Central Limit Theorem

Learning Objective

Why This Matters

Every time Netflix tests a new thumbnail or Spotify tweaks a recommendation algorithm, they sample a few thousand users out of hundreds of millions and compute an average -- watch time, skip rate, session length. The Central Limit Theorem is the reason that average from a sample can be trusted to reflect the full audience. Without it, A/B testing wouldn't work, polling margins wouldn't exist, and no clinical trial could ever reach a conclusion.

How to Use This Simulation

  1. Pick a population shape on the left -- the histogram shows the underlying data, not the sampling distribution.
  2. Choose a sample size n. This is how many values are pulled in each individual sample.
  3. Click Draw 10 or Draw 100 and watch sample means stack up on the right. The red normal curve overlay appears once you've drawn at least 50 samples.
  4. Try the same population at n = 5 and n = 30. The change in shape is the Central Limit Theorem at work.

The shape on the left is the population. The CLT predicts the sampling distribution on the right will look normal regardless of this shape.

Population

The underlying data. Stays the same shape no matter how many samples you draw.

Sampling distribution of x̄

Each dot is one sample mean. The CLT predicts this becomes normal as n grows.

Most recent sample
Draw a sample to see its values plotted here.
Population mean μ --
Population SD σ --
Theoretical SE σ/√n --
Samples drawn 0
Mean of x̄'s -- should approach μ
SD of x̄'s -- should approach σ/√n
Building... Lights up when observed values match theory within 5%

What's Happening

Quick Check

You're running a CLT simulation. The sampling distribution looks lumpy and skewed. Which change is most likely to make it look more normal?

Try This

Set the population to Right-skewed and n = 5. Click Draw 100 Samples. Sketch or describe the shape of the sampling distribution.

Now switch n to 50, click Clear Samples, and Draw 100 Samples again. What shape changed and what stayed the same? Write one sentence explaining which control drove the change: sample size n, or number of samples drawn.

Set the population to Right-skewed. The population standard deviation σ is shown above the left histogram. Set n = 30. Before drawing any samples, calculate the theoretical standard error σ/√30 by hand.

Now Draw 100 Samples, then Draw 100 more (200 total). Compare your calculation to the SD of x̄'s reported in the stats panel. How close are they? Explain why they aren't identical and what would make them closer.

A factory's bolt-length machine has a known standard deviation of σ = 0.05 mm from years of production records. The target bolt length is 10.00 mm. A quality control inspector samples 64 bolts and finds x̄ = 9.92 mm.

  1. If the machine is producing bolts at the target length, what mean and standard error describe the sampling distribution of x̄?
  2. How many standard errors away from 10.00 is the observed 9.92?
  3. Based on (b), should the inspector flag this batch for investigation? Defend your answer in one sentence using what you know about how often sample means land far from the true mean by chance.

Instructor Notes

Teaching Notes

The most powerful moment in this simulation is the side-by-side comparison of the same population at n = 5 versus n = 30. Have students predict what will happen before they change n. Many will say "the sampling distribution will spread out" or "it will get more skewed." Watching it tighten and snap into a normal shape against their prediction is what makes the CLT click.

The Bimodal population is the most diagnostic. A bimodal population looks nothing like a normal distribution, and students often predict the sampling distribution will also be bimodal. Drawing samples at n = 30 from the bimodal population produces a clean bell curve -- this is the demonstration that earns the CLT its central role in inference.

Common Student Errors

Discussion Questions

Exam Connection

Typical exam items provide a population mean μ and standard deviation σ and ask students to find P(x̄ < some value) for a given sample size n. The standardization step uses (x̄ − μ) / (σ/√n), not (x − μ) / σ. Students who haven't internalized that the standard error replaces σ in CLT calculations will systematically get these wrong. The simulation's stats panel shows both values explicitly to reinforce the distinction.