Visual Tools
Calculators
Tables
Mathematical Keyboard
Converters
Other Tools

How to Approximate a Binomial with a Normal

A binomial with many trials has a bell-shaped histogram, so a normal curve with the same mean and standard deviation gives its probabilities. Because the binomial lives on whole numbers, widen each boundary by half a unit first (the continuity correction), then standardize.

Recognize the task

A binomial count with a large nn and a long sum: "100100 coin tosses, P(at least 60 heads)P(\text{at least } 60 \text{ heads})", "200200 voters, 30%30\% support, P(55≤X≤70)P(55 \le X \le 70)". Test: np≥10np \ge 10 and n(1−p)≥10n(1-p) \ge 10. If pp is tiny instead, use the Poisson approximation.
Why sums of many trials become normal is explained in the central limit theorem lesson.

Two ways to do it

Normal with continuity correctionstandard
P(X≤k)≈Φ ⁣(k+0.5−npnp(1−p))P(X \le k) \approx \Phi\!\left(\frac{k + 0.5 - np}{\sqrt{np(1-p)}}\right)
Match the binomial mean and variance, widen by 0.50.5, then use normal probability.
Exact binomial sumsoftware
P(X≤k)=∑j=0k(nj)pj(1−p)n−jP(X \le k) = \sum_{j=0}^{k} \binom{n}{j} p^j (1-p)^{n-j}
Exact, but long by hand; use the binomial probability steps or a calculator.

Steps

StepOn 100100 fair coin tosses, P(X≥60)P(X \ge 60)
1Check the conditions. Both expected counts, successes and failures, must be at least 1010.np=50≥10np = 50 \ge 10, n(1−p)=50≥10n(1-p) = 50 \ge 10 ✓
2Find the mean and standard deviation. μ=np\mu = np, σ=np(1−p)\sigma = \sqrt{np(1-p)}.μ=50\mu = 50, σ=25=5\sigma = \sqrt{25} = 5
3Apply the continuity correction. Each whole number kk is the bar from k−0.5k - 0.5 to k+0.5k + 0.5. Widen the event to cover its bars: X≥60→Y≥59.5X \ge 60 \to Y \ge 59.5, X≤45→Y≤45.5X \le 45 \to Y \le 45.5, X<60→Y≤59.5X < 60 \to Y \le 59.5, X=50→49.5≤Y≤50.5X = 50 \to 49.5 \le Y \le 50.5.P(X≥60)≈P(Y≥59.5)P(X \ge 60) \approx P(Y \ge 59.5)
4Standardize. z=(corrected boundary−μ)/σz = (\text{corrected boundary} - \mu)/\sigma.z=59.5−505=1.9z = \frac{59.5 - 50}{5} = 1.9
5Read Φ and report an approximation. Take left tails, right tails or differences as for any normal.1−Φ(1.9)≈1−0.97128≈0.02871 - \Phi(1.9) \approx 1 - 0.97128 \approx 0.0287 (exact 0.02840.0284)

Worked examples

Example 1 · at most k

k + 0.5
n=100n = 100, p=0.5p = 0.5, P(X≤45)P(X \le 45)
μ=50\mu = 50, σ=5\sigma = 5
P(Y≤45.5)P(Y \le 45.5), z=45.5−505=−0.9z = \frac{45.5 - 50}{5} = -0.9
include the bar at 4545
Φ(−0.9)≈0.18406\Phi(-0.9) \approx 0.18406
P(X≤45)≈0.1841P(X \le 45) \approx 0.1841
exact binomial: 0.18410.1841

Example 2 · exactly k

one bar
n=100n = 100, p=0.5p = 0.5, P(X=50)P(X = 50)
one bar, width 11
P(49.5≤Y≤50.5)P(49.5 \le Y \le 50.5): z=±0.1z = \pm 0.1
±0.5/5\pm 0.5 / 5
Φ(0.1)−Φ(−0.1)≈0.53983−0.46017\Phi(0.1) - \Phi(-0.1) \approx 0.53983 - 0.46017
P(X=50)≈0.0797P(X = 50) \approx 0.0797
exact binomial: 0.07960.0796

Example 3 · a range with p ≠ 0.5

two corrections
n=200n = 200, p=0.3p = 0.3, P(55≤X≤70)P(55 \le X \le 70)
μ=60\mu = 60, σ=42≈6.4807\sigma = \sqrt{42} \approx 6.4807
P(54.5≤Y≤70.5)P(54.5 \le Y \le 70.5): z1≈−0.8487z_1 \approx -0.8487, z2≈1.6202z_2 \approx 1.6202
widen both ends
Φ(1.6202)−Φ(−0.8487)≈0.94741−0.19802\Phi(1.6202) - \Phi(-0.8487) \approx 0.94741 - 0.19802
P(55≤X≤70)≈0.7494P(55 \le X \le 70) \approx 0.7494
exact binomial: 0.74690.7469
Compare with the exact value in the discrete distributions calculator.

See it happen

Draw the binomial PMF as bars of width 11 centered on the whole numbers. The normal curve with the same mean and spread runs through the tops of the bars, so the area of a group of bars is close to the area under the curve over the same stretch, and that stretch ends half a unit beyond the outer counts. Leaving out the half units cuts off half of each end bar. Compare the bars with the curve in the binomial distribution visualizer.

Common mistakes

No continuity correction

✗ P(X≤45)≈Φ(−1)≈0.1587P(X \le 45) \approx \Phi(-1) \approx 0.1587
✓ use 45.545.5: Φ(−0.9)≈0.1841\Phi(-0.9) \approx 0.1841, which matches the exact value

Correcting in the wrong direction

✗ P(X<60)≈P(Y≤60.5)P(X < 60) \approx P(Y \le 60.5)
✓ X<60X < 60 means X≤59X \le 59, so P(Y≤59.5)P(Y \le 59.5)

Using the variance as σ

✗ n=200n = 200, p=0.3p = 0.3: σ=42\sigma = 42
✓ σ=np(1−p)=42≈6.4807\sigma = \sqrt{np(1-p)} = \sqrt{42} \approx 6.4807

Ignoring the conditions

✗ n=20n = 20, p=0.05p = 0.05: normal with μ=1\mu = 1
✓ np=1<10np = 1 < 10; the histogram is skewed, so use the exact binomial or a Poisson