Posters · How it works
How to show a confidence interval on a chart
To show a confidence interval on a chart, draw an error bar from the lower to the upper limit of the interval around each estimate, and say in the caption that the bars are 95 percent confidence intervals. Error bars can also show standard deviations or standard errors, so without that caption they cannot be read.
Error bars are among the most misread marks in science, partly because three different quantities share one symbol. Getting them right on a poster means choosing the quantity deliberately, labelling it and knowing what an overlap does and does not tell a reader.
Indunil Asanka · Co-founder
6 min read · Published
| Bar type | What it shows | Changes with sample size | Use it to |
|---|---|---|---|
| Standard deviation (SD) | Spread of individual values; about two thirds of values lie within one SD of the mean | No, it describes the data | Describe how variable the measurements are |
| Standard error (SE) | Precision of the mean; SD divided by the square root of n | Yes, shrinks as n grows | Rarely on its own; it looks precise but is hard to interpret |
| 95% confidence interval (CI) | Range of values compatible with the data at the 95 percent level, roughly the mean plus or minus two SE | Yes, narrows as n grows | Show how precisely an estimate is known; the usual choice for inference |
- 1
Decide what the bars must say
Use confidence intervals when the reader should judge how precise an estimate is, and standard deviations when the point is how variable individuals are. Do not use standard errors just because the bars look smaller.
- 2
Calculate the interval for each plotted estimate
Compute the lower and upper limits from the analysis you report, so the chart matches the numbers in the results text and the key number.
- 3
Draw the bars and keep the axis honest
Draw a whisker from lower to upper limit with small caps. Start a bar chart's axis at zero, and consider dots with intervals instead of bars when zero is not meaningful.
- 4
Label the bars in the caption
Write, for example, Bars show 95% confidence intervals, n = 160 per group. State whether groups are independent or repeated measures.
- 5
Put the key comparison in numbers too
Report the difference with its own interval beside the chart, because overlap between two groups' intervals is not the same as the interval of their difference.
Three quantities, one symbol
Cumming, Fidler and Vaux's paper Error bars in experimental biology, in the Journal of Cell Biology, starts from a finding that many researchers could not say what error bars in published figures represented. The first rule they give is to describe in every caption what the bars are. Standard deviation bars are descriptive: roughly two thirds of observations fall within one SD of the mean and about 95 percent within two. Standard error and confidence interval bars are inferential: they show how precisely a mean has been estimated. Krzywinski and Altman's Error bars column in Nature Methods makes the same distinction and warns that the three are often confused.
What overlap does and does not tell you
A common belief is that if two confidence intervals overlap, the difference is not significant. That is wrong. For two independent groups each with at least ten observations, Cumming and colleagues give rules of eye. With 95 percent confidence intervals, an overlap of about half the average arm length corresponds to p of roughly 0.05, and bars whose tips just touch correspond to p of about 0.01. With standard error bars, a gap between the bars equal to one standard error corresponds to about p = 0.05. So intervals can overlap moderately while the difference is still significant.
When the rules do not apply
The rules of eye hold for independent groups only. For repeated measurements on the same people, such as before and after, the bars on each time point say nothing reliable about the change, because the correlation between measurements is not shown. Cumming and colleagues are explicit that error bars on the individual means are irrelevant to comparisons within the same group. In that situation, plot the within person differences with their own interval, or report the paired difference and its interval in text next to the chart. The same caution applies to clustered or matched designs, where the analysis accounts for structure that a simple bar per group cannot show; the chart should then carry the model's estimates, not raw group means with naive intervals.
Choosing between bars, dots and bands
A bar chart with whiskers is familiar, but the bar itself adds ink that says little when zero is not a meaningful baseline, such as mean blood pressure or temperature. A dot for each estimate with a line for its interval carries the same information, keeps the eye on the estimate and its uncertainty, and works for ratios, where the reference line sits at one. For estimates over time, a line with a shaded band reads well, provided the band is labelled and no more than two or three bands overlap. Whatever the form, order groups meaningfully, keep the axis label specific, and make the interval level visible in the caption.
Common mistakes
Leaving bars unlabelled is the most frequent error on posters. Showing standard errors because they look tidier than confidence intervals misleads by making estimates look more precise than they are. Drawing bars on a truncated axis exaggerates differences. Using overlapping intervals as a significance test gets the answer wrong in both directions. And putting intervals on the chart but a different set of numbers in the results text, such as unadjusted means on the chart and adjusted estimates in the text, confuses readers comparing the two. A last one is drawing intervals so thick or so pale that they cannot be seen from a metre away, which on a conference board is the distance that matters; a thin dark line with short caps reads best.
Where it shows up in the poster builder
None of the ten chart kinds a poster figure offers draws error bars or interval whiskers. A figure's data is a set of categories and one or more series of values, with no field for a lower and upper limit. So on a board built here the interval has to be carried in words: the key number block, on 27 layouts, has a detail line designed for the interval, the figure caption can state it, and the results text can give each estimate with its limits. A chart with intervals made in statistics software cannot be dropped into a chart figure, because image figures cannot be filled. The human geography example shows the pattern: its hero figure plots commute time, and the key number's detail line carries the 95 percent interval of 7.4 to 10.6 minutes for the nine minute saving.
Questions people ask
Should I use 95 percent or 99 percent intervals on a chart?
Use the level your analysis reports, which is 95 percent in most fields. Mixing levels across a board confuses readers. If the study prespecified a different level, such as 90 percent for an equivalence test, state it in the caption and in the methods. Never switch levels between charts to make one comparison look cleaner than another.
Can I show confidence intervals on a line chart?
Yes. Draw whiskers at each point or a shaded band between the lower and upper limits across time. Bands work well when there are many points, but they can hide individual values and overlap confusingly when several lines share one chart. Limit a chart to two or three lines with bands, and label the band in the caption.
What are the whiskers on a box plot?
They are not confidence intervals. In a standard box plot the box spans the middle half of the data and the whiskers usually reach to the most extreme values within 1.5 times the box's height, with points beyond shown individually. Box plots describe distributions. If a box plot appears on a poster, say in the caption how its whiskers were defined.
How many observations do I need before error bars mean much?
With very small groups, such as three replicates, error bars are unstable and confidence intervals become wide. Cumming and colleagues suggest showing the individual data points when n is small, since three dots tell the reader more than a bar and a whisker. Many journals now ask for individual points to be plotted alongside summary statistics.
Should I put intervals on proportions in a bar chart?
Yes, when the proportion is an estimate for a wider population, such as the share of survey respondents who agree. Use an interval method suited to proportions, particularly when values are near zero or one hundred percent or samples are small, where the simple formula can produce impossible limits below zero or above one hundred.
What if my chart has too many bars for whiskers to be readable?
Reduce the chart before adding clutter. Show the main comparison in the chart and move secondary groups to a table or behind a QR code. Alternatively, switch from bars to dots with intervals, which carry the same information with much less ink and remain readable with many categories.
Make one with posters
The button opens the generator with this use case already described. Change the wording to match your own.
Create a poster with OneCraftRelated questions
- What is a p value?What is a p value: how incompatible data are with a null model, not the chance a finding is true. What 0.05 means and how to report one on a poster.
- What is a forest plot?What is a forest plot: a chart of each study's effect and interval with a diamond for the pooled result. How to read one, and how to show one on a poster.
- What is effect size?What is effect size: how big a difference or association is, not whether it passed a significance test. Cohen's d, risk and odds ratios, and how to report them.
Step by step in the builder: Add charts and diagrams.
Written and checked by the OneCraft team. Last checked .