Sample Size and Power for Correlation Tests
What you'll learn: How to determine the minimum sample size needed to reliably detect a correlation of a given strength.
Why Sample Size Matters for Correlations
Just as with other hypothesis tests, correlation tests require adequate statistical power—the probability of detecting a real correlation when it exists. Too small a sample might miss a meaningful relationship (Type II error), while unnecessarily large samples waste resources.
The required sample size depends on three factors:
- The expected correlation magnitude (effect size): Smaller correlations require larger samples to detect reliably
- Desired significance level (α): Typically 0.05
- Desired power (1 - β): Typically 0.80 (80% chance of detecting the effect)
The Intuition
Think of fishing for a signal in noise. A strong correlation (say, r = 0.7) is like a loud siren—you'll hear it even with few observations. A weak correlation (r = 0.2) is like a whisper—you need many observations to distinguish it from random fluctuation.
Practical Guidelines
For Pearson correlation with α = 0.05 and power = 0.80:
- To detect r = 0.1 (small): ~783 observations needed
- To detect r = 0.3 (medium): ~84 observations needed
- To detect r = 0.5 (large): ~29 observations needed
These calculations use power analysis formulas that account for the correlation coefficient's sampling distribution under the alternative hypothesis. The exact formula involves transforming r using Fisher's z-transformation to stabilize variance.
When Planning Research
Before collecting data, decide the smallest correlation you care about detecting (minimum effect size of interest), then calculate required N using power analysis software or tables.
Key Takeaway: Larger samples are needed to detect smaller correlations with adequate power; plan your sample size before data collection to ensure your correlation test can detect effects that matter.