need in each group to reliably detect a meaningful difference between two means or two proportions.

The Core Challenge

When comparing two groups—say, testing whether a new drug lowers blood pressure more than a placebo—you need enough data to spot a real difference if it exists. Too few observations, and even a genuine effect might look like random noise. The required sample size depends on four key factors:

  1. The effect size you want to detect (how big a difference matters?)
  2. Variability within groups (more scatter = more data needed)
  3. Significance level (α), typically 0.05 (controls false positives)
  4. Power (1 - β), typically 0.80 (your chance of detecting a real effect)

For Comparing Two Means

The formula builds on what you've learned about standard error and confidence intervals:

n = 2 × (Z_α/2 + Z_β)² × σ² / δ²

Where:

  • n = sample size per group
  • Z_α/2 = critical value for your confidence level
  • Z_β = critical value for your desired power
  • σ = standard deviation (pooled or estimated)
  • δ = minimum detectable difference you care about

For Comparing Two Proportions

n = 2 × (Z_α/2 + Z_β)² × p̄(1 - p̄) / (p₁ - p₂)²

Where p̄ is the average of the two proportions you're comparing.

Think of It Like This

Imagine trying to weigh two feathers that differ by 0.1 grams. A cheap scale that varies by ±5 grams won't help—you'd need many, many measurements to average out the noise. A sensitive scale (less variability) needs fewer readings.

Key Takeaway: Larger effect sizes and lower variability require smaller samples; higher confidence and power demand more data. Always plan your sample size before collecting data to ensure you can actually detect the differences that matter.