What a Confidence Interval Is Not, and How to Read It as a Measure of Precision

A confidence interval is a range of values compatible with your data, and its width tells you how precisely you measured the thing you set out to measure. A 95 percent interval is built by a procedure that, run on fresh samples, would trap the true value about 95 times out of 100. That describes the method's long-run reliability, not the single interval in front of you. It is not the probability that the truth sits inside this range, and it is not a verdict on whether the result is real. This is a methods article and not medical advice; for anything about your own care, talk with a clinician who knows your history.

I read these intervals from both sides of the page. As a reviewer I have seen an interval that ran from "barely helpful" to "wildly helpful" sold as a clean win. As an author, including a Diabetes Care meta-analysis I co-wrote, I have had to decide what a wide interval was honestly allowed to claim. It is one of the most useful things in a results section, and one of the most misread.

What is a confidence interval, in one plain sentence

Here is the quotable version. A confidence interval is the range of effect sizes your data do not rule out, given the model you assumed and the level of confidence you chose. A narrow interval means you pinned the answer down; a wide one means the honest answer is "somewhere in this large neighborhood, and we cannot say where." The point estimate in the middle is your best guess. The interval tells you how much to trust it.

The width is the message. Two studies can report the same central estimate, say a risk reduction of 20 percent, and mean completely different things. One reports an interval from 18 to 22 percent; the other reports 2 to 60 percent. Same headline, utterly different evidence: the first has measured something, the second has barely glimpsed it. Read only the middle number, and a thin result passes for a solid one.

Why a confidence interval is not the probability the truth is inside it

This is the misreading I meet most often, and it slips past careful people because the everyday meaning of "confidence" pulls the wrong way. A 95 percent interval invites you to say there is a 95 percent chance the true value lies between these two endpoints. Under the standard framework that built the interval, the true value is a fixed quantity, not a random one; it either sits in your interval or it does not. The 95 percent describes how often the recipe succeeds across many repetitions, not the odds for the one in your hands.

The distinction sounds pedantic until it changes a decision. To make the satisfying statement, that there is a 95 percent chance the truth is in this range, you need different machinery and a set of prior beliefs the ordinary interval never asked for. The honest reading is humbler: the interval marks the values your data are compatible with, computed by a method you can trust on average.

Why an interval is not a row of equally plausible values

A second trap is to read the interval as a flat strip where every value is just as likely as every other, with a hard wall at each end. That is not how the evidence is shaped. Values near the center are better supported than values out at the edges, and the truth does not become impossible one step past the boundary. The endpoints are a convention drawn at a chosen confidence level, not a cliff. So beware the habit of seizing on the most dramatic limit and quoting it as if the data pointed there. They do not; the data point hardest at the center.

Why a confidence interval is not a verdict on significance

Many readers use the interval as a yes-or-no device. If a 95 percent interval for a difference excludes zero, they call the result real; if it includes zero, they call it nothing. That move throws away most of what the interval was telling you.

Two examples show the cost. An interval can exclude zero and still run from "too small to matter to anyone" up to "modestly useful," pairing statistical significance with very little practical promise. The reverse is more painful: an interval can include zero while sitting mostly over a meaningful benefit, its lower edge dipping just below the no-effect line. That usually means the study was too small to settle the question, not that the treatment does nothing. So an interval that includes zero is not proof of no effect; it reports that "no effect" is one of the values your data allow, often alongside effects worth having.

What the width is actually made of

The width of an interval is mostly a story about sample size and variability: small, noisy studies produce wide intervals, larger and cleaner ones produce tighter ones. A wide interval is not a flaw to apologize for. It signals honestly that the question is still open, and the right response is usually more data, not a bolder claim. What worries me is the opposite case: a narrow interval around an effect too good to be plausible, which often points to a problem upstream in how the study was run.

How I read a confidence interval as a reviewer

When I review, I read the interval before the point estimate, and long before any p-value, asking what the two endpoints would mean in the real world. If both ends describe an effect a patient would notice, the result is interesting whether or not it crossed a threshold. If one end is clinically trivial, the study cannot yet rule out a result not worth acting on.

So treat a confidence interval as a precision instrument and a plausibility range, not a truth meter. Read the width to judge how much the study knew, read both endpoints for the best and worst cases the data permit, and resist the pull to collapse the range into a single verdict. The fix is not to distrust the interval but to stop asking it for something it cannot give.