On this concept: Formula

Not sure which lesson to do first? Find your level and get a plan →

Math · Statistics & Probability · Grade 9-12

Central Limit Theorem

The Central Limit Theorem says that for a large enough sample size (n≥30 rule of thumb), the sampling distribution of the sample mean is approximately normal — centered at μ with spread σn — regardless of the population's shape.

Xˉ∼N(μ,  σ2n), so the SD of Xˉ is σn

Quick Answer

The Central Limit Theorem says that for a large enough sample size (n≥30 rule of thumb), the sampling distribution of the sample mean is approximately normal — centered at μ with spread σn — regardless of the population's shape. Use it when the data themselves are skewed or lumpy (incomes, wait times, die rolls) but the question is about the average of many of them. It lets you use z-scores and the normal table for xˉ, which is what makes most confidence intervals and hypothesis tests work.

Why This Matters

Real data are rarely bell-shaped: incomes are skewed, wait times pile up near zero, die rolls are flat. The CLT says averages of many such values are nearly normal anyway, so one set of normal-curve tools works for almost any population. Without it, every population shape would need its own method; misapplied, it tempts people to treat raw skewed data, or tiny samples, as if they were normal.

Intuitive Explanation

Roll one die: each face from 1 to 6 is equally likely, a flat distribution with no peak. Roll two dice and average them. An average of 3.5 can happen 6 ways (1 and 6, 2 and 5, and so on), but an average of 1 happens only one way (1 and 1), so the middle already bulges. Average 10 dice and averages near 1 or 6 almost never happen; the histogram of averages looks like a bell centered at 3.5.

That is the theorem at work. Whatever the population's shape, an average mixes many independent values whose highs and lows tend to offset, so averages pile up near μ in a normal shape with standard deviation σn. One die has σ≈1.71, so averages of 36 rolls are approximately normal with spread about 1.716=0.285.

How large is large enough depends on the population. A symmetric population such as die rolls gives nearly normal averages by n=10; a strongly skewed one, like household incomes, may need far more than 30. The rule of thumb n≥30 is a starting point, not a guarantee.

The theorem says nothing about individual values: 1,000 die rolls are still flat, and only their averages are bell-shaped. It also differs from the law of large numbers, which says xˉ gets close to μ as n grows; the CLT describes the shape of the scatter around μ along the way.

Think of it like this. Averaging works like mixing paint from randomly chosen cans. One can might be any shade, but a mix of 30 cans almost always comes out a middling color, since an extreme mix needs nearly every can to be extreme. The mixes cluster around the average shade whatever the cans looked like; mixes of only two or three cans still vary a lot.

Core idea. The CLT says sample means become approximately normal for large n, whatever the population's shape.


When to Use

Use the CLT when a question involves the mean of a sample (or a sum or proportion) from a population that is not normal or of unknown shape, and the sample is large, about 30 or more. It lets you find probabilities for xˉ with z=xˉ−μσ/n. If the question is about one value from a skewed population, the CLT doesn't apply. If the population is already normal, xˉ is normal for any n and the theorem isn't needed.

Tip. Ask: Am I claiming the distribution of a sample MEAN is approximately normal because the sample is large?

How to Recognize It

Check three things before using the theorem: whether the question is about a mean or a sum rather than one value, how skewed the population is, and how big the sample is.

  1. Is the probability about the average or total of many values, not about one value?

    Yes is where the CLT applies. A question about one customer's wait from a skewed population must use that population's own shape; the theorem says nothing about single values.

  2. Is the population already normal?

    Yes means you do not need the CLT: the sample mean is exactly normal for every n, even n=3. The theorem earns its keep when the population is skewed, lumpy or of unknown shape.

  3. Is the sample large for how skewed the population is, about 30 or more for moderate skew?

    Yes means a normal model for xˉ is reasonable. Strong skew, such as incomes or insurance claims, may need n in the hundreds, and with n=5 from a skewed population the CLT does not justify a normal model.

  4. Does the question ask only whether the sample mean ends up close to μ, with no probability attached?

    Yes is the law of large numbers: averages settle toward μ as n grows. The CLT goes further and describes the bell-shaped spread around μ, which you need for z-scores and probabilities.

Central Limit Theorem compared with Sampling distribution, Normal distribution, and Law of large numbers

All four rows describe sample means and are easy to blur. Ask what is claimed: the distribution of xˉ itself (sampling distribution), its bell shape for large n (CLT), xˉ approaching μ (law of large numbers), or a variable that is normal already (normal distribution).

Central Limit Theorem

Meaning
Use this when you have the mean of a large sample and want to treat its sampling distribution as normal despite a non-normal population.
Key test
Am I claiming the distribution of a sample MEAN is approximately normal because the sample is large?
Formula
Xˉ∼N(μ,  σn)
Example
A skewed population has μ=50, σ=12. For n=36, describe the sampling distribution of xˉ.

Sampling distribution

Meaning
The general concept of a statistic's distribution; CLT is the result about its SHAPE.
Key test
Use when defining the distribution of a statistic, not asserting it's normal.
Example
Histogram of all sample means

Normal distribution

Meaning
The specific bell curve; CLT explains WHY sample means follow it.
Key test
Use when working with an already-normal variable, not the large-n justification.
Formula
N(μ,σ)
Example
Heights, already bell-shaped

Law of large numbers

Meaning
Says the sample mean CONVERGES to μ; CLT says how it's DISTRIBUTED around μ.
Key test
Use when arguing the mean gets close to the truth, not its bell shape.
Example
xˉ→μ as n→∞

Formula & Notation

Xˉ∼N(μ,  σ2n), so the SD of Xˉ is σn
If X1,…,Xn are i.i.d. with mean μ and variance σ2, then Xˉn−μσ/n→dN(0,1) as n→∞

How to read it: σn is called the standard error of the mean.

Worked Examples

Example 1 — Mean of a large sample

Easy

A skewed population has μ=50, σ=12. For n=36, describe the sampling distribution of xˉ.
Show the steps and the answer
  1. Check the sample size: n=36≥30.

    The population is skewed, so a normal shape for xˉ has to come from the CLT, and 36 meets the rule of thumb.

  2. Center: the mean of xˉ is μ=50.

    Sample means center on the population mean whatever the population's shape.

  3. Spread: σn=1236=126=2.

    Averaging 36 values cuts the spread by a factor of 6.

  4. Shape: approximately normal, so Xˉ≈N(50,2).

    Here the second number is the standard deviation, so about 95% of sample means fall within 50±4.

Answer: Xˉ≈N(50,2), approximately normal

Takeaway: The CLT supplies the normal shape; the center and spread come from μ and σn.

Example 2 — Small sample, skewed data

Standard

From the same skewed population, you take ONE sample of n=4. Is xˉ approximately normal?
Show the steps and the answer
  1. Compare n=4 with the rule of thumb of about 30.

    With only 4 values, one value from the long tail can drag the average far, so the skew survives in the averages.

  2. Note what still holds: the mean of xˉ is 50 and its standard error is 124=6.

    The center and spread formulas work for any n; only the normal shape needs a large sample.

  3. Conclude that normal-table probabilities for xˉ would be unreliable here.

    You would need the population's actual shape, or a much larger sample, before using z-scores.

Answer: No — too small for the CLT to apply

Takeaway: Small samples from skewed populations keep some of the skew; the CLT's normal shape is a large-sample result.

Example 3 — Find the mistake: the data do not become normal

Application

Wait times at a clinic are right-skewed with μ=20 min and σ=15 min. Nia has data on 100 patients and writes: "By the CLT, wait times are normal, so the chance one patient waits over 50 min is P(Z>2)≈0.023." Find the mistake and correct it.
Show the steps and the answer
  1. Ask what the CLT describes: the distribution of xˉ, the average of the 100 waits.

    Collecting more data does not change the shape of individual waits; they stay right-skewed.

  2. Test Nia's normal model on individuals: it gives P(wait<0)=P(Z<0−2015)=P(Z<−1.33)≈0.09.

    A model that predicts 9% of patients wait a negative time cannot describe single patients.

  3. Apply the CLT where it belongs: σxˉ=15100=1.5 min.

    With n=100 the average wait is close to normal with this spread.

  4. Example: P(xˉ>23)=P(Z>23−201.5)=P(Z>2)≈0.023.

    The normal table now answers a question about the average, which is what the theorem covers.

Answer: One patient's chance of waiting over 50 min can't come from a normal model; the CLT covers the average instead, for example P(xˉ>23)≈0.023 for the mean of 100 waits.

Takeaway: The CLT makes averages approximately normal, not the data; individual skewed values keep their own distribution.

Where You'll See It

  • Why casinos can budget

    A single $1 bet on red in American roulette is unpredictable, but over 100,000 such bets the casino's average gain per bet is approximately normal, with mean 238≈0.053 dollars and standard error about 1100000≈0.003 dollars. So about 95% of the time its total gain lands between roughly $4,600 and $5,900.

  • Testing batches of batteries

    Battery lifetimes are often right-skewed, but a factory tests the average life of 50 batteries per batch. By the CLT that average is close to normal, so if lifetimes have μ=10 h and σ=4 h, a batch average below 10−2×450≈8.9 h signals a problem.

  • Polls built from yes-or-no answers

    Each voter's answer is a 0 or a 1, about as far from bell-shaped as data can be. A poll's percentage is the average of those 0s and 1s, so with 1,000 voters the CLT makes it approximately normal, which is where the familiar plus-or-minus 3 points comes from.

Common Mistakes

  1. The slip: Claiming the raw data becomes normal

    Instead: the CLT is about the distribution of the MEAN, not the individual values.

  2. The slip: Applying it with a tiny sample from a very skewed population

    Instead: n≥30 is a rule of thumb; heavy skew needs even larger n.

  3. The slip: Forgetting the spread shrinks

    Instead: the sample mean's SD is σn, not σ.


Mini Practice

Try these on your own. Tap Reveal when you want to check.

  1. Wait times are right-skewed with μ=8 min and σ=6 min. For samples of n=36, describe the sampling distribution of xˉ.

    Hint: Give its shape, center and spread.

  2. For the same wait times (μ=8, σ=6, n=36), about what is P(xˉ>10)?

    Hint: Standardize with the standard error, then use the 68-95-99.7 rule.

  3. One die roll has mean 3.5 and SD about 1.71. For the average of 49 rolls, find the mean and SD, and a range that holds about 95% of such averages.

    Hint: Divide the SD by 49, then go 2 SDs each way.

  4. Bags of chips weigh μ=200 g with σ=10 g. A box holds 25 independent bags. Find the mean and SD of the box's total weight.

    Hint: For a sum, the SD grows with n.

  5. A question asks whether the proportion of heads in 10,000 fair flips will be close to 0.5. Is that the law of large numbers or the central limit theorem?

    Hint: Is the question about where the average settles or about the shape of its variation?

  6. Kai samples 100 incomes from a right-skewed population and says: "Since n≥30, the 100 incomes in my sample will form a bell curve." Correct him.

    Hint: Is the theorem about individual values or about sample means?

Want the full set?

50 practice questions for this concept — free to try, every one with a complete worked solution showing the why, not just the answer.

Frequently Asked Questions

Why is 30 the magic number?

It is not magic, just a rule of thumb that works for populations with mild skew. For a symmetric population, sample means look normal with nn as small as 10 or so. For a very skewed one, such as lottery winnings, where almost everyone wins nothing and a few win millions, even n=100n=100 can leave the distribution of xˉ\bar{x} visibly lopsided. The more skewed the population, the larger the sample you need.

Why do so many things in nature, like heights, look normal?

The CLT suggests a reason. A height is shaped by many small, roughly independent influences, such as hundreds of genes and many factors of diet and health, which add together. A sum of many small independent effects tends toward a normal shape whatever each effect looks like on its own. Quantities driven by one big factor, or by effects that multiply, such as incomes, tend to be skewed instead.

My textbook writes N(μ,σ2n), but this lesson writes N(μ,σn). Which is right?

Both, because they use different conventions for the second number. Many textbooks put the variance there, σ2n\frac{\sigma^2}{n}; others, including many AP Statistics books, put the standard deviation, σn\frac{\sigma}{\sqrt{n}}. The distribution is the same. Check which convention your source uses before plugging in, since squaring by mistake, or forgetting to take the square root, changes the zz-score you get.

Does the CLT work for anything other than means?

Yes, for sums and for proportions. A sum of nn values is just nn times their mean, so it is approximately normal with mean nμn\mu and standard deviation n σ\sqrt{n}\,\sigma. A sample proportion is a mean of 0s and 1s, which is why the binomial distribution looks like a bell curve when nn is large. The theorem does not cover statistics like the sample maximum.

Learning Path

Central Limit Theorem

You are here

You need the Sampling Distribution (center μ, spread σn) and the Normal Distribution first. The CLT supplies the missing piece, the shape: for large n the sampling distribution of xˉ is approximately normal, whatever the population looks like. Next, Confidence Intervals and Hypothesis Testing both rely on it to turn a sample mean into probabilities from the normal table.

See Also