Choosing the Right Chart

This chapter assumes the reader, decision, metric and evidence have already been defined. If they have not, return to Part I — Foundations of Effective Chart Design. Chart selection cannot compensate for an unclear purpose or unsound evidence.

The remaining question is practical: which visual form will make the required comparison easiest to see?

The selection path

The interactive taxonomy above follows three levels. Start with analytical intent, examine the shape of the data, then choose the visual form whose encoding best supports the reading task.

Level 1: What are you trying to understand?

Translate the business question into an analytical intent. None of these intents names a chart. Each describes the kind of understanding needed before making a decision.

Uncertainty cuts across all of them. "How sure are we?" may not be the primary question, but it qualifies forecasts, estimates, comparisons and model output. Show it whenever its omission would change the judgement.

If the question does not fit one of these intents, return to the purpose. The decision may need to be clearer, the metric better defined or the information need made more specific.

Level 2: What does the evidence look like?

Intent identifies the reading task. The structure of the evidence determines which version of that task is actually present.

Comparison. How many categories are there? Is the point a ranking, a difference, or performance against a target? Are values positive and negative? Is there one measure or several?

Trend. Is time continuous or divided into distinct periods? Is there one series or many? Does seasonality matter? Is part of the series forecast? Are missing observations genuinely missing?

Distribution. Is there one group or several? How many observations are available? Do readers need the overall shape, summary statistics, outliers or every point?

Relationship. Are there two variables or more? How many observations are there? Is the question about correlation, clusters, drivers or exceptions? Do the measures share a compatible grain and population?

Composition. Is this a snapshot or change over time? Are there a few parts or many? Is there one level of grouping or a hierarchy? Does the reader need precise comparison or a broad sense of share?

Geography. Do values attach to regions, points or routes? Is the story location, density or movement? Should values be normalised for population or another denominator?

Flow. Does quantity move between states, branch between nodes or pass through ordered stages? Is the structure important, the volume, or both?

Hierarchy. How deep is the structure? Do leaf values matter? Must readers compare across branches, navigate the structure, or understand reporting relationships?

These answers remove unsuitable chart families before visual preference enters the discussion. They also expose when the evidence is not ready to chart: incompatible measures, missing denominators, too few observations or an aggregation that hides the variation the decision depends on.

Level 3: Which form makes the answer easiest to see?

Only now should a chart type enter the conversation. The strongest first choice is the form whose visual encoding best supports the intended comparison with the available evidence.

Compare values

• Reach for first: bar and column, dot plot, lollipop. They encode values through aligned position or length, making ranking and differences easy to judge.
• Also useful: bullet for performance against a target; slope for two points in time; table heatmap for a dense matrix; small multiples for repeated comparisons on common scales.

Show change

• Reach for first: line for continuous change; columns for a small number of discrete periods.
• Also useful: area when accumulated magnitude matters; sparkline for compact context; waterfall for an additive bridge; bump for rank over time; calendar heatmap for daily rhythm; trend ribbon for change with uncertainty.

Understand a distribution

• Reach for first: histogram or density for shape; box plot for compact comparison across groups.
• Also useful: violin for shape plus summary; strip / jitter and beeswarm when individual observations matter; ridgeline when distributions change across an ordered dimension.

Explore relationships

• Reach for first: scatter, because position on two quantitative scales reveals association, clusters and outliers; hexbin when point volume causes overplotting.
• Also useful: bubble when a third variable only needs approximate reading; parallel coordinates for multivariate profiles; table heatmap for a correlation matrix.

Understand composition

• Reach for first: stacked bar when parts need precise comparison across groups; waffle for a simple, countable snapshot; treemap for many hierarchical parts.
• Also useful: pie or donut for two or three parts when a rough sense of share is sufficient; marimekko for two dimensions of share; stacked / stream area for composition over time.

Understand geography

• Reach for first: choropleth for normalised values attached to regions; symbol maps for values attached to locations.
• Also useful: value-by-alpha when confidence or population should temper visual prominence; hexbin surfaces for point density.

Explain processes and flow

• Reach for first: sankey for quantities moving between states; funnel or a bar sequence for ordered attrition.
• Also useful: chord for flows between many pairs; network when the structure of connections is the question.

Explain hierarchy

• Reach for first: a tree when structure and paths matter; treemap when the hierarchy also carries magnitude.
• Also useful: sunburst for a compact view of hierarchical share; a table or indented list when readers need exact lookup and navigation.

Reveal uncertainty

• Reach for first: intervals attached directly to the estimate being judged; trend ribbon for uncertainty around a time series.
• Also useful: density and violin when the full distribution matters; fan charts for widening forecast horizons.

A chart selection reference

The matrix compresses the selection path into one reference. Enter from the first column when the analytical intent is clear. Use the second column to match the relevant data shape, then treat the final columns as candidates to evaluate rather than instructions to follow blindly.

Continue into the Chart Index

Every Chart Index article follows the same structure: what the chart is, when to use it, when not to use it, design principles, anatomy, related types and further reading. Use the selection path to identify a small set of candidates, then use their entries to understand the trade-offs.

Comparison — bar and column · grouped bar · stacked bar · lollipop · dot plot · bullet · slope · marimekko · table heatmap · small multiples

Time series — line · area · stacked and stream area · sparkline · candlestick · horizon · calendar heatmap · bump · waterfall · trend ribbon

Distribution — histogram · density · box plot · violin · strip / jitter · beeswarm · ridgeline

Relationship — scatter · bubble · hexbin · parallel coordinates · voronoi

Composition and hierarchy — stacked bar · pie and donut · treemap · sunburst · waffle

Flow — sankey · chord · network · funnel

Geography — choropleth · value-by-alpha · hexbin map

Chart choice is complete only when you can explain why a form suits the question, the evidence and the way the reader must interpret it.