Central Limit Theorem Calculator & Simulator

The Central Limit Theorem (CLT) says that if you take repeated random samples of size n from any population with a finite mean μ and standard deviation σ, the distribution of the sample means approaches a normal distribution as n grows — with mean μ and standard deviation σ / √n (called the standard error). The astonishing part is the phrase "any population." The original data can be uniform, heavily skewed, bimodal, or anything else with a finite mean and variance — the distribution of sample means still becomes bell-shaped.

That is exactly what the simulator above shows. The left histogram is a big sample of raw population values — clearly not normal for the Skewed or Bimodal choices. The right histogram is the distribution of 2,000 sample means (each averaging 30 draws). Notice how it centres on μ and takes on a bell shape even though the raw population histogram on the left doesn't. That is the CLT.

Almost every classical inference procedure you learn in a first stats course — confidence intervals for the mean, one-sample z-tests, t-tests, ANOVA, regression — leans on the assumption that the sample mean (or something like it) is approximately normally distributed. In real data the underlying population rarely is. The CLT is what lets those methods work anyway: as long as n is reasonably large, the SAMPLE MEAN behaves like it came from a normal distribution even when the raw data does not.

This is why textbooks recite the rule of thumb n ≥ 30. Treat it as a rough guideline, not a strict cutoff. For a roughly symmetric population, n = 10 or n = 15 is often more than enough. For a heavily skewed population like the Exponential option in the simulator, n = 30 is only a mediocre approximation and you may want n = 50 or n = 100 to be comfortable. The simulator lets you check for yourself instead of trusting the rule of thumb.

Move the sample size (n) slider and watch two things happen at once in the right-hand histogram:

• It gets narrower. The standard error is σ/√n, so doubling n only shrinks the spread by a factor of √2 ≈ 1.41 — that is why halving the standard error requires four times as much data, not twice as much.
• It looks more normal. At n = 2 or n = 3 the sampling distribution still carries fingerprints of the population's shape — a right skew for the Exponential, a soft double peak for the Bimodal. By n ≈ 30 the shape is close to bell-shaped; by n ≈ 100 it is visually indistinguishable from a normal curve.

The number of samples slider does something different — it does NOT invoke the CLT, it only sharpens the picture of the sampling distribution. Even a million samples of size n = 2 won't make the sampling distribution of the mean look normal for a skewed population. The variable that matters for the CLT is n.

• "The CLT says my data becomes normal." No. The population's shape does not change no matter how much data you collect. The CLT is a statement about the distribution of sample MEANS, not raw data.
• Treating n ≥ 30 as a hard rule. It is a rough guideline. Heavily skewed populations may need much larger n; roughly symmetric ones need much less.
• Confusing "number of samples" with "sample size". Drawing more samples of a tiny size does not invoke the CLT — it only gives you a sharper picture of a sampling distribution that hasn't yet converged.
• Assuming the CLT applies to any statistic. The classical CLT is about the sample mean. Sample maxima, minima, ranges, and other statistics have their own (often non-normal) limit distributions.
• Ignoring finite variance. The CLT needs a finite population variance σ². Heavy-tailed distributions like a Cauchy distribution have infinite variance and the CLT does not apply — no matter how large n is, the sample mean does not converge to a normal.

• Three population shapes (Uniform, Skewed / Exponential, Bimodal) chosen specifically to be visibly non-normal.
• Side-by-side histograms — raw population on the left, sample means on the right — so the CLT effect is obvious at a glance.
• Sample-size slider from n = 2 up to n = 200 with instant re-simulation so you can watch the sampling distribution narrow and become bell-shaped as n grows.
• Draw up to 5,000 samples in a single click — enough to give a smooth histogram of sample means without freezing the page.
• Empirical mean and SD of the sample means displayed alongside the theoretical values (μ and σ/√n) so you can verify the CLT numerically as well as visually.
• Re-run button reseeds the RNG so you can see how much the sample-means histogram wobbles from run to run — another intuition many stats textbooks skip.
• Show/hide step-by-step working that walks through the theorem, plugs in your population's μ and σ, computes σ/√n and compares it to the empirical SD from your simulation.
• Fast enough for 1,000+ draws without freezing the page — the entire loop is pure numeric JavaScript, no charting library overhead.