How to Interpret a 95% Confidence Interval
A 95% confidence interval gives an estimated range around an effect and communicates its precision under the statistical model and assumptions used. Narrow intervals generally indicate greater precision; wide intervals indicate greater uncertainty. Under the standard frequentist interpretation, it is not strictly correct to say there is a 95% probability that the true value lies inside the particular interval you calculated.
Your table says:
RR 0.72, 95% CI 0.60-0.87
Many researchers read:
RR 0.72
and treat everything inside the brackets as statistical furniture.
Those brackets may contain the most useful part of the result.
What does a confidence interval tell me?
A confidence interval combines an effect estimate with information about uncertainty or precision.
Suppose you estimate:
Mean difference = 4.2 points
95% CI = 1.5 to 6.9
The point estimate is 4.2.
The interval shows the range of values around that estimate produced by the interval procedure under your model and assumptions.
A narrower interval generally signals greater precision.
A wider interval signals greater uncertainty.
This is why reporting an estimate without its confidence interval often leaves the reader with too little information.
What does “95% confidence” actually mean?
The common explanation is:
“There is a 95% probability that the true value lies between 1.5 and 6.9.”
Under the conventional frequentist interpretation, that is not strictly correct.
The 95% refers to the long-run performance of the interval-producing procedure.
If the study process were repeated many times under the model assumptions and a 95% confidence interval were calculated each time, approximately 95% of those intervals would contain the target parameter.
Once your data have produced one particular interval, frequentist theory does not treat the fixed parameter as having a 95% probability of being inside that interval.
That distinction is technically important.
For ordinary research writing, a more useful and less misleading formulation is:
The data are compatible with effect values ranging approximately from 1.5 to 6.9, under the model and assumptions used.
You do not need to turn every results paragraph into a philosophy seminar.
You do need to avoid confidently teaching the wrong probability statement.
interpreting RR 0.72, 95% CI 0.60-0.87
Suppose an intervention study reports:
RR = 0.72
95% CI = 0.60 to 0.87
Read it in stages.
1. What is the point estimate?
RR = 0.72.
The estimated risk in the intervention group is 0.72 times that in the comparison group, corresponding to an estimated 28% relative reduction.
2. What is the interval?
0.60 to 0.87.
The data and model produce an interval compatible with relative risks from approximately 0.60 to 0.87.
3. What is the null value?
For a risk ratio, the usual null value is 1.
The interval 0.60-0.87 does not include 1.
4. How precise is the estimate?
That depends on the width of the interval and the substantive scale.
The interval is considerably narrower than, for example:
RR 0.72, 95% CI 0.38-1.36
Same point estimate.
Very different uncertainty.
What is the null value for a confidence interval?
It depends on the effect measure.
For differences, the usual null value is often 0.
Examples:
- mean difference;
- risk difference;
- regression coefficient representing a difference.
For ratios, the usual null value is often 1.
Examples:
- risk ratio;
- odds ratio;
- hazard ratio.
So:
Mean difference 5.0, 95% CI 2.0 to 8.0
does not include the null value of 0.
But:
Mean difference 5.0, 95% CI -1.0 to 11.0
does.
And:
RR 0.70, 95% CI 0.58-0.84
does not include 1.
Whereas:
RR 0.70, 95% CI 0.49-1.04
does.
For corresponding conventional two-sided tests under matching assumptions, a 95% confidence interval excluding the null value is closely related to a p-value below 0.05.
But do not stop there.
Bench Check: using the confidence interval only as a significance detector
A common workflow is:
Does the interval cross zero or one?
No → significant.
Yes → not significant.
Done.
That wastes much of the interval.
A confidence interval can help you ask:
- What is the estimated magnitude?
- How uncertain is it?
- Does it include the null?
- Does it include clinically or practically important values?
- Does it rule out effects that would matter?
- Is it compatible with benefit and harm?
The null value is one landmark inside the interval. It is not the whole landscape.
Two studies with the same estimate
Consider:
Study A
RR 0.72, 95% CI 0.69-0.75
Study B
RR 0.72, 95% CI 0.38-1.36
Both have the same point estimate.
But Study A estimates that effect much more precisely.
Study B remains compatible with a much wider range of effects, including the null value and values on both sides of it.
Calling both studies “RR = 0.72” without the interval makes them look far more similar than they are.
How does sample size affect confidence intervals?
All else equal, more information generally increases precision.
That can mean narrower confidence intervals.
But interval width is not determined by sample size alone.
It can also depend on:
- variability;
- event frequency;
- model specification;
- design;
- clustering;
- measurement quality;
- weighting and other analytic features.
A large sample does not guarantee a narrow interval for every parameter.
And a narrow interval does not guarantee the estimate is unbiased.
Precision and validity are different questions.
Does a confidence interval tell me whether a result is important?
It helps, but it cannot decide importance by itself.
Suppose:
RR 0.98, 95% CI 0.97-0.99
This is a very precise estimate close to the null.
Whether a 1-3% relative reduction matters depends on the outcome and context.
A tiny relative change in a rare, mild outcome may be unimportant.
A small change in mortality across a very large population may matter greatly.
Interpretation requires context.
What do I actually write in my results section?
Instead of:
There was a significant association between intervention exposure and the outcome.
Write the estimate:
Participants exposed to the intervention had an estimated 28% lower risk of the outcome than those in the comparison group (RR 0.72, 95% CI 0.60-0.87).
If appropriate, add the p-value.
The confidence interval allows the reader to see the uncertainty directly.
For a mean difference, you might write:
Mean scores were 4.2 points higher in the intervention group than in the comparison group (95% CI 1.5-6.9).
Use language that matches your design and effect measure.
Do not describe an association as causal merely because its interval is narrow.
A four-question confidence interval check
Whenever you see a confidence interval, ask:
1. Where is the point estimate?
What is the study's best single estimate under the model?
2. How wide is the interval?
How much uncertainty remains?
3. Does it include the null value?
Useful, but not the end of the interpretation.
4. What important values does it include or exclude?
Could the true effect still be large enough to matter?
That fourth question is often the most scientifically useful.
Open your main results table.
Choose one confidence interval and write four notes beside it:
- point estimate;
- width/precision;
- null value;
- practically important values still compatible with the interval.
If your current results narrative mentions only whether the interval “crosses one” or “crosses zero,” add the substantive interpretation.
Frequently asked questions
- What does a 95% confidence interval mean?
Under the standard frequentist interpretation, 95% describes the long-run coverage of the interval procedure. It is not strictly a 95% probability statement about the fixed parameter after your particular interval has been calculated.
- What does it mean if a confidence interval includes zero?
For a difference measure where zero is the null value, the interval includes values compatible with no difference under the model. Interpret the full interval rather than declaring “no effect.”
- What does it mean if a confidence interval includes one?
For ratio measures such as risk ratios or odds ratios, one is usually the null value. An interval including one is typically consistent with a corresponding two-sided p-value above 0.05 under matching assumptions.
- Is a narrow confidence interval always better?
A narrow interval indicates greater precision, but a precise estimate can still be biased by poor design, bad measurement or inappropriate modelling.
- Should I report confidence intervals if I already report p-values?
Usually yes. Major reporting guidance emphasizes estimates and their precision because they communicate more than a significance label alone.
Use the Methods Bench Manuscript Outline Planner to check whether your results section reports the estimate, confidence interval and p-value where appropriate.
References and further reading
- 1.Gardner MJ, Altman DG. Confidence intervals rather than P values: estimation rather than hypothesis testing. BMJ. 1986. https://www.bmj.com/content/292/6522/746
- 2.Greenland S, Senn SJ, Rothman KJ, et al. Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations. https://pmc.ncbi.nlm.nih.gov/articles/PMC4877414/
- 3.What Do P Values and Confidence Intervals Really Represent? https://pmc.ncbi.nlm.nih.gov/articles/PMC5811238/
- 4.BMJ. How to obtain the P value from a confidence interval. https://www.bmj.com/content/343/bmj.d2304
- 5.STROBE Statement. https://www.equator-network.org/reporting-guidelines/strobe/
Take it further
Better research, once a week.
Practical methods guidance, research tools and funding opportunities from Methods Bench.
Get practical research notes and new opportunities
Short, useful emails for researchers in Uganda and East Africa. No spam.