Limitations and When the CLT Fails
What you'll learn: You'll discover when the Central Limit Theorem breaks down and can't rescue your analysis.
When the Magic Stops Working
The Central Limit Theorem feels like a statistical superpower—it promises that sample means become normally distributed regardless of the population shape. But this promise comes with conditions. When those conditions aren't met, the CLT fails, and assuming normality can lead you astray.
Key Situations Where the CLT Fails
1. Sample sizes that are too small with highly skewed distributions
If your population is extremely skewed (think income data with billionaires, or failure times), you need much larger sample sizes for the CLT to kick in. A sample of n=30 might work for mildly skewed data, but you might need n=100+ for heavily skewed distributions. The more extreme the skewness, the larger the sample you need.
2. When independence breaks down
The CLT assumes your observations are independent. If your data points are correlated—like daily stock prices, measurements from the same person over time, or clustered samples—the CLT's guarantees vanish. Dependent observations don't average out in the same predictable way.
3. Heavy-tailed or infinite-variance distributions
Some distributions (like Cauchy or certain Pareto distributions) have no finite variance. The CLT literally doesn't apply here because it relies on the population having a defined mean and variance. Averaging these samples won't converge to normal behavior.
The Bottom Line
Think of the CLT like training wheels that work on smooth pavement but fail on rocky terrain. Before trusting it, check your sample size against your distribution's skewness, verify independence, and confirm your population has finite moments.
Key Takeaway: The CLT requires sufficient sample size relative to skewness, independent observations, and finite variance—violate these, and you can't rely on approximate normality.