01
What the evidence says
For a difference measure, an interval crossing zero includes no difference; for a ratio measure, crossing one does. That threshold answers a narrow significance question. More importantly, inspect whether the interval includes effects that would be clinically important, trivial, or harmful. A narrow interval around a small effect differs from a wide interval containing both major benefit and meaningful harm.
Sources for this section: [1]
02
How to use the finding
Precision generally improves with more informative events and better measurement, not merely a large headline sample. Clustered data, repeated outcomes, loss to follow-up, multiple analyses, and model assumptions alter the interval. A 95% method means that across repeated compatible studies, about 95% of intervals constructed this way would cover the target parameter; it does not assign 95% probability after observing this interval.
Sources for this section: [1]
03
Limits and open questions
Do not treat two results as different merely because one interval excludes the null and the other does not. Directly test the interaction or difference. Confidence intervals also do not capture bias, selective reporting, confounding, poor outcome definition, or generalizability. A precise estimate from a biased design can be confidently wrong, so design appraisal comes before decimal places.
Sources for this section: [1]
Sources
- Statistical tests, P values, confidence intervals, and power: a guide to misinterpretationsEuropean Journal of Epidemiology · 2016DOI 10.1007/s10654-016-0149-3