Make a clearer scientific graph

Start with the scientific question

Decide what the reader should compare before choosing colors or fitting a curve. A graph should usually answer one main question rather than display every available variable.

• Use a scatter plot to show the relationship between two measured variables.
• Use a line plot when observations have a meaningful continuous order, such as time.
• Use error bars when each point has measured or estimated uncertainty.

Make labels do real work

• Use axis labels that name the quantity and unit, such as Temperature (°C).
• Write a short, specific title. Avoid repeating the axis labels as “Y vs X” when the scientific relationship can be named.
• Give series meaningful labels instead of Data 1 or Column B.
• Keep tick labels, legends, and annotations readable at the exported image size.

Represent the evidence honestly

• Show uncertainty and sample size when they affect interpretation.
• Use logarithmic axes only when multiplicative changes are scientifically meaningful, and label them clearly.
• Do not remove points only because they weaken a trend. Review outliers relative to the model and document exclusions.
• Inspect residuals before accepting a fit. A high R² can coexist with systematic model errors or overfitting.

A fitted line is a model of the selected data, not proof of the underlying mechanism.

Use color and space carefully

Use distinct colors to separate meaningful series, but do not rely on color alone. Prefer a restrained palette with strong contrast, avoid crowded legends, and leave enough space around labels and data.

Before exporting, check the graph in light and dark surroundings and inspect the final PNG, SVG, or PDF at the size where readers will see it.

Final checklist

1. Can a reader identify X, Y, units, and groups without guessing?
2. Is the chosen graph type appropriate for the measurement and ordering?
3. Are uncertainty and excluded data handled transparently?
4. Does the fit remain plausible when residuals and validation evidence are considered?
5. Are text, markers, and legends readable at final size?