The problem with bar charts
Bar charts are the default in most biology papers, but they hide more than they show. A bar whose height represents the mean tells you nothing about the distribution of the underlying data, whether it is symmetric, skewed, or whether one outlier is pulling the mean up.
Switch between the views below: the same two groups of ELISA readings, drawn three ways. The bar gives no hint of the outlier the dot plot makes obvious.
The bar shows only the mean ± SD. The treated group's outlier at 1.35 is completely hidden.
If you have fewer than 10 data points per group, show every individual point. A dot plot or strip chart communicates the same mean while being honest about sample size.
Matching the graph to your data
| Graph type | Best for | Avoid when |
|---|---|---|
| Bar + error bar | Large n (>20), normally distributed | Small n, where the bar hides the data |
| Box plot | n ≥ 5, showing spread and outliers | Very small n, where boxes look precise but aren't |
| Dot plot / strip chart | Any n, especially small | Hundreds of points, which become unreadable |
| Violin plot | Large n, comparing distribution shapes | n < 10, where the shape is meaningless |
What error bars actually mean
Error bars without a label are useless. Always specify which measure you are displaying:
| Error bar type | What it shows | When to use it |
|---|---|---|
| SD (standard deviation) | Spread of the data | Describing the sample distribution |
| SEM (standard error of the mean) | Precision of the mean estimate | Generally discouraged as a default; it's the smallest bar and easily misleading, so prefer a 95% CI |
| 95% CI | Range likely to contain the true mean | Inferential comparisons between groups; effect size |
Toggle the error-bar measure on the same means and watch the whiskers shrink. The data is identical; only the claim changes.
SD. SD describes the spread of the data itself, the widest bars here.
SEM is always smaller than SD and can make noisy data look precise. Many journals now require 95% CI or raw data points instead.
Representing paired data
If your experimental design pairs samples (the same animal before and after treatment, or matched donors), the graph must reflect the pairing. An unpaired bar chart throws away the most important structure in your data.
For paired data, connect the individual points with lines between conditions. The slope of each line communicates the direction and magnitude of the individual effect far better than two separate bars.
Each line is one donor. Most rise, but D7 falls; pairing exposes individual variation. Hover a line to read its change.
Colour and accessibility
Avoid using red/green combinations as the primary way to distinguish groups, since approximately 8% of men have red-green colour blindness. Use:
- Shape and fill pattern alongside colour
- Colour palettes designed for colour-blind readers (e.g. Okabe-Ito)
- Direct labels on the chart rather than relying solely on the legend
Exporting from Platelet
Every graph in Platelet can be exported as a high-resolution PNG or SVG. The SVG format is vector-based and scales without loss for journal figures. Use the Graphs tab to configure axis labels, units, and colour scheme before exporting.