Open the abstract of almost any study about children and you'll find a sentence like this: "Children in the program gained 4.2 points (95% CI 1.1 to 7.3)." News stories tend to report the 4.2 and drop the brackets. The brackets hold the part that tells you how far to trust the 4.2. This note explains how to read them.
What is a confidence interval?
A confidence interval is the range of values that fit a study's data reasonably well, given the play of chance in who happened to take part.
No study measures every child. It measures a sample, and a different sample would give a slightly different answer. The single number a study reports, called the point estimate, is its best guess. The confidence interval puts a range around that guess.
In the example above, the best guess is a gain of 4.2 points. The data fit any true gain from about 1 point to about 7 points. A gain of 1 point would barely register in a classroom. A gain of 7 might be worth the cost of the program. The study can't yet say which end is closer to the truth.
Most papers use 95% intervals. Some use 90% or 99%. A 99% interval is wider, because you're asking for more certainty.
Why does the width of the interval matter?
A narrow interval means the study pinned the effect down; a wide one means the true effect could be small or large.
Width mostly depends on two things: how many participants there were and how much their results varied. Big studies with consistent results give narrow intervals. Small studies, or outcomes that vary a lot from child to child, give wide ones. Our sample size guide shows how margins of error shrink as samples grow. A confidence interval works the same way.
Two studies can report the same point estimate and mean very different things. Here are two made-up examples:
| Study | Estimate | 95% CI | What it tells you |
|---|---|---|---|
| A, 2,000 children | +4 points | 3 to 5 | A modest gain, measured precisely |
| B, 60 children | +4 points | -3 to 11 | Could be harmful, nothing, or a large gain |
Both studies "found a 4-point gain". Only study A tells you much.
What does it mean when the interval crosses zero or one?
When the interval includes the value that means "no effect", the study can't rule out that nothing happened, so the result isn't statistically significant at that level.
For differences (points, minutes, kilograms), no effect is zero. Study B's interval runs from -3 to 11, so zero sits inside it. For ratios, such as relative risk or odds ratios, no effect is one. An odds ratio of 1.4 with an interval from 0.9 to 2.1 includes one, so the study hasn't shown a difference in odds.
This is the link between intervals and p-values. A 95% interval that excludes the no-effect value matches a p-value below 0.05. Our note on what statistically significant means covers the p-value side.
The interval gives you more than a yes or no, though. It shows how big the effect might plausibly be. An interval of 0.98 to 3.5 and one of 0.4 to 1.05 both include one, yet the first leans toward a real increase and the second leans the other way.
How should you read both ends?
Read the low end and the high end separately, and ask whether either one would change your decision.
This habit turns a statistical range into something practical:
- Find the point estimate and the two ends of the interval.
- Translate all three into plain terms. Points on a reading test, minutes of sleep, extra children per 100.
- Ask: if the low end were true, would this matter? If the high end were true, would it?
- If both ends would lead you to the same choice, the study is useful for that choice even if the interval is wide.
- If the two ends point to different choices, the question is still open, and one more study may settle it.
Imagine a sleep study that reports teens with a phone curfew slept 12 minutes longer (95% CI 2 to 22 minutes). The low end is barely noticeable. The high end is a meaningful amount of sleep for a teenager. A reasonable reading: there's probably some benefit, of uncertain size.
What are the common mistakes with confidence intervals?
The most common mistake is treating the point estimate as the answer and ignoring the range around it.
A few others turn up often:
- Reading "95%" as a 95% chance the true value is inside this particular interval. That isn't quite what it means, though as a working shorthand it does little harm.
- Comparing two groups by checking whether their separate intervals overlap. Overlapping intervals can still hide a significant difference. Look for the interval of the difference itself.
- Assuming the interval covers every source of error. It only covers chance. A biased sample or a badly measured outcome can produce a narrow, precise and wrong interval.
That last point is why design still matters. The study types guide explains which designs are more likely to point at a cause.
Where will you see intervals outside single studies?
Systematic reviews show intervals for every included study on one chart, called a forest plot, with the combined result at the bottom.
Each line on a forest plot is one study's interval. A vertical line marks no effect. You can see at a glance which studies are precise, which are noisy and whether they agree. The combined estimate at the bottom usually has the narrowest interval, because it pools the most data.
When you next see a study in the news, look for the brackets. The study checklist includes a question about effect size and precision, so you can weigh the result with the range in view.
Questions readers ask
- What does the 95% in a 95% confidence interval mean?
- It describes the method. If researchers ran the same kind of study many times and calculated a 95% interval each time, about 95 of every 100 intervals would contain the true value. Any single interval either contains it or doesn't. In everyday reading, treat it as the range of results that fit the data well.
- Why does a confidence interval sometimes include zero?
- Because the data are compatible with no effect at all. For a difference between groups, zero means no difference. For ratios such as relative risk or odds ratios, the no-effect value is one.
- Is a confidence interval the same as a margin of error?
- They are closely related. A poll's margin of error is usually half the width of a 95% confidence interval. Saying 52% plus or minus 3 points is the same as an interval from 49% to 55%.
- Where do I find the confidence interval in a paper?
- Usually in the abstract's results sentence, written as "95% CI" followed by two numbers, and in the results tables. If a paper reports only p-values and no intervals, that's worth noting.