Visual Tools
Calculators
Tables
Mathematical Keyboard
Converters
Other Tools

How to Use the Central Limit Theorem

The sum or the average of many independent, identically distributed values is approximately normal, whatever the shape of one value. Scale the mean and standard deviation to the sum or the average, standardize, and read the normal CDF.

Recognize the task

A question about a total or an average of many independent repeats: the mean lifetime of 36 bulbs, the total weight of 50 packages, the share of heads in 400 flips. You know the mean μ\mu and standard deviation σ\sigma of one term, but not the exact distribution of the total. Test: is nn large (often n≥30n \geq 30, less for symmetric data) and are the terms independent? Then use a normal with the scaled parameters.
The theorem is stated in the lesson section Formal Statement of the Central Limit Theorem.

Two ways to do it

For a sample meanaverage
Xˉ≈N ⁣(μ, σ2n)\bar{X} \approx N\!\left(\mu,\ \frac{\sigma^2}{n}\right)
The average keeps the mean μ\mu and its standard deviation shrinks to σn\frac{\sigma}{\sqrt{n}}.
For a sumtotal
Sn≈N ⁣(nμ, nσ2)S_n \approx N\!\left(n\mu,\ n\sigma^2\right)
The total has mean nμn\mu and standard deviation σn\sigma\sqrt{n} (variances of independent terms add).

Steps

StepOn the mean lifetime of 36 bulbs, μ=1000\mu = 1000 h, σ=120\sigma = 120 h, P(Xˉ<970)P(\bar{X} < 970)
1Identify one term, the number of terms, and sum or mean. The terms must be independent with the same distribution.one lifetime; n=36n = 36; sample mean
2Write the mean and standard deviation of one term. From the problem, or from the expected value and standard deviation pages.μ=1000, σ=120\mu = 1000,\ \sigma = 120
3Scale them to the sum or the mean. Mean: μ\mu and σn\frac{\sigma}{\sqrt{n}}. Sum: nμn\mu and σn\sigma\sqrt{n}.σXˉ=12036=20\sigma_{\bar{X}} = \frac{120}{\sqrt{36}} = 20
4Standardize. Subtract the scaled mean and divide by the scaled standard deviation (standardization).z=970−100020=−1.5z = \frac{970 - 1000}{20} = -1.5
5Read the normal CDF. Use Φ(z)\Phi(z) for "below" and 1−Φ(z)1 - \Phi(z) for "above", as in normal probability.P(Xˉ<970)≈Φ(−1.5)≈0.0668P(\bar{X} < 970) \approx \Phi(-1.5) \approx 0.0668

Worked examples

Example 1 · a sample mean

average
n=36n = 36, μ=1000\mu = 1000, σ=120\sigma = 120: σXˉ=20\sigma_{\bar{X}} = 20
divide by 36=6\sqrt{36} = 6
z=970−100020=−1.5z = \frac{970 - 1000}{20} = -1.5
standardize
P(Xˉ<970)≈Φ(−1.5)≈0.0668P(\bar{X} < 970) \approx \Phi(-1.5) \approx 0.0668
rounded to 4 places

Example 2 · a total weight

sum
50 packages, μ=2\mu = 2 kg, σ=0.5\sigma = 0.5 kg
E[S]=100E[S] = 100, σS=0.550≈3.5355\sigma_S = 0.5\sqrt{50} \approx 3.5355
scale for a sum
z=105−1003.5355≈1.4142z = \frac{105 - 100}{3.5355} \approx 1.4142
standardize
P(S>105)≈1−Φ(1.4142)≈0.0786P(S > 105) \approx 1 - \Phi(1.4142) \approx 0.0786
upper tail, 4 places

Example 3 · a share of heads

proportion
400 flips, one flip: μ=0.5\mu = 0.5, σ=0.5\sigma = 0.5
0-1 variable, p(1−p)\sqrt{p(1 - p)}
σp^=0.5400=0.025\sigma_{\hat{p}} = \frac{0.5}{\sqrt{400}} = 0.025
share is a mean
z=0.55−0.50.025=2z = \frac{0.55 - 0.5}{0.025} = 2
standardize
P(p^>0.55)≈1−Φ(2)≈0.0228P(\hat{p} > 0.55) \approx 1 - \Phi(2) \approx 0.0228
upper tail, 4 places
Get Φ(z)\Phi(z) in the continuous distributions calculator. For counts, see normal approximation, which adds a continuity correction.

See it happen

Adding independent values smooths out their individual shapes: the peaks and gaps of one value fill in, and the total piles up around nμn\mu in a bell shape. The spread of the total grows only like n\sqrt{n}, because highs and lows partly cancel, so the average gets tighter as nn grows. Overlay the matching bell curve in the normal distribution visualizer.

Common mistakes

Using the standard deviation of one term

✗ z=970−1000120=−0.25z = \frac{970 - 1000}{120} = -0.25
✓ use σn=20\frac{\sigma}{\sqrt{n}} = 20: z=−1.5z = -1.5

Scaling a sum by n instead of the square root of n

✗ σS=50⋅0.5=25\sigma_S = 50 \cdot 0.5 = 25
✓ σS=0.550≈3.5355\sigma_S = 0.5\sqrt{50} \approx 3.5355

Applying it to one observation

✗ one bulb is normal by the CLT: P(X<970)≈Φ(−0.25)P(X < 970) \approx \Phi(-0.25)
✓ the CLT is about sums and averages; one lifetime keeps its own distribution
See when it applies in the CLT conditions.