Central Limit Theorem Explained Simply (With a Dice Experiment)

The central limit theorem says that averages of random samples form a bell shaped curve, even when the original data doesn’t. The bigger your samples, the tighter and more normal that curve gets. We’ll test it with dice, then look at when the famous “n = 30” rule works and when it doesn’t.

What is the central limit theorem?

Take any population. Draw a random sample and work out its average. Repeat that many times, then plot all the averages.

Those averages will form a normal distribution, which is the classic bell curve. This holds even if the population itself is lopsided. That’s why we can use the normal curve for sample means, even when raw data looks nothing like it.

The first version came from de Moivre in 1733. He was studying the binomial distribution.

A dice experiment

One fair die is flat. Each number from 1 to 6 is equally likely. That’s about as far from a bell curve as you can get.

Now roll n dice and take the average. Do that 20,000 times. I ran this on a computer. Here’s what happened to the spread of the averages.

The average of the averages stayed near 3.5 every time. That’s the true mean of a die. Only the spread shrank, and the shape grew bell like. The formula predicted every row.

The formula in plain words

The spread of the sample averages has a name. It’s called the standard error. Here’s the formula:

standard error = σ ÷ √n

The σ is the standard deviation of the population. The n is your sample size.

This tells you something useful. Going from n = 100 to n = 400 cuts the error in half. To halve your error, you need four times the data. So precision gets more expensive as you go.

Conditions

The classic version needs four things:

1. The samples are random.

2. The observations are independent of each other.

3. The population has a finite mean and variance.

4. The sample is large enough.

Some versions loosen these rules. For a first course, stick to the four above.

Is n = 30 really enough?

Many books say a sample of about 30 is enough. It’s a handy rule of thumb. It isn’t a law.

I also tested a strongly skewed population with a long right tail. Skew measures how lopsided a curve is. Zero means perfectly symmetric.

At n = 30, some lopsidedness remains. At 100, it’s much closer to symmetric. So 30 is a starting point, not a finish line.

Central limit theorem vs law of large numbers

People mix these up. They answer different questions.

Both are about large samples. The first tells you where. The second tells you how the averages spread around that spot.

Why it matters

Many tools assume the average behaves normally. That includes confidence intervals and common tests like the t test, ANOVA, and regression. The theorem is a big reason those tools work even when raw data isn’t normal.

What it doesn’t say

• Big datasets don’t become normal. Your raw data keeps its shape.

• It doesn’t shrink the spread of the data. It shrinks the spread of the average.

• It doesn’t fix a biased sample. If the sample isn’t random, you can still be wrong.

When it breaks

Some distributions have heavy tails and no finite variance. The Cauchy distribution is the classic case. Data collected over time can also be linked from one point to the next, which can weaken the result. And a strongly skewed population with a small sample may not look normal yet.

Try it yourself

You can run a quick test with Python.

python

import random, statistics as st
n, reps = 30, 20000
means = [sum(random.randint(1, 6) for in range(n)) / n for in range(reps)]
print(st.mean(means), st.pstdev(means))

You should see about 3.5 and about 0.31. Change n and watch the spread change.

Want to see these ideas in real studies? Try Literature Help to find papers. You can also use research paper summaries to check what a study found before reading it in full.

Final thoughts

The central limit theorem is simple at heart. Averages of random samples look normal, and the spread shrinks as samples grow. Run the code above with n = 5, 30, and 200 and watch the bell shape appear. For more beginner topics, check our more guides.

FAQs

Does the population have to be normal?

No. That’s the point. If it already is normal, any sample size works.

Is the standard error the same as the standard deviation?

No. The standard deviation measures the spread of the data. The standard error measures the spread of the sample average.

Is 30 always enough?

No. Mildly skewed data is fine near 30. Strongly skewed data needs more.

Why is it called “central”?

It sits at the heart of probability theory and statistics. It isn’t about the middle of the curve.

What is Canyam?

Canyam lets you search more than 65 million research papers and read short AI summaries of them.