Statistics Interpretation
How to Read Confidence Intervals and Effect Sizes
Start with the size of the observed effect, then read how much uncertainty surrounds it. A confidence interval helps with both questions but does not prove the result is true or clinically important.
Educational content only. Not medical advice.
Start with the effect estimate
An effect estimate quantifies the observed contrast or association: a mean difference, risk ratio, odds ratio, correlation, standardized difference, or another measure tied to the study question. Its sign or position relative to a null value shows direction, while its magnitude uses the measure's own scale. A large numerical value is not automatically important; readers need the outcome definition, measurement scale, baseline level, and a pre-specified threshold for scientific or practical relevance.
A confidence interval communicates precision and compatible values
A 95% confidence interval is produced by a procedure that, under its model assumptions and repeated sampling, has 95% coverage of the target parameter. For a completed study, it is useful to read the interval as a range of effect values reasonably compatible with the data and model. It is not a 95% probability that the fixed true value lies inside this particular interval, and it does not include every source of bias or uncertainty.
Width matters as much as whether the interval crosses the null
A narrow interval suggests greater statistical precision than a wide one on the same scale and under comparable assumptions. An interval that crosses the null may still include scientifically important benefit and harm, indicating inconclusive precision rather than proof of no effect. An interval entirely away from the null can still describe an effect too small to matter. Read both boundaries against practical thresholds, not only against zero or one.
Design quality determines what the estimate can mean
Confidence intervals do not correct confounding, selection bias, outcome switching, missing data, multiplicity, poor measurement, or an inappropriate statistical model. Compare the estimate with the pre-specified primary outcome, analysis population, protocol, and sensitivity analyses. Also distinguish relative from absolute measures because they frame magnitude differently. Estimation supports disciplined interpretation; it does not predict an individual's result or establish medical guidance.
Evidence limits
- Coverage and interpretation depend on the statistical model, study design, and assumptions.
- Confidence intervals generally omit unmeasured bias and may be affected by multiplicity or selective reporting.
- Population estimates do not forecast an individual outcome.
Sources and further reading
These sources ground the definitions and evidence boundaries on this page. A citation is a route for verification, not an endorsement of a product or personal use.
The BMJ and CONSORT Group
CONSORT 2025 Statement: Updated Guideline for Reporting Randomised Trials
Primary reporting guideline emphasizing transparent estimates, outcomes, analyses, and uncertainty.
Open sourceJornal Vascular Brasileiro via PubMed Central
P-value and Effect-size in Clinical and Experimental Studies
Peer-reviewed explanation of joint interpretation of effect sizes, confidence intervals, and p-values.
Open sourceCommon questions
Does a 95% confidence interval contain 95% of individual outcomes?
No. It describes uncertainty around a population parameter estimate, not the distribution of individual observations.
Does crossing the null prove no effect?
No. The interval may remain compatible with meaningful effects in more than one direction or simply be imprecise.
Can a statistically precise effect be unimportant?
Yes. A narrow interval can surround an effect that is too small to be scientifically or practically meaningful.
Continue with context
Build a more rigorous reading workflow
Explore the complete PeptideSchool research workspace, including source trails, study organization, and the Premium Calculator. The tools are educational and are not a substitute for professional judgment.