What "95 % confident" means
◈ 5 cardsThe 95 % belongs to the method, not to μ: in repeated sampling 95 % of intervals built this way would contain μ. Wider for more confidence or a larger σ; narrower for a larger n; a claimed value outside the interval is not plausible.
Two sentences the marker accepts, one it does not
Kitchener Ridge's 95 % interval for mean days-to-pay is . What does the 95 % refer to?
Not to . The population mean is a fixed number — unknown, but not random. It either lies in or it does not; there is no probability about it. What is random is the interval: a different sample gives a different and a different pair of limits. The 95 % describes the method: of all the intervals that could be built this way, one sample at a time, 95 % would contain and 5 % would miss it. We do not know which kind ours is.
So the two acceptable sentences are:
- The client's sentence. We are 95 % confident that the mean days-to-pay of all current invoices is between 44.24 and 48.16 days.
- The statistician's sentence. If we drew many samples of 64 and built an interval from each, about 95 % of those intervals would contain the true mean.
And the boxed wrong one: there is a 95 % probability that lies in . It puts the probability on , which has none. A related wrong one: 95 % of invoices take between 44.24 and 48.16 days. The interval is about the mean, not about individual invoices — most invoices lie well outside it, since .
What moves the width
The margin is , so three things drive it:
- Confidence level. At 90 % the critical value is and the margin falls to : . Less confidence, narrower interval — you can be surer of a wider range or less sure of a narrower one, never both.
- Sample size. With instead of 64 the SE doubles to and the 95 % margin doubles to 3.92. Quadrupling halves the width (L8.2's square root again). This is the only lever the analyst controls without giving up confidence — hence Module 9.6.
- Population spread. A larger widens everything; nothing to be done about it.
The sample mean itself is not a driver: a larger shifts the interval but leaves its width unchanged.
Using the interval to judge a claim
A consultant tells Kitchener Ridge that its mean days-to-pay is 50. Is that consistent with the data? 50 lies outside , so at the 95 % level 50 is not a plausible value for — the sample would be an unusually low one if the mean were really 50. Had the claim been 47, inside the interval, it would be plausible — not proven, merely not contradicted.
This is a hypothesis test in disguise: "is inside the interval?" is exactly the two-sided test of Module 10, and L10.7 makes the equivalence explicit. For now, the paper wants the three-part answer: where the claim sits, what that says about plausibility, and the confidence level at which you say it.