Lesson 8 of 8 · 13 min

Intervals for values and for the mean

Use x̄ and s to estimate an interval that covers a given proportion of individual values of a normal variable, and contrast it with the much narrower confidence interval for μ.

What you will be able to do

  • Use xˉ\bar{x} and ss as estimates of μ\mu and σ\sigma to find an approximate interval xˉ±zs\bar{x} \pm zs covering a proportion of values of a normal variable.
  • Compare it with the approximate confidence interval xˉ±zsn\bar{x} \pm z\frac{s}{\sqrt{n}} for μ\mu, and explain the factor n\sqrt{n} between their widths.
  • Choose the right interval for a question in context.

A fisheries officer measures 100 randomly caught flathead from an estuary: xˉ=42.0\bar{x} = 42.0 cm and s=5.0s = 5.0 cm, and the lengths look normally distributed. Two different people ask her two different questions:

  • A recreational angler asks: "What length is a typical fish? Between what lengths do most fish fall?"
  • A scientist asks: "What is the mean length of all flathead in the estuary?"

Both answers are intervals, and both use 95%, but they are very different intervals.

Where most values lie

If X∼N(μ,σ2)X \sim N(\mu, \sigma^2), about 95% of values lie in μ±1.96σ\mu \pm 1.96\sigma. Estimate μ\mu and σ\sigma by xˉ\bar{x} and ss:

xˉ±1.96s=42.0±9.8\bar{x} \pm 1.96s = 42.0 \pm 9.8

About 95% of flathead are roughly between 32.232.2 and 51.851.8 cm long.

Where the mean lies

The approximate 95% confidence interval for μ\mu is

xˉ±1.96sn=42.0±0.98\bar{x} \pm 1.96\frac{s}{\sqrt{n}} = 42.0 \pm 0.98

We are 95% confident that the mean length of all flathead is between 41.0241.02 and 42.9842.98 cm.

Same data, same 95%. The interval for individual lengths (grey) is 100=10\sqrt{100} = 10 times as wide as the confidence interval for the mean (orange).

Key idea

Two intervals, two questions

  • Individual values of a normal variable: about a proportion CC lie in xˉ±zs\bar{x} \pm zs. Its width does not shrink as nn grows; a bigger sample just estimates μ\mu and σ\sigma better.
  • The population mean: the confidence interval xˉ±zsn\bar{x} \pm z\frac{s}{\sqrt{n}}. Its width shrinks like 1n\frac{1}{\sqrt{n}}.

The ratio of their widths is n\sqrt{n}.

Worked example

An interval for fish lengths and an interval for their mean

  • Complex familiar
  • Technology active

A random sample of 100100 flathead from an estuary has mean length xˉ=42.0\bar{x} = 42.0 cm and standard deviation s=5.0s = 5.0 cm. Flathead lengths are approximately normally distributed.

(a) Estimate an interval that contains about 95% of the lengths of flathead in the estuary.

(b) Calculate an approximate 95% confidence interval for the mean length μ\mu.

(c) The legal minimum length is 3535 cm. Estimate the proportion of flathead under the legal length, correct to 3 decimal places.

4 steps

Try it yourself first, then check the working one step at a time.

Method

  1. Individual values: xˉ±zs\bar{x} \pm zs (needs a roughly normal variable).

  2. Population mean: xˉ±zsn\bar{x} \pm z\frac{s}{\sqrt{n}}.

  3. For a proportion of individuals, model X∼N(xˉ,s2)X \sim N(\bar{x}, s^2).

Watch out for

  • Using the confidence interval to say where most fish lengths are.

  • Using sn=0.5\frac{s}{\sqrt{n}} = 0.5 in part (c), which gives a proportion of essentially zero.

Times to complete a puzzle are normally distributed. A random sample of 5050 times has xˉ=30\bar{x} = 30 s and s=4s = 4 s. Use these to estimate an interval containing about 95% of all completion times. Give the endpoints correct to 2 decimal places.

From a random sample of 4949 values, Kai calculates an interval containing about 90% of individual values and an approximate 90% confidence interval for μ\mu. How many times wider is the first interval than the second?

Common mistake

The interval for values needs a normal population

The central limit theorem makes Xˉ\bar{X} approximately normal, but it does nothing for individual values. xˉ±1.96s\bar{x} \pm 1.96s covers about 95% of values only if the variable itself is roughly normal. For skewed data such as incomes it can be badly wrong, even when the confidence interval for μ\mu is fine.

Exam tip

Read which question is being asked

"Most customers", "a randomly chosen bag", "an individual" → interval for values, xˉ±zs\bar{x} \pm zs. "The mean", "on average across all", "the population mean" → confidence interval, xˉ±zsn\bar{x} \pm z\frac{s}{\sqrt{n}}. Say which one you are using and why.

Summary

  • For a normal variable, about a proportion CC of values lie in xˉ±zs\bar{x} \pm zs, with xˉ\bar{x} and ss estimating μ\mu and σ\sigma.

  • The confidence interval for μ\mu, xˉ±zsn\bar{x} \pm z\frac{s}{\sqrt{n}}, is n\sqrt{n} times narrower.

  • More data narrows the confidence interval but not the spread of individual values.

  • The interval for values needs a normal population; the confidence interval needs only a large sample.

Finished the lesson?

Try the 2 quick checks above and the lesson completes itself, or mark it complete now.

Lesson 8 of 80% read

Save your progress with a free account