A distribution has heavy tails (kurtosis much greater than 3). A sample of n = 100 is taken from this distribution. The CLT predicts that the sampling distribution of x-bar will be approximately normal. How much confidence should we have in this approximation compared to a sample from a normal population?
- ANo confidence — CLT does not apply to heavy-tailed distributions.
- BThe same confidence — CLT always gives exact normality for n = 100.
- CLess confidence — heavy-tailed populations require larger samples for the CLT approximation to be accurate.Correct
- DMore confidence — heavy tails actually make the CLT more accurate.
Why C is right
While the CLT applies broadly, the speed of convergence to normality depends on the population's distribution. Heavy-tailed distributions (high kurtosis) converge more slowly than distributions close to normal. For heavy-tailed populations, n = 100 may not provide as accurate a normal approximation as for a symmetric, light-tailed distribution. More observations are needed for heavy-tailed cases.
Why the other 3 choices miss
- A
- Incorrect because the CLT does apply to heavy-tailed distributions (as long as the distribution has finite variance). It just requires a larger sample for a good approximation compared to distributions closer to normal.
- B
- Incorrect because the CLT provides an approximation, not exact normality, for non-normal populations. The accuracy of the approximation depends on both n and the population's shape. Heavy tails require larger n.
- D
- Incorrect because heavy tails actually slow down convergence to normality in the sampling distribution. More extreme observations in heavy-tailed distributions introduce more variability in x-bar.