Choosing the Right Chart
What are you comparing? Then pick the chart
Match the chart to the comparison: one dimension, change over time, part-to-whole, distribution, correlation, ranking, or flow. The right chart answers the right question 90% of the time.
What you will learn
- Map the seven classic comparisons to their chart archetypes
- Avoid chart-vs-message mismatches
- Annotate the takeaway so the chart speaks
What is this?
Chart selection is question-driven: change over time → line; magnitude compare → bar; part-to-whole → stacked bar or donut (sparingly); distribution → histogram/boxplot; correlation → scatter; ranking → ordered bar; flow → sankey/beeswarm. When the comparison type and the chart type disagree, readers reverse-engineer the wrong message.
Why do you need it?
- A mismatched chart smudges the exact comparison the user needs.
- A small catalog of archetypes covers most production analytics.
- Question-first selection is the defensible answer in chart reviews.
- Using the right shape makes the takeaway pop without annotation crutches.
Mental model
A toolbelt with one tool per job
Chart types
You wouldn't use a wrench to hammer — but people do it with charts daily (scatter for trends, pie for rankings). Each chart is shaped exactly for one comparison type; the belt is small and you pick by the task, not by style.
How it works
- Time-series → line/area (x = time), scatter for point-in-time correlation.
- Category magnitude → vertical bars; many categories → horizontal bars.
- Part-to-whole splits → stacked bars or a (documented) donut for 2-5 shares.
- Distribution → histogram/binned bars, boxplots for summary stats.
- Ranking → ordered horizontal bars; flow → sankey only when real flow exists.
Comparison → chart type.
Examples
The 90% chart chooser
IF the question is: "How did it move over time?" -> Line chart "Which is bigger?" -> Vertical bar "Which ranks highest?" -> Horizontal ordered bar "Parts of the whole?" -> Stacked bar (or donut, 2-5 slices) "How distributed?" -> Histogram / boxplot "Do two things correlate?" -> Scatter + trend line "What flows where?" -> Sankey (only real flows)A small decision tree covers nearly every analytics chart request. Deviations need a written reason.
In production
- An exec trend page uses lines for monthly revenue and bars for plan-vs-actual deltas.
- A pricing analysis uses a scatter of price vs volume to show the demand curve.
- Support aging uses a histogram of ticket age bins instead of a line, so the distribution is visible.
Best practices
- Answer the comparison type first; choose from the archetype catalog.
- One chart, one message; resist combo-chart spaghetti.
- Sorted categories when ranking; chronological when time.
- Annotate the conclusion — charts should want to say something.
Common mistakes
- A scatter for time series (reorders the story).
- A line for categories (implies continuity that isn't there).
- Pies with 12 slices; stacked 100% bars hiding absolute drops.
- Dual y-axes on a line chart that visually correlate unrelated scales.
Debugging tips
- Charts 'feel off' to stakeholders → re-check the comparison-to-type match.
- Line suggests a trend the data doesn't have → category data got a line; switch to bars.
- Review caught a scatter without correlation → add the trend line/CI or drop it.
Interview checkpoint
Hands-on lab
- Map question types to chart archetypes from facts
- Compute the trend stats that justify line vs bar
- Build the histogram supports for distribution claims
Pick the Chart from the Data Story
40 min · intermediate
Quick quiz
Which chart emphasizes continuous change over time?
A pie with ten slices is a good way to compare category shares.