Confidence Intervals in Scientific Results
An estimate needs a range
A study usually reports an estimated difference based on a sample rather than a complete population. Another sample collected under comparable conditions would generally produce a different estimate. A confidence interval describes uncertainty associated with the estimation procedure and its assumptions.
The Cochrane Handbook, Chapter 15: Interpreting results and drawing conclusions discusses using intervals when interpreting scientific findings. Read the lower limit, upper limit and point estimate together. An interval that permits only small differences says something different from one that spans a substantial decrease and a substantial increase, even if both studies are described as inconclusive.
Avoid the usual probability shortcut
In the conventional frequentist interpretation, a 95% confidence procedure would cover the true parameter in 95% of intervals produced over repeated sampling under its assumptions. The specific interval already calculated is not usually interpreted as assigning a 95% probability to the fixed parameter lying inside it.
This distinction need not make a reading note complicated. Write that the interval indicates the precision of the estimate under the stated analysis. Avoid saying that 95% of participants fall within it: an interval for a group difference does not describe the distribution of individual measurements.
Read the scale before the limits
For a difference measure, zero commonly represents no difference. For a ratio measure, one commonly represents no relative difference. Confusing these reference values can reverse the interpretation of a figure. Check the axis label and whether the analysis used a transformed scale.
The CONSORT 2025 explanation and elaboration: updated guideline for reporting randomised trials includes reporting of precision alongside effect estimates. In a hypothetical peptide assay, a mean difference and its interval should also retain the original measurement unit. Without that unit, the reader cannot judge what the permitted values imply for the laboratory question being studied.
Recognise what an interval leaves out
An interval calculated from a statistical model does not automatically incorporate uncertainty from selective reporting, misidentified samples, confounding or an unsuitable experimental model. More observations may narrow sampling uncertainty while leaving a systematic problem intact. Precision therefore deserves attention alongside design and material identity.
When consulting our research overview, copy the interval rather than only the headline estimate. Then add a separate note on study limitations. This keeps two useful ideas visible: what the data say about the size of a comparison, and what features of the experiment may restrict the trust or scope of that statement.
Related research reading
- Multiple Endpoints: Which Question Was the Study Designed to Answer?
- Effect Size and Statistical Significance
Explore the research blog and original batch-document library for further context.