Chart Chooser

Chart Chooser

Overview

This skill turns a dataset plus an analytical question into the right chart, and ensures the result is clean, accurate, and not misleading. It covers two jobs:

1. Choose — map the question intent and data shape to a chart type.
2. Polish — apply encoding, labeling, axis, color, and honesty rules so the chart reads correctly at a glance.

Keywords: data visualization, chart type, graph selection, bar chart, line chart, scatter plot, histogram, heatmap, pie chart, dashboard, axis truncation, misleading charts, color accessibility, data-ink, encoding.

Use this skill whenever someone needs to decide how to show data, or wants an existing visual critiqued.

Workflow

Follow these steps in order. Do not jump to a chart type before clarifying intent and data shape.

1. Identify the question intent. Pick the dominant one (a chart should answer one question well):
   • Comparison (which is bigger/smaller across categories)
   • Trend / change over time
   • Distribution (shape, spread, outliers of one variable)
   • Relationship / correlation (two+ numeric variables)
   • Part-to-whole (composition of a total)
   • Ranking (ordered comparison)
   • Geospatial (values across places)
   • Flow / part-to-whole over stages
2. Profile the data shape. Note for each variable: type (categorical / ordinal / quantitative / temporal / geographic), cardinality (number of distinct values), and whether values can be summed to a meaningful total (needed for part-to-whole).
3. Match intent + shape to a chart. Use the decision table in references/chart-decision-matrix.md. Resolve ties with the "pick the simplest that answers the question" rule.
4. Choose encodings. Assign data fields to position, length, color, and size using the accuracy ranking in Best Practices. Position/length beat color/area for quantitative accuracy.
5. Apply honesty + clarity rules. Run the checklist in references/clarity-and-honesty-checklist.md: axis baselines, sorting, direct labels, color accessibility, decluttering.
6. Produce the output. Either describe the spec (chart type, x, y, color, sort, annotations) or generate code. Use scripts/suggest_chart.py to get a programmatic recommendation from a quick data profile.
7. Self-review. Verify the chart answers the original question in under 5 seconds and contains no misleading element.

Quick Decision Cheats

• Over time, continuous: line chart (one or few series), area only when part-to-whole over time matters.
• Compare categories: bar chart (horizontal if labels are long or many). Sort by value unless order is inherent (time, age bands).
• Distribution of one number: histogram; box/violin to compare distributions across groups.
• Two numbers, relationship: scatter plot; add trend line only if a relationship is real and labeled.
• Part-to-whole, few parts (2-4), single total: consider a stacked bar; a pie is acceptable only with ≤4 slices and clear differences.
• Part-to-whole, many parts: sorted bar of the parts, not a pie.