Charts provide efficient ways to communicate complex information. A chart earns its place when the reader’s question is comparative — bigger than what, changing how, clustered where. Choosing a chart is choosing which comparisons become effortless and which become impossible.
No single rule picks the chart. Four things matter: what the reader needs to do, what the data looks like, how much of it there is, and how it is drawn. They are independent — start from whichever is clearest.
What the reader needs to do
Two ways to name the goal. The first breaks it down into what the reader is looking for and what they will do with it; the second just names what the chart is about. Use whichever comes to mind first.
Looking for what, to do what
| Target | Aim | Query |
|---|---|---|
| Trends — overall shape or direction | Explain — communicate a finding | Compare — magnitudes, ranks, change |
| Features — specific patterns or peaks | Discover — explore for unknowns | Identify — locate a specific value or outlier |
| Outliers — anomalies and exceptions | Discover and explain — both | Summarise — aggregate, describe distribution |
What the chart is about
| Subject | What it shows |
|---|---|
| Time | Change across a sequence; temporal order on the x-axis |
| Distribution | How values spread; shape, spread, skew, outliers |
| Association | Relationship between two or more continuous variables |
| Comparison | Relative magnitude across discrete categories |
| Flow | Volume moving between states or nodes |
| Hierarchy | Containment and parent–child structure |
| Network | Connections with no implicit hierarchy |
| Spatial | Values tied to geographic or positional coordinates |
What the data looks like
Structure rules out whole chart families before anything else gets a say.
Dataset type
| Type | Description |
|---|---|
| Tables | Rows and columns; the most common case |
| Tree | Parent–child hierarchy with a single root |
| Network | Nodes and edges; no single root |
| Spatial data | Geographic or positional coordinates |
Value types
The mix of categorical, ordinal, and quantitative variables decides how many chart types remain.
- Cat — pure categories, no inherent order
- Quant — continuous numeric values
- Cat, Quant — the most common combination; unlocks bar, scatter, line families
- Cat, Ord, Quant — adds ranked categories; enables parallel coordinates and heatmaps
- Quant, Quant — two numeric dimensions; scatter and bubble territory
How much data
Scale knocks out candidates on its own, whatever the data type or goal.
| Dimension | Values |
|---|---|
| Variables | 1 / 2 / 3 / 4 or more |
| Categories | under 10 / 10–99 / 100–999 / 1,000+ |
| Data values | under 20 / dozens / hundreds / thousands / effectively infinite |
A bar chart works under 10 categories; beyond that it degrades. A choropleth needs a fixed region set. A scatter plot tolerates thousands of points; a pie chart does not tolerate more than five.
Marks and channels
What a chart is made of. Some families are defined by their mark; within a family, which channel carries which variable is the real design decision. Position is the strongest channel — if a value matters, put it on an axis.
Marks
| Mark | Typical chart families |
|---|---|
| Points | Scatter, bubble, dot plot, strip plot |
| Lines | Line chart, area, parallel coordinates, step chart |
| Areas | Area chart, treemap, choropleth, streamgraph |
| Connection | Network, chord, arc, Sankey |
| Containment | Treemap, circle packing, Euler/Venn |
Channels
| Channel | What it encodes well |
|---|---|
| Position | Quantitative values; highest accuracy for readers |
| Colour | Category membership or magnitude (use sparingly for magnitude) |
| Size | Quantitative values; works for magnitude, not fine differences |
| Shape | Category membership; limited to ~5–6 distinguishable shapes |
In the catalogue
- Bar chart — the one built chart mechanism; it owns the comparison branch
- Elements — the rendering conventions every chart family shares: categorical colour, time and number formatting, symbology
- TODO: the rest…
Resources & references
- Tamara Munzner, Visualization Analysis and Design — the source of the target/aim/query breakdown and the marks-and-channels vocabulary
- IBM Carbon
- WHO Data Design Language
- Apple HIG
- Salesforce Lightning
- Tree of charts by Lars Verspohl
Related patterns
Related
- Needs-based view — the monitoring variant is largely composed of charts; the needs-based framing decides which comparisons each chart must support
- Data view — a chart is one of the representations a collection framing can take — the analytic sibling of table and cards