In statistics we routinely treat the average of a large sample as if it were normally distributed — for example, when building confidence intervals for a mean. Why is that justified? Explain precisely **what quantity** becomes approximately normal as the sample size grows, state the theorem that justifies it along with its conditions, give an intuitive argument (or a proof sketch) for why it is true, and describe concrete situations where the normal approximation fails or is poor. ```hint Be precise about what converges "A large sample is approximately normal" is a common misstatement — the data's distribution never changes with $n$. The object that becomes normal is the **standardized sample mean** (equivalently, the standardized sum). Naming the Central Limit Theorem is the start, not the answer. ``` ```hint A route to "why" Consider the characteristic function (or moment generating function) of a standardized sum of independent variables: what does raising a second-order expansion to the $n$-th power converge to? Alternatively, think about why convolving many independent distributions keeps smoothing the result toward one universal shape. ``` ### Constraints & Assumptions - Assume independent, identically distributed draws unless you explicitly relax that. - Whiteboard-level rigor is expected: a precise statement plus a convincing sketch, not a measure-theoretic proof. - You should address both the "why it works" and the "when it breaks" sides. ### Clarifying Questions to Ask - Do you want the formal statement with conditions, the intuition for why it holds, or both? - May I assume i.i.d. sampling with finite variance, or should I discuss what happens when those assumptions are relaxed? - Are you also interested in *how fast* the approximation becomes good (rates of convergence), or just the limiting statement? ### What a Strong Answer Covers ```premium-lock What a Strong Answer Covers ``` ### Follow-up Questions - Give a distribution for which the sample mean is *never* approximately normal, no matter how large $n$ is, and explain why the theorem's hypotheses fail. - How fast does the approximation improve with $n$, and what feature of the underlying distribution controls that rate? - The observations in a time series are dependent. Does anything like the CLT still hold, and what has to be true for it to? - If you suspect the normal approximation is poor for your sample size, what would you do instead to get a confidence interval for the mean?