Image: Metro Centric, Wikimedia Commons, CC BY 2.0
A sample statistic is never the truth about the population, so the honest report is a number with a range around it. This unit builds that range: what makes an estimate good, why the sample variance divides by n - 1, the confidence interval for a mean and what 95 per cent confidence promises, Student's t distribution for small samples, the interval and the margin of error of a poll, planning a sample size, paired data, and the comparison of two groups.
Preview: the beginning of the unit
A poll says that 58 per cent of adults disapprove of their president, a laboratory reports that a drug adds 1.6 hours of sleep, and a survey puts the mean height of adult men at 172 cm. None of these numbers is the truth about the population. Each is a sample statistic, and Unit 5 showed that a statistic moves from sample to sample. The honest way to report it is therefore a number with a range around it, and this unit explains how the range is built, what it promises, and what it does not. The tools are the standard error of Unit 5, the normal curve of Unit 4, and one new distribution that a brewer worked out in 1908 because his samples were too small for the normal curve.
1 Estimates and their precision
Unit 1 separated a parameter, a number that describes the population, from a statistic, a number computed from the sample. When a statistic is used to guess a parameter it is called an estimator, and the value it takes in one particular sample is an estimate. The sample mean \(\bar{x}\) estimates the population mean \(\mu\), the sample proportion \(\hat{p}\) estimates the population proportion \(p\), and the sample standard deviation \(s\) estimates \(\sigma\). The hat on \(\hat{p}\) is the general sign for “estimate of”, and from now on a parameter with a hat always means the estimate computed from the data.
Two questions decide whether an estimator is a good one. The first is whether it aims at the right target. An estimator is unbiased when its expected value over all possible samples equals the parameter, so that it is right on average even though every single estimate is a little off. Unit 5 derived that \(\mathrm{E}[\bar{X}] = \mu\), which says that the sample mean is unbiased for the population mean, and the same argument gives \(\mathrm{E}[\hat{p}] = p\) for a proportion. The sample variance is unbiased too, and this is the promise Unit 2 made when it divided by \(n - 1\). The derivation below keeps that promise.
The second question is how far a single estimate is likely to be from the parameter, and the answer is the standard error: the standard deviation of the estimator over all possible samples, written \(\mathrm{SE}\). Unit 5 derived the standard error of the sample mean, \(\sigma/\sqrt{n}\), from the rule that variances of independent observations add. In practice \(\sigma\) is unknown, and \(s\) takes its place.
The full unit is part of the program
Want to see a complete unit first? Unit 5, Sampling and the Central Limit Theorem, is free to read.