SD vs SEM Error Bars (and Confidence Intervals) Explained
Two bar charts can show identical means with error bars of very different lengths, and both can be correct, because they are drawing different things. The choice between SD vs SEM error bars, or a confidence interval, decides what a reader can conclude from a figure. Picking the wrong one, or failing to say which one you used, is among the most common presentation errors in lab reports and papers. This guide explains what each bar represents, when to use it, and how to read one without falling for the usual traps.
Describing data versus describing the estimate
Error bars answer one of two questions. The first is descriptive: how spread out are the individual measurements? The second is inferential: how precisely have we estimated the mean? Standard deviation answers the first. Standard error and confidence intervals answer the second. Mixing up the two questions is the root of most confusion.
Standard deviation: the spread of the data
The SD measures how far individual values typically sit from the mean. For roughly normal data, about two thirds of values fall within one SD of the mean and about 95% within two. Crucially, the SD does not shrink as you collect more data; it settles toward the true variability of the system.
Use SD when the point of the figure is to show variability itself: how consistent a cell line is from experiment to experiment, how much a plate reader varies across wells, or how much purity results differ between vials of a lot.
Standard error of the mean: precision of the average
The SEM is the SD scaled down by √n, where n is the number of independent measurements. It describes how much the sample mean would be expected to vary if the experiment were repeated many times. Because it divides by the square root of n, it gets smaller as the sample grows, even if the underlying spread is unchanged.
This is why SEM bars are popular: they are always shorter than SD bars and make data look tidier. That is also why they are criticised. An SEM bar alone does not show the reader how variable the individual measurements were, and with small n it can give a misleading sense of precision.
Confidence intervals: the most direct inferential choice
A 95% confidence interval gives a range of values for the true mean that is compatible with the data, under the model’s assumptions. For large samples it is roughly the mean plus or minus two SEMs. For small samples it is wider, because it uses the t-distribution: with n = 3, the multiplier is about 4.3 rather than 2.
That small-sample widening is the reason many statisticians prefer confidence intervals over SEM for typical lab experiments. With three biological replicates, SEM bars can look narrow while the corresponding 95% interval spans a far wider range.
SD vs SEM error bars at a glance
| Bar type | Answers | Changes with n? | Best used for |
|---|---|---|---|
| SD | How variable are individual measurements? | No (stabilises) | Showing spread, assay consistency, lot-to-lot variability |
| SEM | How precise is the mean estimate? | Yes, shrinks with √n | Rarely the best choice; acceptable if n is stated and large enough |
| 95% CI | What range of true means is compatible with the data? | Yes, shrinks with n | Comparing groups and reporting effect sizes |
Reading error bars without being misled
Error bars invite visual comparisons, and several common shortcuts are wrong:
- Overlapping SEM bars do not prove the means are similar, and non-overlapping SEM bars do not prove a significant difference. With small n, gaps between SEM bars can occur by chance.
- Overlapping 95% CIs do not rule out a significant difference either. Two intervals can overlap moderately while a formal test still finds p below 0.05.
- Bars on paired or repeated data can hide the real comparison. If each experiment includes both control and treatment, the relevant variability is in the paired differences, which the bars on separate groups do not show.
- Normalised data can erase variability. When every experiment’s control is set to 100%, the control bar has no spread at all, and the treated bar’s spread absorbs everything.
Counting n correctly
Every error bar depends on n, so it matters what was counted. If the bars were calculated from technical replicate wells of a single experiment, both SEM and CI will be misleadingly small, because they describe measurement precision rather than reproducibility. Bars should normally be built from independent biological replicates, with technical replicates averaged first.
A reporting checklist
- State in the legend whether bars show SD, SEM or a confidence interval, and at what level.
- State n and what one unit of n represents.
- Where n is small, plot the individual points on top of the bars.
- Prefer SD to show spread, and CI to show uncertainty in a mean or a difference.
- For comparisons, consider showing the difference between groups with its own confidence interval.
Spread as a quality signal for incoming material
The same thinking applies outside the experiment. When a lab receives a large, multi-vial order, the relevant question for QC is often the spread between vials rather than the average. Reporting the SD of any in-house checks across vials of one lot gives a clearer picture of consistency than a single mean. Bulk Peptides products are third-party tested for purity by HPLC, certificates are published for some products, and each vial’s cap and crimp colour ties it to its certificate, so vial-level results can be logged against the correct lot.
Bulk Peptides sells research peptides for in-vitro laboratory use only. They are not intended for human or animal use, and this article is a guide to presenting laboratory data.
