Bar Chart (Mean ﾂｱ Error)

The bar chart with error bars is the most commonly misused figure type in biology. A bar showing the mean with SEM error bars hides the distribution, makes non-normally distributed data look symmetric, and routinely misleads about the precision and variability of the measurement. Licklider applies conditional-use rules and mandatory guards before generating a bar chart.

STEP 1 —When to Use

Use a bar chart when: the mean is the primary quantity of interest (not the distribution); n is large enough (>= 30) that the mean is a reliable summary; you are comparing multiple conditions and space efficiency is required; the journal specifically requires bar charts (rare in modern high-impact journals).

Licklider's bar chart guard fires when: n < 10 and no individual data overlay is present; the error bar type is not labeled (SD, SEM, or CI are not interchangeable); the bars start at a non-zero baseline without explicit justification; bars are used when the data are ordinal or non-normally distributed with n < 30.

[Image placeholder: Four panels showing the same underlying data (n=8, bimodal distribution) as: (1) bar + SEM —looks clean and symmetric. (2) bar + SD —shows more variability. (3) bar + SEM with data points overlaid —reveals bimodal structure. (4) violin + points —shows the full truth. All have identical bar heights but communicate radically different information.]

The same data, four representations. Bar + SEM (top left) hides everything that matters. Violin + points (bottom right) shows it all.

STEP 2 —Journal Requirement

Error bar type must be labeled explicitly: "Error bars represent SEM" or "Error bars show 95% CI." SD, SEM, and CI are not interchangeable and have different meanings.
n per group must be stated in the legend.
Nature Methods, Cell, and eLife require individual data points overlaid for n < 20.
Zero-baseline: if the y-axis does not start at zero, this must be clearly visible (broken axis) and stated in the legend.
Statistical test, p-value, and effect size required for any inferential comparison.

STEP 3 —Licklider Implementation

Input

Continuous outcome and group variable
Error bar type selection: SD (default for descriptive), SEM (for precision of mean estimate), 95% CI (recommended)
Individual data overlay: auto-enabled for n < 20

Output

Bars from zero baseline (configurable, with axis break notation)
Error bars with type explicitly labeled in legend and figure
Individual data overlay (jitter/beeswarm) for n < 20 —cannot be disabled for n < 10
n per group labeled
Significance annotation with p-value and effect size
Error bar type mismatch guard: if SEM is selected but SD is reported in the legend, the guard flags the inconsistency

Guard Summary

n < 10 without individual data overlay xBlocked
Error bar type unlabeled xBlocked
SEM displayed when describing natural variability xWarning (SEM describes precision of the mean, not variability of the data)

[Image placeholder: Licklider bar chart with 3 groups (n=22 each). Shows bars with 95% CI error bars, individual data points overlaid (small transparent dots), n=22 labeled below each group, significance bracket, and legend stating "Error bars: 95% CI. Individual data points overlaid (n=22 per group)."]

Licklider bar chart: 95% CI error bars (labeled), individual data overlay for n=22, significance bracket with effect size.

STEP 4 —Draft Output (Draft / Needs review)

Bar heights show group means; error bars represent 95% confidence intervals
of the mean. Individual data points are shown overlaid (n = 22 per group).
Statistical comparison by one-way ANOVA with Holm post-hoc correction:
F(2, 63) = 7.4, p = 0.001; pairwise comparisons shown in figure.
Effect size: eta² = 0.19 (95% CI [0.04, 0.34]).