The probability a test correctly detects a real difference (1-beta); higher power requires more samples.