What the CLT Says
The Central Limit Theorem
The CLT says that if you take many samples and compute their means, those means will form a normal distribution - even if the original population is skewed, uniform, or weird. This is statistical magic.
Watch It Happen
Start with any population shape. As you increase sample size, the distribution of sample means gets more and more bell-shaped. By n=30, it's usually close enough to normal.
Key Properties
The CLT guarantees three things: (1) normality of x̄, (2) x̄ is centered at μ (unbiased), and (3) the spread decreases with √n. These enable all of inferential statistics.
Standard Error
SE vs SD
SD measures variability of individuals. SE measures variability of sample means. Because means average out individual variation, SE is always smaller than SD.
Why Square Root?
| n | √n | SE (if σ=10) | Reduction |
|---|---|---|---|
| 1 | 1.0 | 10.0 | - |
| 4 | 2.0 | 5.0 | 50% |
| 16 | 4.0 | 2.5 | 75% |
| 100 | 10.0 | 1.0 | 90% |
| 400 | 20.0 | 0.5 | 95% |
The √n relationship means diminishing returns: going from n=4 to n=16 (4×) halves SE. Going from n=100 to n=400 (4×) halves it again. Precision gets expensive.
Using s Instead of σ
In practice, we estimate SE using the sample standard deviation s. This adds uncertainty, which is why we use the t-distribution (heavier tails) instead of z for confidence intervals and hypothesis tests.
How Large is Large Enough?
n ≥ 30
When n < 30: Need approximately normal population OR use non-parametric methods
The n ≥ 30 rule is a rough guideline. For mildly skewed populations, n = 30 usually works. For heavily skewed or outlier-prone populations, you may need n > 100.
Population Shape Matters
| Population Shape | n Needed |
|---|---|
| Already normal | Any n works! |
| Symmetric, no outliers | n ≥ 15 |
| Mildly skewed | n ≥ 30 |
| Heavily skewed | n ≥ 50-100 |
| Extreme outliers possible | n ≥ 100+ or use median |
The more non-normal your population, the larger n you need for CLT to apply. With extreme skew or outliers, consider using the median (and bootstrap) instead of the mean.
How to Verify
Always look at your data. If you have n < 30 and the data looks heavily skewed, the CLT may not apply well enough. Consider log transformations or non-parametric methods.
Why CLT Matters
Enables All of Inference
The CLT is why we can use normal-based methods (z, t) even when data isn't normal. It's the foundation of confidence intervals, hypothesis tests, and most inferential statistics.
Income Distribution
Income is famously right-skewed (most people earn modest amounts, few earn millions). But if you sample 50 people and take the mean, that mean will follow a normal distribution. This lets us make inferences about average income.
The Big Picture
We only need large enough n for the CLT to make our sample mean approximately normal.
Then we can use normal-based methods (z, t, confidence intervals, p-values).
The CLT liberates us from needing to know (or assume) the population distribution. With sufficient n, sample means are normal regardless. This is why statistics works in practice.
Real-World Examples
Apply your knowledge to practical scenarios
Rolling Dice
A single die roll has a uniform distribution (each outcome 1-6 has probability 1/6). What happens to the distribution of the mean when we roll multiple dice?
Key Insights
- Even a completely flat (uniform) distribution produces normal means with enough observations
- You can simulate this yourself - roll dice and track means
- The transformation from uniform to normal is dramatic
Factory Quality Control
A factory produces bolts. Individual bolt lengths vary with σ = 0.5mm. Quality control samples n = 25 bolts each hour and measures the mean length.
Key Insights
- SE quantifies how much sample means vary
- Quality control uses CLT to set control limits
- An unusual sample mean signals a process change
Central Limit Theorem Quiz
Test your understanding • Passing score: 70%