An estimate needs a measure of uncertainty
Suppose you want to know what fraction of seeds in a large batch will sprout. Testing every seed would defeat the purpose, so you test a sample and calculate the fraction that sprouts. That sample gives you an estimate of the fraction for the whole batch. A different sample might give a slightly different answer, even if you follow the same procedure carefully.
A confidence interval places a range around an estimate to reflect that sampling variation. For example, imagine a hypothetical report that estimates a batch’s sprouting rate at 60% and gives a 95% confidence interval from 52% to 68%. The interval does not say that every seed has a sprouting probability between those endpoints. It concerns a characteristic of the batch as a whole: its sprouting rate under the conditions being studied.
Confidence intervals can also be calculated for differences between groups, average measurements and many other quantities. In each case, the first question is: What quantity is this interval intended to estimate?
What the “95%” describes
A 95% confidence interval gets its name from the long-run behavior of the method used to construct it. Imagine repeatedly taking new samples from the same population and calculating an interval from each one in the same way. If the method’s assumptions hold, about 95% of those intervals would contain the population value being estimated. The intervals change from sample to sample; that population value does not.
This is subtly different from saying, “There is a 95% probability that the true value lies in this particular interval.” Under the usual frequentist interpretation, once the sample has been collected and its interval calculated, that interval either contains the fixed population value or it does not. The 95% describes the procedure’s expected coverage over repeated sampling, not a probability assigned to this one completed interval.
The distinction may feel fussy, but it prevents a common mistake: treating a confidence label as a guarantee. Even a correctly calculated 95% interval will sometimes miss its target. Other statistical approaches can assign probabilities to a range of possible parameter values, but those probabilities rely on a different framework and its own assumptions.
Why some intervals are wider than others
An interval’s width reflects how precisely the available data and chosen method can estimate the target. Larger, well-collected samples often produce narrower intervals because random sampling variation tends to shrink. Measurements that vary greatly from one observation to the next often produce wider intervals. The requested confidence level matters too: using the same data and method, a higher confidence level generally requires a wider interval to capture the target more often in repeated sampling.
A narrow interval is not automatically a trustworthy one. Imagine measuring the sprouting rate only among seeds stored in unusually favorable conditions, then applying the result to every seed in the batch. Collecting more seeds from that same unrepresentative group might narrow the interval without fixing the mismatch. Similarly, a faulty measuring tool can yield precise-looking results centered on the wrong value.
The advertised coverage depends on assumptions about how the data were obtained and how the interval was calculated. When those assumptions are poor approximations, an interval labeled 95% may not achieve 95% coverage in practice.
What an interval leaves unanswered
A confidence interval does not tell you whether a study was well designed, whether an effect is important, or whether one factor caused another. An interval for a difference between two groups may help show which differences are compatible with a statistical model, but it cannot by itself rule out confounding or measurement errors.
It also does not describe where most individual observations lie. An interval for an average can be quite narrow while individual measurements vary widely. If you need to anticipate the result of one future measurement, you need a method aimed at prediction rather than an interval for the population average.
Finally, values inside a confidence interval should not all be treated as equally supported, nor should values just outside it be dismissed as impossible. The endpoints mark a result of a particular calculation at a chosen confidence level. They are useful boundaries for interpreting an estimate, not a complete map of uncertainty.
How to read one responsibly
Start by identifying the estimated quantity, its units and the population it is meant to represent. Then look at both endpoints, not just the central estimate. Ask whether the range is narrow enough to answer the question you care about. A range can be statistically informative yet still leave room for outcomes that would lead to very different practical conclusions.
Next, consider how the sample was selected and whether important sources of error are missing from the interval. Check that comparisons use the same outcome definitions and conditions. If a report presents several intervals, remember that choosing the most striking one after looking at the data can complicate their interpretation.
A confidence interval is best read as a conditional statement: given the data-collection process, statistical model and calculation used, this method produces ranges with a stated long-run coverage. It makes uncertainty visible—but understanding what that uncertainty includes, and what it leaves out, is part of reading the result.
