What they tell you that p-values don't
A p-value tells you one thing: is this result likely due to chance? It says nothing about how big the effect is, how precise your estimate is, or whether the effect matters clinically. Confidence intervals fix this.
How to read confidence intervals to understand the range of plausible effects, why width matters as much as the point estimate, and how to use CIs to make better clinical decisions than p-values alone would allow.
A 95% confidence interval means: if we repeated this study 100 times, about 95 of those intervals would contain the true population value.
It's a range of plausible values for the true effect, given your data.
Point estimate: Your best guess (the middle of the CI)
Lower bound: The smallest plausible effect
Upper bound: The largest plausible effect
The true effect could be anywhere in that range. A CI of 2% to 40% is very different from 18% to 22%, even if both have the same point estimate of 20%.
The confidence interval tells you what you don't know. A wide CI means high uncertainty; a narrow CI means you've pinned it down.
The width of a CI tells you how precise your estimate is—which depends mainly on sample size.
Same point estimate (20% reduction), different sample sizes:
A wide CI doesn't mean no effect—it means we don't know yet. A narrow CI around zero means we're confident there's no meaningful effect.
If the 95% CI excludes the null value (0 for differences, 1 for ratios), the result is significant at p < 0.05. You get significance information for free.
The point estimate tells you the magnitude. P-values don't.
The width tells you how confident to be. P-values don't distinguish between precise and imprecise estimates.
You can see if the entire CI is above or below clinically meaningful thresholds. P-values can't do this.
Every p-value can be converted to a CI, but not vice versa. The CI is strictly more informative.
Effect: 25%, 95% CI: 20% to 30%
CI excludes zero, narrow width. Strong evidence of a real, clinically meaningful effect.
Effect: 25%, 95% CI: 2% to 48%
CI excludes zero (significant), but the true effect could be anywhere from trivial (2%) to huge (48%). Needs more data.
Effect: 15%, 95% CI: -10% to 40%
CI includes zero, but also includes clinically meaningful effects. We can't rule out benefit OR harm. Underpowered.
Effect: 0.5%, 95% CI: -2% to 3%
CI includes zero but is narrow. Even the extremes are clinically irrelevant. This is a confident null—no meaningful difference.
A common mistake: "The CIs overlap, so the groups aren't different."
Two groups can have overlapping CIs and still be significantly different from each other. The correct comparison is the CI of the difference between groups, not whether individual CIs overlap.
Group A: 40% success, 95% CI: 30% to 50%
Group B: 55% success, 95% CI: 45% to 65%
The CIs overlap (both include 45-50%), but the difference (15%) has its own CI: 2% to 28%. Since this excludes zero, the groups are significantly different.
Never judge significance by eyeballing overlap. Look for the CI of the difference, or the p-value for the comparison.
For each scenario, choose the best interpretation.
When reading a study, look at the CI before the p-value. Ask: "What's the smallest plausible effect? What's the largest? Is this range narrow enough to be informative?"