Standard Deviation Error Bars in Charts: SD vs SE vs CI

Quick Answer

Use standard deviation (SD) error bars when the chart should show how spread out the raw observations are. Use standard error (SE) error bars when the chart should show uncertainty in the estimated mean. Use confidence interval (CI) error bars when readers need a plus-or-minus range around the mean at a stated confidence level.

A student or analyst often reaches this question after making a bar chart of group means and realizing the software can draw several kinds of error bars. My role as a statistics educator is to make the chart answer one specific question: are we showing variation among observations, precision of the mean, or a confidence interval for the population mean?

If you need the raw spread first, calculate it with the mean and standard deviation calculator or the sample standard deviation calculator. If the chart is about uncertainty in a mean, pair this guide with Standard Error vs Standard Deviation, Building Confidence Intervals, and the standard error of the mean calculator.

What Error Bars Answer

The same dataset can produce three valid-looking charts with three different messages. The mistake is not the arithmetic; it is choosing an error bar that answers the wrong question.

For charts aimed at nontechnical readers, 95% CI bars are often easier to defend than SE bars because the caption can explain the interval directly. SD bars are still the right choice when the visual purpose is spread, such as comparing variability in test scores, response times, fill weights, or lab replicates.

SD vs SE vs CI

Worked Example

Here is a concrete teaching dataset we use when reviewing chart captions. An analyst compares completion times, in seconds, for two checkout flows. Each flow was tested by eight users.

This example also shows why SE bars can be misleading in presentations. They are much shorter than SD bars, so a casual reader may think the raw data are tightly clustered. The smaller SE only means the sample mean is estimated more precisely. For more on that distinction, see Margin of Error vs Standard Error.

How to Label the Chart

A good caption removes ambiguity before the reader has to guess. Put the statistic, sample size, and interval level directly in the figure note.

• For SD bars:Bars show mean +/- 1 sample standard deviation; n = 8 users per flow.
• For SE bars:Bars show mean +/- 1 standard error; n = 8 users per flow.
• For CI bars:Bars show 95% confidence intervals for the mean using the t distribution; n = 8 users per flow.

Do not infer too much from visual overlap alone. Two 95% CI bars that overlap slightly can still correspond to a statistically detectable difference, and two SD bars that overlap tell you very little about the difference between means. If the decision is formal, use the chart as a summary and report the test or interval that matches the study design.

Reporting Checklist

• State whether bars show SD, SE, or CI; never rely on the visual alone.
• Report n for every group, especially when group sizes differ.
• Use SD bars for raw variability and CI bars for inference about means.
• Use t-based CI bars for small samples unless a known population standard deviation justifies a z interval.
• Avoid bar charts for small samples when dot plots or boxplots would show the data pattern more honestly.
• Check whether outliers, skew, paired observations, or unequal variances change the appropriate analysis.

Weakest Section Rewrite

Weak version: "Error bars help show uncertainty in charts."

Concrete version: "In the checkout-flow chart above, Flow A's SD bar is +/- 2.66 seconds, its SE bar is +/- 0.94 seconds, and its 95% CI half-width is +/- 2.22 seconds. Those three bars answer different questions, so the caption must name the one used."

That concrete substitution is the standard to use in your own reporting: name the statistic, show the number, and connect the visual choice to the decision.