The Distinction Between Confidence Intervals, Prediction Intervals, and Tolerance Intervals

Confidence intervals, prediction intervals, and tolerance intervals are all expressed as a range, but they answer fundamentally different questions. Choosing the wrong one leads to incorrect scientific conclusions.

The Core Question Each Interval Answers

All three intervals are computed from sample data, but they express uncertainty about different things:

Confidence Interval

Captures uncertainty about the mean of the population. A 95% CI means that if you repeated the experiment many times, 95% of the computed intervals would contain the true population mean. The CI becomes narrower as you collect more data, because your estimate of the mean improves.

Prediction Interval

Captures where a single future observation is expected to fall. A 95% PI means that 95% of new individual measurements from the same population will land within the interval. The PI is always wider than the CI because it incorporates both the uncertainty in the mean and the natural variability of individual observations. Even with an infinite sample, the PI does not collapse to a point — individual variation remains.

Tolerance Interval

Covers a specified proportion of the population with a specified confidence. A 95%/99% tolerance interval means you are 95% confident that at least 99% of the population falls within the interval. Tolerance intervals are common in quality control and manufacturing, where you need to certify that a product spec covers nearly all units produced.

Comparison Table

IntervalAnswersCapturesWidth vs. N
Confidence Interval (CI)Where is the true population mean?The mean of the populationLarger samples → narrower CI
Prediction Interval (PI)Where will the next single observation fall?A single future observationAlways wider than CI; N has less impact
Tolerance Interval (TI)What range covers at least X% of the population?A specified proportion of the populationWider as coverage proportion increases

The Most Common Mistake

Mistaking a CI for a PI

The most common error is interpreting a confidence interval as if it tells you where individual measurements will fall. It does not. A CI tells you about the mean; a PI tells you about individuals. If you say "our treatment gives a blood pressure of 120 mmHg (95% CI: 118-122)" and a colleague assumes that means 95% of patients will have BP between 118 and 122 — that is an incorrect interpretation. The prediction interval for individual patients would be much wider.

Practical Examples

Use a CI when: you want to report the mean effect of a drug and how precisely it has been estimated from your sample.
Use a PI when: you want to predict what will happen to the next patient you treat, not just what the average response is.
Use a TI when: you need to certify that a manufacturing process produces items within specification for nearly all units (quality control, regulatory submissions).

Related Resources