Confidence intervals from small samples

Confidence intervals from small samples

Spec SH4 · AQA A-Level Further Maths: Revision notes

Studying this course? Keep your progress in one place.

Key facts

  • For a small sample from a normal population with unknown variance, Xˉ−μs/n∼tn−1\dfrac{\bar X-\mu}{s/\sqrt n}\sim t_{n-1}.
  • The interval is xˉ±tνsn\bar x\pm t_{\nu}\dfrac{s}{\sqrt n} with ν=n−1\nu=n-1 degrees of freedom.
  • For a C%C\% interval read the column for tail probability 100−C2%\dfrac{100-C}{2}\%.
  • tt critical values are larger than zz values, so small-sample intervals are wider.
  • Check whether a claimed mean lies inside the interval, and answer in context.

Why the t-distribution is needed

With a small sample, ss can be well away from σ\sigma, so the standardised mean follows a tt-distribution with heavier tails.

For a normal population with unknown variance, Xˉ−μs/n∼tn−1\dfrac{\bar X-\mu}{s/\sqrt n}\sim t_{n-1}, the t-distribution with ν=n−1\nu=n-1 degrees of freedom.

It is symmetric about 00 with heavier tails than the standard normal, so its critical values are larger. As nn grows it approaches the normal distribution, which is why the normal can be used for large samples.

−4−3−2−112340.10.20.30.4xyt, 3 degrees of freedomstandard normal
Heavier tails: t with 3 degrees of freedom against the standard normal (dashed).

Compared with the standard normal, the t-distribution has

The interval

Go either side of xˉ\bar x by a critical value tνt_\nu times the standard error sn\frac{s}{\sqrt n}.

The critical value is the tt value from the table with ν=n−1\nu=n-1 and tail probability 100−C2%\frac{100-C}{2}\%.

For ν=7\nu=7: 95%95\% uses the 2.5%2.5\% column (2.3652.365) and 99%99\% uses the 0.5%0.5\% column (3.4993.499).

Find xˉ\bar x and ss first, using the unbiased variance. The standard error is sn\frac{s}{\sqrt n}.

012345678910x̄ = 5 (illustrative)
  • Interval for μ\muxˉ±tνsn\bar x\pm t_{\nu}\dfrac{s}{\sqrt n}
  • Unbiased variances2=1n−1(∑x2−(∑x)2n)s^2=\dfrac{1}{n-1}\left(\sum x^2-\dfrac{(\sum x)^2}{n}\right)

For a 90%90\% interval, which tail probability column do you use?

A small-sample interval

Using tt instead of zz widens the interval to allow for the small sample.

Chocolate bars: n=8n=8, xˉ=36.2\bar x=36.2, s=2.4s=2.4. The margin of error is tsnt\dfrac{s}{\sqrt n}, so the interval is xˉ±\bar x\pm margin of error.

The number line compares the tt interval with the too-narrow interval from z=1.96z=1.96.

3333.53434.53535.53636.53737.53838.539x̄ = 36.2

Worked example

Chocolate bars: n=8n=8, xˉ=36.2\bar x=36.2, s=2.4s=2.4. Find a 95%95\% confidence interval for the mean mass.

For ν=7\nu=7, which critical value gives a 99%99\% interval?

Width, assumptions and inference

The interval is wider for higher confidence, larger ss or smaller nn, and it assumes a normal population and a random sample.

A smaller nn also means a larger tt value. The method assumes a normal population, which matters most when nn is small.

If a claimed mean lies inside the interval, the data are consistent with the claim. If it lies outside, there is evidence at the matching level that the mean is different, and the interval shows whether it is higher or lower. Always answer in context.

02468101212345678910Degrees of freedom νCritical t value
Critical t value for a 95% interval (2.5% column) against degrees of freedom: smaller samples need a larger t, so the interval is wider.

Wider interval

  • Higher confidence level
  • Larger ss
  • Smaller nn (and larger tt)

Narrower interval

  • Lower confidence level
  • Smaller ss
  • Larger nn

A 95%95\% interval for a mean is (34.19, 38.21)(34.19,\ 38.21). A company claims the mean is 4040. What do you conclude?

Try an exam question

The times, in seconds, taken by a random sample of 1212 pupils to complete a puzzle have mean 8.48.4 and unbiased standard deviation s=1.5s=1.5. Assuming times are normally distributed, find a 90%90\% confidence interval for the mean time. (For ν=11\nu=11 the 5%5\% critical value is 1.7961.796.)

[4 marks]

That's the notes covered.

Carry on to the next subtopic.

Try a question

State the number of degrees of freedom of the tt-distribution used to construct a confidence interval for the population mean.

1 mark, marked against the mark scheme

Revision notes