In simple terms
An estimate built from a sample is never exact. A confidence interval is the range of values that is likely to contain the quantity you were trying to measure.
The US Census Bureau describes it as the uncertainty surrounding an estimate. A wide interval is a warning. A narrow one is a claim of precision.
How it works
The confidence level describes the method, not the single interval in front of you.
The NIST statistical handbook is explicit. At a 95 percent level, if the same population were sampled many times and an interval built each time, roughly 95 percent of those intervals would bracket the true value. Any particular interval either contains it or does not.
Two things set the width. Wider spread in the data widens the interval. A bigger sample narrows it, because sample size sits in the denominator of the calculation. Intervals are also wider when the standard deviation has to be estimated from limited data rather than known in advance.
The level itself is a choice, and conventions differ by field. Medical and scientific work usually reports 95 percent. The Census Bureau publishes 90 percent intervals for its income and poverty estimates. Higher confidence costs precision: for the 1995 US poverty count, the 90 percent interval ran from 35,534,124 to 37,315,094 people, while the 95 percent interval stretched from 35,363,606 to 37,485,612.
Why it matters
A single number invites false certainty. An unemployment rate, a poll result or a treatment effect is an estimate, and the interval is where the honesty sits.
It also shapes how comparisons should be read. The Census Bureau publishes separate guidance on comparing estimates precisely because two figures that look different may not be distinguishable once uncertainty is taken into account.
And a wide interval is not a failure. On a small sample it is an accurate report of how much the data can actually support.
Where you’ll see it
- Official statistics releases, printed next to the estimate itself.
- Medical papers reporting a treatment effect with a 95 percent interval.
- Opinion polls, where the margin of error is a confidence interval in other clothing.
- Error bars on charts, which should always state what they represent.
Example
A survey puts average weekly household spending at 420 units, with a 95 percent confidence interval of 400 to 440. Run the same survey design many times and about 95 percent of the intervals produced would contain the true average.
Often confused with
The chance that the true value lies in this interval. The confidence level says how often the procedure succeeds across repeated samples. It is not a probability attached to the one interval you are looking at.
Key facts
- The NIST/SEMATECH handbook defines a confidence interval as a range of values likely to contain the population parameter of interest.1
- At a 95 percent confidence level, repeated sampling and interval building would bracket the true parameter in approximately 95 percent of cases.1
- The confidence level of 1 minus alpha is the inverse of the significance level alpha.1
- Intervals are necessarily wider when the standard deviation is estimated from limited data, and sample size appears in the denominator, so larger samples narrow the interval.1
- The US Census Bureau describes a confidence interval as a range of values that describes the uncertainty surrounding an estimate, and treats larger intervals as a signal for more caution.2
- The Census Bureau routinely publishes 90 percent intervals; for the 1995 US poverty estimate the 90 percent interval was 35,534,124 to 37,315,094 and the 95 percent interval 35,363,606 to 37,485,612.2
Related concepts
In the news
Quick checkWhat does a 95 percent confidence level actually describe?Show answer
The method. Across many repeated samples, about 95 percent of the intervals built this way would contain the true value.
Sources
- NIST/SEMATECH. e-Handbook of Statistical Methods, 7.1.4 What are confidence intervals?. Undated (accessed 15 September 2026)
- US Census Bureau. A Basic Explanation of Confidence Intervals. Undated (accessed 15 September 2026)
Editorially reviewed by Specialty Digest Editorial TeamLast reviewed September 16, 2026Researched and drafted with AI assistanceReport an issue