Term
Confidence Interval
The range a true effect plausibly occupies. The single most useful number on this site, and the one most often left out of supplement marketing.
A confidence interval is a range of values that the true effect plausibly falls within, given the data. The 95% figure means that if the same study were repeated many times, about 95% of the resulting intervals would contain the true value.
What it tells you that an average does not
An average alone hides how sure the finding is. Two trials can both report “−0.35 mmol/L”:
| Reported | 95% CI | What it means |
|---|---|---|
| −0.35 mmol/L | −0.38 to −0.32 | A tight, consistent effect |
| −0.35 mmol/L | −0.90 to +0.20 | An effect that may be nothing |
Same headline number, different information. The second interval includes zero, which means the data are compatible with no effect at all.
The zero-crossing rule
If a confidence interval for a difference crosses zero, the trial did not establish a difference. For ratios — odds ratios, risk ratios — the equivalent test is whether the interval crosses 1.
This rule is applied on every table on this site. A row whose interval crosses zero is graded C regardless of how good the point estimate looks, because the point estimate is the middle of a range rather than a finding. The grading standard is published in full on how we grade evidence.
Why “showed a trend” usually means this
The phrase “a trend toward improvement” appears throughout supplement writing. In trials it almost always describes a result whose interval crosses the threshold and whose p-value sits above the conventional cut-off. It is a way of reporting a non-finding without saying so.
When you see the phrase, look for the interval. If it crosses zero, the honest description is that the trial did not show the effect.
What it does not tell you
A narrow interval does not mean the effect is large, and a wide interval does not mean the effect is small. Width describes precision; the centre describes size. A precise estimate of a tiny effect is still a tiny effect.
Nor does a confidence interval account for bias. A pooled estimate from trials that were never published for negative results can have a tight interval built on a skewed sample.
Facts checked: 29 October 2026 · VitalRoundup Editorial Team · How we review