Statistics › Section H · Confidence intervals › Topic H.2
Interval for a mean with known sigma
A point estimate names one number; an interval tells how far that number deserves to be trusted. This topic builds the confidence interval for a mean in the rare, clean case where the population's spread is known in advance - the prototype for every interval that follows.
From point to interval
The battery estimate of 412 hours came with a standard error, but a reader still has to ask what range of true means the data allow. The confidence interval for a population mean adds and subtracts a margin from the sample mean: the estimate, plus an honest admission of its wobble. This topic builds the version for the case where the population's spread \(\sigma\) is known in advance - from long records or engineering specifications - which makes it the prototype for every interval in the rest of the course.
The formula and its margin
When \(\sigma\) is known, the interval is
The plus-minus piece, \(z^*\frac{\sigma}{\sqrt{n}}\), is the error bound for the mean (EBM): the standard error \(\sigma/\sqrt{n}\) from the CLT chapter, stretched by a critical value \(z^*\) chosen to match the confidence level. The interval's centre is the estimate \(\bar{x}\); its width is pure arithmetic on \(z^*\), \(\sigma\) and \(n\).
Confidence levels and critical values
A confidence level CL is the central area of the standard normal reserved for capturing \(\mu\); the remaining \(\alpha = 1 - \text{CL}\) splits equally between the two tails, \(\alpha/2\) each. The critical value \(z^*\) is the z-score leaving \(\alpha/2\) beyond it: 90% confidence puts 5% in each tail and \(z^* = 1.645\); 95% puts 2.5% each side and \(z^* = 1.96\); 99% puts 0.5% each side and \(z^* = 2.576\). Higher confidence demands a taller \(z^*\), and these three values are worth knowing by heart.
What "confident" means
The confidence statement is about the method, not the parameter. \(\mu\) is a fixed number; the interval is what varies from sample to sample. "95% confident" means that if the study were repeated with new random samples, about 95% of the resulting intervals would capture \(\mu\) - and about 5% would miss. Any one interval either contains \(\mu\) or does not, with no way to know which: the 95% is the procedure's long-run capture rate, not a probability attached to one particular interval. The distinction feels philosophical until a report misquotes it - then it is simply an error.
Worked example: coffee deliveries
A coffee shop's delivery times have a known \(\sigma = 4\) minutes from years of data; a random \(n = 36\) orders average \(\bar{x} = 28\) minutes. The 95% interval: 1. Standard error: \(4/\sqrt{36} = \frac{4}{6} \approx 0.667\). 2. Margin: \(1.96 \times 0.667 \approx 1.31\) minutes. 3. Interval: \(28 \pm 1.31 = (26.69, 29.31)\) minutes. 4. Interpretation: we are 95% confident that the mean delivery time of all orders lies between 26.69 and 29.31 minutes. Check: the margin is the SE stretched by 1.96, and the interval is symmetric around 28 - the estimate always sits dead centre.
Worked example: one data set, three levels
Same deliveries; change only the confidence level. 90%: margin \(1.645 \times 0.667 \approx 1.10\), interval \((26.90, 29.10)\). 95%: margin 1.31, interval \((26.69, 29.31)\). 99%: margin \(2.576 \times 0.667 \approx 1.72\), interval \((26.28, 29.72)\). The pattern is the trade-off made visible: more confidence buys a wider interval - a price paid in precision. Check: all three intervals share the same centre, 28, and differ only in width, exactly as the formula predicts.
Design leverage: the sample size
The margin shrinks with \(\sqrt{n}\): quadruple the orders and the margin halves. To halve the delivery interval's 1.31-minute margin at 95%, the shop would need \(36 \times 4 = 144\) orders, not \(72\) - the square root makes precision expensive, exactly as the CLT chapter warned. This leverage runs backwards too, which is the sample-size topic's whole business: state the margin you want, solve the formula for \(n\), and know the price before the study is run rather than after.
The known-sigma caveat
Everything above leans on \(\sigma\) being known. In most studies it is not, and the sample's \(s\) must stand in for it - a substitution that adds uncertainty this formula ignores. The next topic therefore replaces the normal's \(z^*\) with a wider critical value from the \(t\) distribution. Two conditions otherwise apply here: a random sample, and either a normal population or \(n \ge 30\) so the central limit theorem has done its work on the sample mean's shape.
Common mistakes to avoid
First, interpreting the interval as covering individuals: "95% of deliveries take 26.7 to 29.3 minutes" is wrong - the interval estimates the mean. Second, attaching a 95% probability to this one interval containing \(\mu\): the 95% is the method's long-run rate. Third, using this formula when \(\sigma\) is unknown - that is the \(t\)-interval's job. Fourth, quoting a level with no interpretation: name the level, the parameter, the interval and the units. Fifth, forgetting that \(\bar{x}\) sits at the interval's exact centre. Sixth, computing the margin with \(\sigma\) instead of \(\sigma/\sqrt{n}\).
Key takeaways
With known \(\sigma\), the interval for \(\mu\) is \(\bar{x} \pm z^*\frac{\sigma}{\sqrt{n}}\), its margin the standard error stretched by the critical value: \(z^*\) equals 1.645, 1.96 or 2.576 for 90%, 95% or 99% confidence. Confidence is the method's long-run capture rate - intervals vary, \(\mu\) does not. The deliveries: \(28 \pm 1.31 = (26.69, 29.31)\) at 95%, widening to \(\pm 1.72\) at 99%. Margins shrink with \(\sqrt{n}\). When \(\sigma\) must be estimated, this formula's licence expires and the \(t\)-interval takes over.
Flashcards
Flashcard
The known-σ confidence interval for a mean
Test your understanding
Question 1 of 5
Delivery times have known σ = 4 minutes; a sample of 36 orders gives x̄ = 28. What is the 95% error bound for the mean?
Question 2 of 5
A 95% interval for the mean delivery time is (26.69, 29.31) minutes. Which statement is correct?
Question 3 of 5
Which critical value gives a 99% confidence interval?
Question 4 of 5
Keeping the same data, moving from 95% to 99% confidence:
Question 5 of 5
To halve the margin of a 36-order interval (other things equal), the sample size must become:
