A scientific figure should represent data and uncertainty accurately, not decorate a result. Automated plotting can accelerate exploration, but defaults may conceal distributions, exaggerate differences, or apply unsuitable aggregation. Generated imagery is not a substitute for experimental images or unobserved data.
When this workflow is useful
- Exploring distributions, relationships, time, or space
- Preparing figures for papers and reports
- Building graphics that regenerate with the analysis
A practical sequence
- 01
Define the comparison
State whether the figure compares distributions, shows a trend, explains structure, or communicates uncertainty. Give each figure one principal analytical purpose.
- 02
Match encoding to data
Choose position, length, colour, or shape according to variable type and scale. Treat truncated axes, areas, and three-dimensional effects cautiously.
- 03
Show observations and uncertainty
Where feasible, expose observations, sample size, distribution, and an appropriate interval. Define error bars, aggregation, normalization, and statistical treatment.
- 04
Produce reproducibly and review
Generate final plots from data and code, preserving transformations and export settings. Check colour accessibility, type size, units, captions, and numerical consistency.
Quality-control questions
- Could visual encoding exaggerate a difference?
- Are sample size, aggregation, and error definitions clear?
- Can the figure be regenerated?
- Could generated material be mistaken for an observation?