Error Bar Type Enforcement

A bar chart with error bars and no label is uninterpretable. SD describes individual variability. SEM describes the precision of the mean estimate. 95% CI defines a plausible range for the true mean. These are fundamentally different quantities —and a reader looking at the figure cannot tell which one is shown. Licklider requires an explicit declaration before any error bar can be displayed.

STEP 1 —The Pitfall: The Unlabeled Error Bar

A 2014 study in PLOS Biology examined 878 figures in a random sample of papers from top journals and found that fewer than 40% of figures with error bars included a label identifying the error bar type in the figure legend. The remainder showed error bars with no indication of whether they represented SD, SEM, or CI.

The confusion between SD and SEM is particularly consequential. SEM is smaller than SD by a factor of 竏嗜 —so for n = 9, SEM is exactly one third of SD. A figure using SEM visually appears to show much less variability than the same data displayed with SD. Reviewers and readers who assume they are looking at SD will dramatically underestimate the spread in the data. Reviewers who assume SEM when SD is shown will overestimate it.

• SD (Standard Deviation): Describes the spread of individual observations around the group mean. Appropriate when the reader needs to understand individual variability.
• SEM (Standard Error of the Mean): Describes the precision of the mean estimate. Appropriate when the reader needs to assess the reliability of the group mean —but does not convey individual variability.
• 95% CI: A range that, under repeated sampling, would contain the true mean 95% of the time. The most information-rich error bar for inferential purposes.

STEP 2 —Journal Requirement

Nature's figure preparation guidelines state: "In figures showing error bars, the legend must state clearly what the error bars represent." Cell requires that "all error bars must be defined in the figure legend." PLOS ONE requires "explicit statement of what error bars represent" as a condition of publication.

The British Medical Journal and JAMA both mandate that error bars representing SEM be accompanied by a note explaining the difference between SEM and SD, given the common confusion between the two. eLife's reporting checklist explicitly asks authors to confirm that "error bars are defined in figure legends."

There is also a substantive concern beyond labeling: many journals now prefer or require 95% CI over SEM for figures showing group comparisons, because CI directly supports inferential conclusions while SEM does not. Licklider surfaces this preference when SEM is selected for a figure with significance annotations.

STEP 3 —Licklider's Solution

Input

• An explicit selection of error bar type: SD / SEM / 95% CI / IQR / Range / None (with justification). The selection is required before error bars can be enabled on any figure.
• The selection is made in the figure configuration panel and is locked per figure. Changing the error bar type after setting significance annotations requires acknowledgment of the impact.

Output

• Labeled error bars on figure. The selected type is displayed as part of the figure itself (small label beneath the error bar, if space permits) and always in the auto-generated figure legend.
• SEM advisory. When SEM is selected for a figure with significance annotations or group comparisons, Licklider displays an advisory: "SEM describes the precision of the mean, not the spread of individual observations. For figures comparing groups, 95% CI is often preferred. Consider switching."
• Legend contribution. The error bar type label is automatically included in the auto-generated figure legend in journal-specific formatting (e.g., "Error bars, s.d." for Nature; "error bars indicate SD" for JAMA).

Guard

STEP 4 —Draft Output (Draft / Needs review)

Figure legend contribution —Nature format

Data are presented as mean ﾂｱ s.d. (n = 12 per group).
Horizontal lines indicate the group median.
Statistical comparison: two-sample Welch t-test (two-sided);
**p = 0.003; Cohen's d = 0.91 (95% CI [0.35, 1.47]).

Figure legend contribution —JAMA format

Data are presented as mean (SD). n = 12 per group.
Error bars indicate 1 SD. Two-sample Welch t-test (2-tailed);
P = .003; Cohen d = 0.91 (95% CI, 0.35-1.47).

Related Features