Group Comparison Mean and SEM

A group comparison mean figure shows the mean of each group alongside an error bar that communicates either the precision of the mean estimate or the spread of the underlying data. It is one of the most commonly used figures in life sciences — and one of the most commonly misinterpreted.

Choosing the right error bar type is not a stylistic decision. It determines what your figure actually claims about your data.

Error bar types

Licklider supports three error bar types. They answer different questions and should not be used interchangeably.

SEM is not a measure of data spread. A small SEM means the mean is estimated precisely — not that the observations are tightly clustered. As n increases, SEM shrinks regardless of whether the data become more or less variable. Figures using SEM can appear more precise than the underlying data warrants.

SD describes your sample, not your inference. It does not shrink with larger n in the same way SEM does. Use SD when the biological variability itself is the quantity of interest.

95% CI is the most interpretable for inference. It shows the range of mean values consistent with your data, and its width reflects both sample size and variability.

How Licklider calculates each

The 95% CI uses the t-distribution with n−1 degrees of freedom. For small samples this produces wider intervals than the normal approximation (1.96 * SEM), which is the statistically correct behavior. When n = 1, CI is undefined and not displayed.

Licklider can calculate these error bars from the visible numeric data, but that does not mean the mean is always the right summary. If the data are strongly skewed, dominated by a few extreme values, or not independent, a correctly computed SEM, SD, or CI can still support a misleading figure.

Changing the error bar type

The error bar type can be changed in two ways:

• Inspector — select SEM, SD, or 95% CI directly from the error bar control when a bar or dot chart is active
• Chat — describe the change in plain language

The figure and the Inspector summary table update immediately when the selection changes.

Chart style: bar vs dot

The same mean and error bar can be displayed as:

• Bar chart — a filled bar extending from zero to the mean, with the error bar on top
• Dot chart — a point at the mean with the error bar extending above and below

Dot charts are generally preferred in modern scientific publishing. Bar charts imply that the area of the bar is meaningful, which it is not for most continuous outcomes — the baseline of zero is arbitrary. Use bar charts only when zero is a meaningful reference point for your outcome variable.

Neither bar nor dot style reveals the full shape of the distribution. If the groups are multimodal, strongly skewed, or contain clustered substructure, the same mean and error bar can hide very different raw data patterns.

Chart style can be changed in the Inspector or via Chat.

Individual data points

Raw data points can be overlaid on either chart style. When n per group is small, showing individual points is strongly recommended — error bars alone can obscure the actual distribution of the data.

What the Inspector shows

When a group comparison mean figure is active, the Inspector displays a summary table with the following statistics for each group:

All five statistics are shown regardless of which error bar type is currently displayed. This allows direct comparison and reporting without switching between error bar types.

Sample size and warnings

Licklider checks sample size before displaying this figure.

These thresholds are display heuristics for stability and readability, not universal scientific cutoffs. They are meant to reduce cases where mean-based summaries look more precise than the visible data can support.

SD with small n triggers an additional warning. When SD is selected and sample size is below the admissibility threshold, Licklider surfaces the mean_sd_bar gate, which explains the limitation and offers alternatives. SD-based error bars with small n are visually indistinguishable from larger-n results but carry substantially more uncertainty.

SEM and 95% CI with small n also trigger a warning at the same thresholds. Small-n SEM bars appear narrow and precise regardless of the actual data variability.

Licklider does not automatically determine from this figure alone whether the rows are independent, whether a few outliers dominate the mean, or whether a median- and distribution-based figure would be more honest than a mean summary. Those are interpretation questions that may require the linked quality checks or a different figure type.

Reporting

When reporting results from this figure, state the error bar type explicitly. Licklider generates a methods text draft that includes this language automatically.

Recommended format:

Data are presented as mean ± SEM (n = [n per group]).

Data are presented as mean ± SD (n = [n per group]).

Data are presented as mean with 95% confidence intervals, calculated using the t-distribution with n−1 degrees of freedom (n = [n per group]).

When to use this figure

Use mean with error bars when:

• Your analysis is based on group means (t-test, ANOVA)
• You want to communicate the precision of the mean estimate (SEM or CI) or the spread of the data (SD)
• n per group is sufficient to estimate the mean reliably
• The mean is a defensible summary of the underlying outcome

Consider alternatives when:

• n per group is small (< 10) — show individual data points with a strip plot or dot plot
• You want to show the full distribution — use a box plot or violin plot
• You want to show both distribution and raw data — use a raincloud plot
• The outcome is non-normal and medians are more appropriate — use a box plot
• The data are repeated, clustered, or otherwise non-independent — review the independence and repeated-measures checks before using a mean summary as if each row were an unrelated observation

Design rationale and references

Licklider keeps SEM, SD, and 95% CI separate because they answer different scientific questions: precision of the mean estimate, spread of the observed data, and inferential uncertainty around the mean. The page is written to prevent the common mistake of reading these bars as interchangeable visual decorations.

Licklider also uses the t-distribution for 95% CI because small-sample mean uncertainty should widen when degrees of freedom are limited. That keeps the interval aligned with standard small-sample inference rather than making it look artificially precise.

The small-n warnings and alternative-figure suggestions are display guardrails, not claims that mean summaries become mathematically invalid at a single cutoff. They are intended to reduce cases where bar-or-dot summaries hide sparse, skewed, or irregular raw data that would be better shown directly.

References

1. Cumming, G., Fidler, F., & Vaux, D. L. (2007). Error bars in experimental biology. The Journal of Cell Biology, 177(1), 7-11. https://doi.org/10.1083/jcb.200611141
2. Weissgerber, T. L., Milic, N. M., Winham, S. J., & Garovic, V. D. (2015). Beyond bar and line graphs: time for a new data presentation paradigm. PLOS Biology, 13(4), e1002128. https://doi.org/10.1371/journal.pbio.1002128

What this page does not cover

• Statistical tests for group comparison — see t-Test, One-Way ANOVA and Post Hoc, Non-parametric Alternatives
• Showing individual observations — see Strip Plot, Dot Plot
• Distribution-based figures — see Box Plot, Violin Plot
• Error bar reporting requirements — see Error Bar Type Enforcement, Effect Size, CI, and N Reporting