Continuity Correction — Definition, Formula & Example
Continuity correction is the practice of adding or subtracting 0.5 to a whole-number boundary when you use a smooth normal curve to estimate probabilities for a count. It treats each whole number as the interval from to , so the area you measure matches the bar you meant to include.
Let be an integer-valued random variable, typically , approximated by a continuous random variable with and . A continuity correction replaces each event about with an event about whose boundary is shifted by . The shift includes or excludes the full unit interval that represents the value .
Key Formula
Where:
- = The discrete random variable, such as a binomial count
- = The approximating normal random variable with the same mean and standard deviation as X
- = An integer boundary value of X
Worked Example
Problem: You flip a fair coin 100 times. Use the normal approximation with continuity correction to estimate the probability of getting at most 45 heads.
Step 1: Let be the number of heads, so . Find the mean and standard deviation. Both and are at least 10, so the normal approximation is reasonable.
Step 2: Apply the correction. "At most 45" includes the whole bar for 45, which reaches up to 45.5.
Step 3: Standardize the boundary and read the area from a standard normal table.
Answer: . The exact binomial probability is also about 0.1841. Without the correction you would get and a probability of about 0.1587, which is noticeably too low.
Why It Matters
In AP Statistics and introductory college statistics, you often estimate binomial probabilities with a normal curve. Without the correction, those estimates can be off by several percentage points. The same idea appears as Yates's correction in chi-square tests on 2×2 tables.
Common Mistakes
Mistake: Shifting strict inequalities the same way as non-strict ones, for example writing .
Correction: First rewrite the event using whole numbers: means . Then apply the correction to get . Ask yourself whether the bar at the boundary value belongs in the event.
