Charts That Tell the Truth
Generated charts default to decoration. How to choose the right form, what quietly misleads, and how to write a title that says the finding.
Ask for a chart and you will get one — usually with a legend nobody needs, a default palette, and a title that names the columns rather than the finding. It will be technically correct and communicate almost nothing.
The fix is to say what the chart is for before asking for it.
Match the form to the comparison
- Change over time → line chart. Not bars, unless the periods are few and discrete
- Comparing categories → horizontal bar chart, sorted by value. Sorting is what makes it readable and is almost never the default
- Parts of a whole → stacked bar, or just a table. Pie charts fail above three slices and nobody can compare angles
- Relationship between two numbers → scatter plot
- Distribution → histogram or box plot. An average alone hides everything that matters about a distribution
- Small precise numbers → a table. Not everything needs to be a chart, and a table of six numbers beats a chart of six numbers
What misleads, usually accidentally
- A truncated y-axis on a bar chart. Bars encode length, so starting at 40 rather than 0 turns a 3% difference into a visual doubling. Line charts may be truncated; bars should not be
- Dual axes. Two lines on two different scales can be made to appear correlated or not by choosing the scales. Use two stacked charts instead
- Cherry-picked windows. Starting the chart at an unusually low month makes any recovery look like growth
- Absolute counts where rates belong. 'More errors in December' means nothing without December's volume
- 3D effects and unnecessary colour, which distort perception of size for no benefit
Asking for a chart that communicates
Make one chart from this data.
Point to communicate: repeat purchase rate for the direct
channel has risen while resellers stayed flat.
Audience: our team, in a meeting, on a shared screen.
Requirements:
- Line chart, monthly, two lines
- Y axis is a percentage starting at 0
- Title states the finding, not the columns
- Label the two lines directly on the chart, no legend
- Annotate March, when the pricing change happened
- Muted colours, one highlighted line, everything else grey
- Readable at a distance: minimum 12pt labels
Output the code so I can adjust it.Grey is the most underused technique
When one series matters and the others are context, colour only the one that matters and make the rest grey. This does more for comprehension than any other single choice, and generated charts never do it unless asked.
Say what the chart does not show
A chart is an argument. If the data covers only paying customers, or excludes one region, or stops before a known event, that belongs in a caption under it.
Analysts who state their limitations get trusted more, not less — and the alternative is someone else finding the limitation during your presentation.
What to take from this chapter
- Choose the chart form from the comparison you are making, and sort category bars by value
- Write titles that state the finding, not the column names
- Never truncate a bar chart axis; avoid dual axes entirely
- Grey out context and colour only the series that matters
- Caption what the data excludes — stated limitations build trust
Try it
Take your most recent chart and rewrite its title as a full sentence stating the finding. If you cannot, the chart is showing data rather than making a point, and it needs rethinking rather than restyling.