Scientific Figures & Visualization — Craft Guide

Figures carry a paper's evidence. Make them accurate, clear, accessible and reproducible. A misleading or unreadable figure undermines an otherwise sound study.

1. Choose the chart for the message + data

• Distribution: histogram, box plot, violin/strip/beeswarm (show the raw points for small n).
• Comparison across groups: bar with error bars/CIs, or dot/forest plots (cleaner than many bars).
• Relationship: scatter (+ regression line & CI), bubble (3rd var by size), hexbin for dense data.
• Trend / time / energy series: line with uncertainty band; small multiples for many series.
• Composition: stacked/100% bar; avoid pie charts (hard to compare) except for 2–3 large slices.
• Part-to-whole over categories: grouped bars or heatmap.
• Multidimensional: small multiples (faceting), parallel coordinates, Pareto front scatter for multi-objective optimization.
• Networks: node-link (bibliometric maps, BIM/data graphs) with meaningful layout — see bibliometric-analysis.