1.9 Summary
- Randomness creates variation, and variation is the raw material of statistics. Where nothing varies there is no distribution, no uncertainty, and nothing to infer.
- Descriptive statistics summarises the data in hand. Inferential statistics uses that data to make claims about a population it did not observe, and reports how uncertain those claims are.
- Different samples give different answers. That is sampling variation — unavoidable, and not a defect in the data.
- A quantity whose value changes from one repetition to the next is a random variable, written \(X\). Its observed value is a realisation, written \(x\). The sample mean is \(\bar{X}\) before the sample is drawn and \(\bar{x}\) after.
- \(E(X) = \sum_i x_i P(X = x_i)\) is the expected value — a long-run average, not a value you expect to see. For a population draw, \(E(X) = \mu\).
- \(\operatorname{Var}(X) = E[(X - \mu)^2]\) measures spread, and \(\sigma = \sqrt{\operatorname{Var}(X)}\) is the standard deviation.
- Parameters (\(\mu\), \(\sigma^2\), \(\sigma\)) describe the population and are fixed but unknown. Statistics (\(\bar{x}\), \(s^2\), \(s\)) are computed from a sample and change with every sample. Inference uses the second to learn about the first.
- The distribution of a statistic across repeated samples is its sampling distribution. Every statistic has one, not only the mean.
- Law of Large Numbers: as \(n\) grows, the sample mean becomes increasingly likely to lie close to \(\mu\). It does not say any particular large sample is correct, and it offers no comfort to a gambler.
- Central Limit Theorem: for sufficiently large \(n\), the sampling distribution of \(\bar{X}\) is approximately normal whatever the shape of the population. The rule \(n \geq 30\) is a rule of thumb, not a guarantee — a heavily skewed population needs far more.
- No formula warns you when the sample is drawn from the wrong population, drawn non-randomly, measured badly, or asked a causal question it cannot answer.
Taken together, these amount to a single change of perspective. We stopped asking what a particular sample says, and began asking how a statistic behaves across the many samples that might have been drawn. Everything that follows — estimation, confidence intervals, hypothesis testing, regression — rests on that shift.