Free graph maker & data converters

Error Bars: SD, SE or Confidence Interval?

By Plotiq Editorial Team · Published 14 November 2026 · Updated 14 November 2026 · 4 min read

Error bars look simple but mean different things depending on what they are calculated from. Standard deviation, standard error and confidence intervals answer different questions, and a graph without a statement of which one it shows cannot be interpreted. This guide explains each and when to use it.

What an error bar is trying to tell you

An error bar is a short line through a data point or bar that shows how uncertain or variable that value is. The bar has no meaning on its own. It might describe the spread of the individual measurements, the uncertainty in the mean, or a range you are 95% confident contains the true value. These are very different statements, and they differ in length by a factor of two or more for the same data. That is why every figure with error bars needs a legend or caption that says what they are.

Standard deviation: how spread out the measurements are

The standard deviation (SD) describes the spread of individual measurements around their mean. For n values the sample SD is the square root of the sum of squared deviations from the mean divided by n − 1. It does not shrink as you collect more data; it estimates a property of the thing you are measuring, such as natural variation between organisms or the precision of a method.

Use SD when you want the reader to know how variable the data are. About 68% of values in a roughly normal distribution lie within one SD of the mean and about 95% within two. A calibration standard measured three times with an SD of 0.004 tells you the readings repeat to about that level.

Standard error: how well the mean is known

The standard error of the mean (SE or SEM) is the SD divided by the square root of n. It describes how precisely the mean has been estimated, and it shrinks as n grows: quadrupling the sample size halves the SE. It does not describe the spread of the data.

This is the source of a widely cited misuse. Because SE is smaller than SD, it makes the bars look tighter, and some authors choose it for that reason. SE is legitimate when the point of the figure is the precision of the means, but it should not be used to suggest that the underlying data are less variable than they are. Whichever you pick, state n.

Confidence interval: a range for the true mean

A 95% confidence interval (CI) is the mean plus or minus a multiple of the SE, using a critical value from the t distribution: roughly 2 for large samples, but 2.78 for n = 5 and 4.30 for n = 3. For small samples the CI is therefore much wider than ±1 SE. A 95% CI is a range that, if you repeated the whole experiment many times, would contain the true mean in 95% of the repetitions.

Confidence intervals are the most informative choice when you want to compare means, because non-overlap and overlap carry meaning that is easier to interpret than SE bars. Many medical and psychology journals prefer them.

A worked example with five measurements

Suppose five replicate measurements of a sample give 9.8, 10.2, 10.0, 10.4 and 9.6. The mean is 10.00. The deviations from the mean are −0.2, 0.2, 0, 0.4 and −0.4, with squares summing to 0.40. Dividing by n − 1 = 4 gives a variance of 0.10, so SD = 0.316. SE = 0.316 / √5 = 0.141. The 95% CI uses t = 2.776 for four degrees of freedom: 2.776 × 0.141 = 0.39, so the interval is 10.00 ± 0.39.

MeasureValueError bar drawn
Mean10.00—
SD0.3169.68 to 10.32
SE0.1419.86 to 10.14
95% CI±0.399.61 to 10.39

Same data, three different bar lengths. The caption must say which one is shown.

Which one should you use?

Use SD to describe variability in your data, for example when showing replicate readings or a population. Use SE or a 95% CI when the figure is about comparing means or how accurate an estimate is. If you are not sure, SD is the safer default because it is the honest description of the data and the reader can derive SE from it if they know n.

Do not use error bars at all for n = 1; there is nothing to estimate. With n of 2 or 3, consider showing the individual points as well, since a mean with an SD from three values is a weak summary. Most journals now ask for this.

How to add them in Plotiq

Put your error values in a column beside the measurements and name it SD, SE or Error; the tool pairs it with the column before it. Alternatively choose a fixed value or a percentage of each value. The figure below shows the calibration standards with their SDs as error bars on a scatter plot. Remember to write what the bars represent in your figure caption.

0246810120.00.10.20.30.40.50.60.7Concentration (mg L-1)Absorbance
Each point is a mean of three readings; bars show ±1 SD. Made with the Error Bar Graph Maker.

Frequently asked questions

Do overlapping error bars mean the difference is not significant?
Not reliably. SD bars overlapping tells you little. For 95% CIs, overlap of up to roughly a quarter of the bar length can still be a significant difference. Use a proper statistical test for significance.
Should I plot SD or SE in a bar chart?
SD if you want to show variability; SE or CI if the point is the precision of the means. Say which in the caption and give n.
What do I do when n is very small?
Show the individual data points on the graph in addition to, or instead of, the bars. With n = 3 a mean and an error bar hide more than they show.
How do I calculate SD in Excel?
Use STDEV.S for a sample (divides by n − 1) and STDEV.P only if you have the whole population. SE is =STDEV.S(range)/SQRT(COUNT(range)).

Keep reading

Written and reviewed by the Plotiq Editorial Team. Spotted a mistake? Tell us and we will correct it. See About Plotiq for how these guides are made.