The Continuity Correction
You'll learn when and how to apply the continuity correction to improve normal approximations of discrete probability distributions.
Why We Need It
When you approximate a discrete distribution (like the binomial) with a continuous distribution (the normal), you face a fundamental mismatch: discrete variables take exact integer values (0, 1, 2, 3, ...), while the normal distribution is continuous and spreads probability smoothly over intervals.
For example, if you want P(X = 5) for a binomial variable, you're asking for the probability at one exact point. But in the normal distribution, the probability of any single exact point is zero! The solution is the continuity correction.
How It Works
Think of each discrete value as occupying a "bin" of width 1, centered on that integer. The value 5 actually represents the interval from 4.5 to 5.5. When approximating:
- P(X = k) becomes P(k – 0.5 < X < k + 0.5)
- P(X ≤ k) becomes P(X < k + 0.5)
- P(X ≥ k) becomes P(X > k – 0.5)
- P(X < k) becomes P(X < k – 0.5)
The key is to add or subtract 0.5 to "capture" the full width of the discrete bar you're approximating.
Example in Action
Suppose X ~ Binomial(n=20, p=0.5) and you want P(X ≤ 10).
Without correction, you'd compute P(Z ≤ standardized 10).
With correction, you compute P(Z ≤ standardized 10.5), giving the normal curve a fair chance to capture the entire bar at X = 10.
Key Takeaway: The continuity correction adds or subtracts 0.5 to discrete values before applying the normal approximation, accounting for the fact that discrete bars have width while the normal curve is continuous.