Why is 95% the usual figure rather than some other number?

It is a convention, not a law of nature. Researchers could report 90% or 99% intervals instead, and some fields do. Ninety five percent became standard partly through historical habit and partly because it offers a reasonable balance between being informative and being overly cautious.

Is a confidence interval the same as a margin of error?

They are closely related. A margin of error is usually the distance from the estimate to each edge of the interval, so an estimate of 50% with a margin of error of 3 points gives a confidence interval of 47% to 53%. Polls tend to use margin of error language, while scientific papers tend to report the interval directly.

Does a very wide confidence interval mean the study was badly done?

Not necessarily. Width often reflects sample size and natural variation rather than poor methodology. A well-designed study on a small or highly variable population can still produce a wide interval, it simply means more data would be needed to narrow things down.

Can I calculate roughly how big a study needs to be for a tight interval?

There are standard formulas for this, and they depend on how variable the thing you are measuring is and how narrow you want the final range to be. As a rough rule, cutting an interval's width in half generally requires roughly four times as many participants, since precision improves with the square root of sample size.

What should I actually do when I see a confidence interval in a news article?

Check whether it includes the point of no effect, such as zero for a difference or one for a ratio, and check how wide it is relative to the headline number. If the range swings from trivial to dramatic, treat the single reported figure as provisional rather than settled.

Confidence Intervals Explained: How to Read Them at a Glance

31 October 2026

A confidence interval is the range of values that a result could plausibly sit within, given the data collected. A narrow range means the estimate is precise. A wide one means there is a lot of uncertainty, even if the headline number looks impressive.

What the interval actually means

Say a study reports that a new diet led people to lose an average of 4 kilograms, with a 95% confidence interval of 2 to 6 kilograms. That range is not a guess pulled from nowhere. It comes from the spread of the data and the size of the sample. The interval tells you the band of values that are consistent with what was measured.

The 95% refers to the method, not to this one result. If you repeated the study many times, each time calculating a new interval, about 95% of those intervals would contain the true average effect. It does not mean there is a 95% chance the true value lies in this particular range, though almost everyone reads it that way, including some researchers who should know better.

Despite that technical wrinkle, the practical habit is sound. Treat the interval as the set of values the data cannot rule out. Everything inside it is plausible. Everything outside it is unlikely, though not impossible.

Reading the width at a glance

Width is the first thing to check. A tight interval, say 3.8 to 4.2 kilograms, tells you the study pinned down the effect with real precision. A loose one, say minus 1 to 9 kilograms, tells you the study barely narrowed things down at all, even if the midpoint looks the same.

Width is mostly driven by sample size and how much the measurements vary from person to person. Small studies, or studies measuring something noisy, produce wide intervals almost by default. This is why a flashy headline number from a trial of 40 people deserves more scepticism than the same number from a trial of 4,000.

A useful trick is to glance at the interval before you glance at the estimate itself. If the range is wide enough to include both a trivial effect and a dramatic one, the single number in the headline is doing more rhetorical work than the evidence supports.

Where overlap matters most

Confidence intervals become especially useful when you are comparing two things, such as a drug versus a placebo, or this year's results against last year's. If the two intervals barely overlap, or do not overlap at all, that is a strong visual signal that the groups genuinely differ. If they overlap heavily, the apparent difference might just be noise.

This matters for the line between a result and nothing at all. If an interval for a difference includes zero, for example minus 1 to 5 kilograms, the data cannot rule out there being no effect whatsoever. The study might still point toward a real effect, but the zero sitting inside the range is a flag that the finding is not settled.

Charts that show intervals as horizontal bars or whiskers are built for exactly this kind of eyeballing. Scan across several studies at once and you can often tell which ones agree with each other and which ones are outliers, faster than reading any paragraph of text.

Why it beats a p-value alone

A p-value answers a narrow question: how surprising would this data be if there were truly no effect. It collapses everything into a single yes-or-no style threshold, usually 0.05, and gives you no sense of size or precision. Two results can both clear that threshold while meaning very different things in practice.

A confidence interval carries more information in the same space. It shows the direction of the effect, its likely size, and how confident you can be in that size, all at once. An interval of 0.5 to 1 kilogram tells a different story from an interval of minus 3 to 10 kilograms, even if both results technically count as statistically significant.

This is why many journals and statisticians now push for confidence intervals to be reported alongside, or instead of, bare p-values. The interval forces the question that actually matters for decisions: not just whether an effect exists, but whether it is big enough to care about.

Common misreadings to avoid

The most common mistake is reading the interval as a probability statement about the true value, as in 95% chance the real effect is between 2 and 6. Strictly, the true value either is or is not in that range, there is no probability left to assign once the data are collected. The 95% describes the reliability of the method across repeated use, not this specific interval.

A second mistake is assuming a narrow interval means a large or important effect. Precision and size are separate things. An interval of 0.1 to 0.3 kilograms is very precise but describes a change nobody would notice.

A third mistake is comparing two intervals that overlap slightly and concluding there is definitely no difference between groups. Overlapping intervals make a difference less certain, but formal overlap does not automatically mean the groups are statistically indistinguishable. The safest approach is to look at the interval for the difference itself, when it is reported, rather than inferring it from two separate intervals.

Common questions

Why is 95% the usual figure rather than some other number?: It is a convention, not a law of nature. Researchers could report 90% or 99% intervals instead, and some fields do. Ninety five percent became standard partly through historical habit and partly because it offers a reasonable balance between being informative and being overly cautious.
Is a confidence interval the same as a margin of error?: They are closely related. A margin of error is usually the distance from the estimate to each edge of the interval, so an estimate of 50% with a margin of error of 3 points gives a confidence interval of 47% to 53%. Polls tend to use margin of error language, while scientific papers tend to report the interval directly.
Does a very wide confidence interval mean the study was badly done?: Not necessarily. Width often reflects sample size and natural variation rather than poor methodology. A well-designed study on a small or highly variable population can still produce a wide interval, it simply means more data would be needed to narrow things down.
Can I calculate roughly how big a study needs to be for a tight interval?: There are standard formulas for this, and they depend on how variable the thing you are measuring is and how narrow you want the final range to be. As a rough rule, cutting an interval's width in half generally requires roughly four times as many participants, since precision improves with the square root of sample size.
What should I actually do when I see a confidence interval in a news article?: Check whether it includes the point of no effect, such as zero for a difference or one for a ratio, and check how wide it is relative to the headline number. If the range swings from trivial to dramatic, treat the single reported figure as provisional rather than settled.

Lucent, once a week

We read the new research and write up what actually holds, in the same plain language as this page. No jargon, no hype, three papers a week.

Read the latest issue