Appendix C: Chart Selection Cheatsheet
This cheatsheet distills the decision framework from Chapters 5 and 35 into a compact reference. Start with the question you are trying to answer, then select from the recommended chart types.
Decision Table: Question Type to Chart Type
| Question | Recommended Charts | Avoid |
|---|---|---|
| How is a single variable distributed? | Histogram, KDE plot, box plot, violin plot, strip plot | Pie chart, 3D bar |
| How do two distributions compare? | Overlaid KDE, side-by-side box plots, violin plot, back-to-back histogram | Stacked area (hard to compare) |
| How do many distributions compare? | Ridgeline (joy plot), violin grid, letter-value plot | Individual histograms (too many panels) |
| What is the relationship between two continuous variables? | Scatter plot, hexbin (large N), 2D KDE, regression plot | Line chart (implies ordered x) |
| What is the relationship among three+ continuous variables? | Pair plot, scatter matrix, bubble chart (3rd var as size), parallel coordinates | 3D scatter (hard to read) |
| How do values compare across categories? | Bar chart (horizontal for many categories), dot plot, lollipop chart | Pie chart with more than 5 slices |
| What are the parts of a whole? | Stacked bar (few time points), pie/donut (2--5 slices), treemap, waffle chart | 3D pie, exploded pie |
| How does a value change over time? | Line chart, area chart, slope chart (two time points) | Bar chart with many time points |
| How do multiple series change over time? | Multi-line chart (up to 5 lines), small multiples, stacked area (if parts-of-whole over time) | Spaghetti chart (too many lines) |
| What is the ranking? | Horizontal bar chart (sorted), bump chart (ranking over time), dot plot | Vertical bars for long category names |
| How are values distributed geographically? | Choropleth, graduated symbol map, cartogram, hexbin map | 3D terrain map for thematic data |
| What are the connections or relationships in a network? | Node-link diagram, adjacency matrix, arc diagram, chord diagram | Hairball networks (too many edges) |
| What is the flow or process? | Sankey diagram, alluvial plot, funnel chart | Circular flow with many nodes |
| How does this text corpus look? | Word cloud (exploratory only), bar chart of term frequencies, dispersion plot | Word cloud as sole analysis |
| What is the correlation structure? | Heatmap (correlation matrix), clustered heatmap, pair plot | Tables of numbers |
| How do groups differ across multiple dimensions? | Radar/spider chart (up to 8 axes), parallel coordinates, small multiples | Radar with 15+ axes |
| What is the composition at each level of a hierarchy? | Treemap, sunburst, icicle chart | Nested pie charts |
Quick Decision Flowchart
Start here and follow the branches:
Step 1 -- How many variables?
- One variable --> Step 2a
- Two variables --> Step 2b
- Three or more --> Step 2c
Step 2a -- One variable: what kind?
- Continuous --> Histogram, KDE, box plot
- Categorical --> Bar chart, waffle chart
Step 2b -- Two variables: what kinds?
- Both continuous --> Scatter, hexbin, regression plot
- One categorical + one continuous --> Bar chart, box plot, violin plot, strip plot
- Both categorical --> Heatmap (counts), grouped bar, mosaic plot
- One temporal + one continuous --> Line chart, area chart
Step 2c -- Three or more variables:
- All continuous --> Pair plot, parallel coordinates, bubble chart
- Mix of types --> Small multiples, faceted plots (use categorical for faceting)
- Hierarchical --> Treemap, sunburst
- Spatial --> Choropleth, symbol map
Audience and Context Modifiers
The table above gives defaults. Adjust based on context:
| Context | Adjustment |
|---|---|
| Executive audience | Prefer simple bar and line charts; add annotations with key numbers; reduce chart count per page |
| Technical / peer review | Show uncertainty (error bars, confidence bands); use precise axis labels; include sample sizes |
| Public / journalism | Use familiar forms (bar, line, map); add explanatory subtitles; highlight the main takeaway in the title |
| Exploratory (for yourself) | Use pair plots, faceted grids, and interactive tools; polish is not required |
| Print (grayscale possible) | Use shape and line-style encodings in addition to color; test in grayscale |
| Interactive medium (web / dashboard) | Prefer Plotly or Altair; add hover tooltips, zoom, and filter controls |
| Presentation (slides) | One chart per slide; large fonts; bold the takeaway; minimize axis detail |
Common Pitfalls
- Using a pie chart for more than five slices. Switch to a horizontal bar chart sorted by value.
- Dual y-axes. They invite misinterpretation of correlation. Use two stacked panels with aligned x-axes instead.
- Truncated y-axis on a bar chart. Bars encode magnitude by length; a non-zero baseline distorts the message. Truncation is acceptable for line charts where the goal is to show change.
- Rainbow colormaps on continuous data. Rainbow palettes (jet, hsv) are not perceptually uniform and fail for colorblind readers. Use viridis, plasma, or another perceptually uniform alternative.
- 3D charts for 2D data. The third dimension adds no information and obscures values through perspective distortion.