An estimated effect becomes more useful when its uncertainty is visible. A confidence interval provides lower and upper limits calculated from the data and a statistical model. Reading it well means identifying the parameter being estimated, the assumptions behind the calculation and the range of conclusions the interval leaves open.
Understand what the confidence level describes
NIST explains confidence through repeated use of the interval method: under its assumptions, a 95% procedure would cover the true parameter in about 95% of repeated samples. It does not assign a 95% probability to the fixed parameter lying in one already calculated interval.NIST/SEMATECH — Confidence Limits for the Mean (opens in a new tab)
In ordinary reading, the limits help convey the precision of an estimate. The parameter might be a group mean, a difference between conditions or a ratio. Locate that target before interpreting the numbers.
| Summary | What it concerns |
|---|---|
| Confidence interval | Uncertainty in an estimated parameter |
| Observed range | Smallest and largest recorded values |
| Standard deviation | Dispersion of observations |
| Prediction interval | Uncertainty about a future observation under a model |
These summaries can appear on similar-looking plots. An unlabeled error bar is therefore insufficient: the caption needs to identify what was calculated and which observations or independent units contributed.
Work through an interval for paired changes
For an original numerical illustration, suppose ten independent experimental units each have a paired before-and-after change. Their mean change is +5 units and the sample SD of those changes is 4 units. Assume the changes meet the conditions for a one-sample t interval.
The standard error of the mean change is 4/√10, approximately 1.265 units. Using the 95% two-sided t multiplier for nine degrees of freedom, approximately 2.262, gives a margin of about 2.861 units.
| Quantity | Value |
|---|---|
| Mean change | +5 units |
| Standard error | About 1.265 units |
| Margin | About 2.861 units |
| 95% interval | About +2.14 to +7.86 units |
This applies NIST’s mean-interval formula to the set of paired changes. It estimates the mean change in the relevant population of units; it does not say that 95% of individual changes fall between those limits.NIST/SEMATECH — Confidence Limits for the Mean (opens in a new tab)
Read more than whether the interval crosses zero
For a difference, zero is the no-difference value. For a ratio, the corresponding value is usually one. Always check the scale, since using zero as the reference for every effect measure leads to incorrect interpretations.
In the hypothetical example, the interval is entirely positive, but it spans changes from a little over 2 to nearly 8 units. If the scientific question requires distinguishing a small change from a larger one, considerable uncertainty may remain.
Conversely, an interval spanning negative and positive values can leave several effects unresolved. It does not prove that the effect is exactly zero. Read how large those values are in the actual outcome scale.
Inspect what the calculation cannot repair
The worked calculation assumed ten independent units. Treating multiple readings from each unit as additional independent units would change the calculation without necessarily adding the information that it assumes.
The interval also does not automatically account for selective exclusions, a biased comparison, an invalid measurement or unreported analyses. These problems concern how the data and result arose, not simply the width of the displayed interval.
For complex designs, the reported interval may come from a model that accounts for pairing, clustering or other structure. Read the model description instead of trying to reconstruct every interval from a simple formula intended for a different design.
A useful summary states the estimated parameter, limits, confidence level and key design assumptions. That gives uncertainty a specific meaning and prevents a precise-looking number from being mistaken for a complete assessment of the evidence.
Sources and further detail
- NIST/SEMATECH — Confidence Limits for the Mean (opens in a new tab)
Repeated-sampling interpretation and t-interval formula read. The ten-unit paired-change example is original, assumes suitable independent changes and uses an approximate t multiplier; it is not a reanalysis of a peptide paper.
Sources checked 19 September 2026. Worked examples are illustrative unless a supplied report is explicitly identified. This article has not undergone independent scientific peer review.