Normal Approximation to the Binomial Distribution: Conditions, Formulas & Continuity Correction

What Is Normal Approximation to the Binomial?

The binomial distribution gives the exact probability of k successes in n independent trials when each success has probability p. The formula involves factorials that grow rapidly with n, making hand calculation impractical for large samples. The normal approximation sidesteps this by treating the binomial as continuous, converting the target value to a Z-score, and reading from the standard normal distribution table.

The theoretical basis is the Central Limit Theorem: because a binomial random variable is the sum of n independent Bernoulli trials, it converges to a normal distribution as n increases. This is why accuracy improves with larger samples and why the approximation is less reliable near the condition thresholds.

Conditions: np ≥ 5 and n(1−p) ≥ 5

The normal approximation is only reliable when the binomial distribution is reasonably symmetric. Symmetry depends on p: distributions with p near 0.5 are almost perfectly bell-shaped even at moderate n, while distributions with extreme p are skewed and need much larger n. The standard check is the Rule of 5, also called the np ≥ 5 and nq ≥ 5 rule (where q = 1 − p).

Why the np ≥ 5 Condition Matters

When p is very small or very large, the binomial is skewed. For example, with n = 20 and p = 0.05: np = 1, far below 5. The distribution is strongly right-skewed — a symmetric bell curve fits it poorly. As p approaches 0.5, the binomial becomes symmetric and the approximation works well even at moderate n. The Rule of 5 catches the worst cases; the Rule of 10 adds safety margin for applied work where precision matters. Full coverage of this skew behavior is in the binomial distribution guide.

Normal Approximation Formulas: μ = np and σ = √(np(1−p))

Three quantities define the approximating normal distribution: the mean, the standard deviation, and the Z-score after applying the continuity correction. These formulas are directly tested in AP Statistics and A-Level Maths exams.

Once you have the Z-score, find the area using the standard normal (Z) table, a TI-84 normalcdf, or stats.norm.cdf() in Python. For right-tail probabilities, subtract the table value from 1. For between-two-Z-scores problems, subtract the smaller area from the larger.

Continuity Correction ±0.5 — Complete Rules

The binomial distribution counts integers: 0, 1, 2, 3, …. The normal distribution covers every real number. When mapping integer k onto the continuous curve, you are treating a single point as a strip of width 1 — specifically from k − 0.5 to k + 0.5. The continuity correction expands each integer to that strip, so the continuous area matches the discrete bar.

Without the correction, you estimate the probability at a single point, which has measure zero on a continuous distribution. With it, you read the area of the strip that corresponds to the discrete count. For large n the difference is minor; near the np ≥ 5 threshold it can shift the result noticeably.

Accuracy: With vs. Without Continuity Correction

The table below compares the normal approximation — with and without the ±0.5 correction — against the exact binomial probability for a selection of values with n = 100, p = 0.5.

The continuity correction essentially eliminates the error at the center of the distribution (near the mean) and reduces it substantially elsewhere. For n = 100 and p = 0.5, the uncorrected approximation at P(X ≤ 50) is off by 4 percentage points; the corrected version matches exactly. There is no reason to skip the correction — it takes one arithmetic step and always improves accuracy.

How to Use Normal Approximation to the Binomial — 8 Steps

Worked Examples

Each example follows the 8-step procedure with full arithmetic shown. Z-score lookups use the four-decimal-place table from the NIST/SEMATECH Engineering Statistics Handbook and Penn State STAT 414 course material.

Example 1 — Airline No-Show Problem: P(X < 15)

Example 2 — Exact Single Value: P(X = 45) from 100 Coin Flips

Example 3 — Poisson to Normal: P(X > 22)

Example 4 — Chi-Square Normal Approximation

Normal Approximation to the Hypergeometric Distribution

The hypergeometric distribution (sampling without replacement from a finite population of size N, K successes, n draws) can also be approximated by a normal distribution when the sample size n is large relative to the population and both expected successes and failures are at least 5. The mean is μ = nK/N and the variance is σ² = nK(N−K)(N−n) / [N²(N−1)]. The continuity correction ±0.5 applies just as with the binomial.

A simpler path: when n/N < 0.05 (the 5% condition), the hypergeometric is well approximated by the binomial with p = K/N, and then the binomial normal approximation applies directly. See the binomial distribution guide for details on this substitution.

Normal Approximation Calculator

Enter the binomial parameters, select the inequality type, and click Calculate. The calculator shows every intermediate step: np ≥ 5 conditions check, mean μ = np, standard deviation σ = √(np(1−p)), continuity-corrected boundary, Z-score, and final probability.

For a standalone tool you can bookmark, see the full Normal Approximation Calculator and the related Binomial Distribution Calculator.

Method Comparison: Exact Binomial vs. Normal Approximation vs. Poisson

For a deeper comparison of the normal and t-distributions, see the T-distribution vs. normal distribution guide. Coverage of the empirical rule is in the empirical rule article. For hypothesis testing for proportions, the same np ≥ 5 conditions apply when using the Z-test for a proportion.

Common Mistakes to Avoid

Python and R Implementation

Software eliminates arithmetic risk. Both examples compute the approximated probability and the exact binomial result side by side. Penn State STAT 414 covers the theoretical derivation in more depth at online.stat.psu.edu/stat414/lesson/27.

Python (SciPy)

import numpy as np
import scipy.stats as stats

n, p, k = 100, 0.4, 35

# Step 1–2: Check conditions np >= 5 and n(1-p) >= 5
np_val = n * p
nq_val = n * (1 - p)
print(f"np = {np_val:.1f}, n(1-p) = {nq_val:.1f}")
assert np_val >= 5 and nq_val >= 5, "Conditions np≥5 and n(1-p)≥5 not met"

# Steps 3–4: Mean mu = np and sigma = sqrt(np(1-p))
mu = n * p
sigma = np.sqrt(n * p * (1 - p))

# Step 5: Continuity correction for P(X <= k)
k_corrected = k + 0.5

# Step 6–7: Z-score and probability
z = (k_corrected - mu) / sigma
prob_approx = stats.norm.cdf(z)

# Exact binomial for comparison
prob_exact = stats.binom.cdf(k, n, p)

print(f"mu = {mu}, sigma = {sigma:.4f}, Z = {z:.4f}")
print(f"Approximated P(X ≤ {k}) = {prob_approx:.4f}")
print(f"Exact P(X ≤ {k})        = {prob_exact:.4f}")

R

n <- 100
p <- 0.4
k <- 35

# Step 1–2: Check conditions np >= 5 and n(1-p) >= 5
np_val <- n * p
nq_val <- n * (1 - p)
cat("np =", np_val, " nq =", nq_val, "\n")
stopifnot(np_val >= 5, nq_val >= 5)

# Steps 3–4: mu = np and sigma = sqrt(np(1-p))
mu <- n * p
sigma <- sqrt(n * p * (1 - p))

# Step 5–7: Continuity correction and Z-score for P(X <= k)
k_corrected <- k + 0.5
prob_approx <- pnorm(k_corrected, mean = mu, sd = sigma)

# Exact binomial
prob_exact <- pbinom(k, size = n, prob = p)

cat("mu =", mu, " sigma =", round(sigma, 4), "\n")
cat("Approximated:", round(prob_approx, 4), "\n")
cat("Exact:       ", round(prob_exact, 4), "\n")

Normal Approximation and the Sampling Distribution of p̂

When working with sample proportions, the same normal approximation logic applies. For a sample of size n drawn from a population with true proportion p, the sample proportion p̂ = X/n has an approximately normal sampling distribution when np ≥ 10 and n(1−p) ≥ 10 (most applied courses use the stricter threshold here).

The Z-score for a sample proportion is Z = (p̂ − p) / √(p(1−p)/n). This is the test statistic used in one-proportion Z-tests and in constructing confidence intervals for proportions.

Related Concepts

FAQs