DataslopeChoosing the Right Chart
A decision-oriented guide to picking a chart that answers your question — without misleading the reader.
The right chart depends on the question you're answering. This page is a practical decision flow plus a tour of common charts and their failure modes.
The decision flow
Print it. Reference it. Internalize it.
The chart families
Bar charts
Best for: comparing a numeric metric across distinct categories.
Do:
- Sort bars by value, not alphabetically.
- Start the y-axis at 0.
- Label units.
Don't:
- Use 3D bars.
- Compare too many categories — top-N + "Other" is often better.
Histograms
Best for: showing the distribution of a single numeric column.
Do:
- Try several bin counts (10, 30, 100) — different bin choices reveal different patterns.
- Annotate the median or mean if it adds context.
Don't:
- Confuse histograms with bar charts. Histogram x-axis is continuous; bar chart x-axis is categorical.
Box plots
Best for: comparing distributions across several groups.
Do:
- Show all points (
points="all"in Plotly) when sample sizes are small — boxes alone can be misleading.
Don't:
- Use box plots for fewer than ~10 points per group.
Scatter plots
Best for: relationships between two numeric variables.
Do:
- Use transparency for dense plots.
- Add a trend line if you're showing a relationship.
Don't:
- Read causation from a scatter — only correlation.
Line charts
Best for: change over time, with continuous x-axis.
Do:
- Sort by x.
- Use color to distinguish multiple series.
- Always include a 0 baseline unless the value can't reach 0.
Don't:
- Use lines for categorical x-axes (use bars).
Pie and donut charts
Best for: almost nothing. Bar charts almost always do the job better.
If you must:
- Use only 2–3 slices.
- Show percentages.
Bad choices and their consequences
A practical anti-checklist
Before you ship a chart, ask:
- Can a reader misread the axis range and reach the wrong conclusion?
- Are the bars / lines / points doing the work, or is the decoration?
- Would a less-fancy chart be more honest?
- Is the title a statement of what the reader should take away, not just a description of what the axes are?
A great title — "Sales grew 18% in Q4 despite flat units" — does more for understanding than a beautiful chart with the title "Sales over time".
Communicating uncertainty
When you show an average, also show how much it varies:
- Error bars on bar charts
- Confidence bands on line charts
- All points beside box plots
- Sample size in the caption
Hiding uncertainty makes patterns look stronger than they are.
When in doubt — show the data
If you can't decide between two charts, pick the one that shows more of the underlying data. Strip plots, jittered scatter, overlaid points on top of boxes — when in doubt, show the points.
Check your understanding
You want to show how monthly active users changed over the last 12 months. Best chart?
Pie chart
Stacked bar
Line chart — time is continuous and a line emphasizes the trend
Box plot
Top 5 product categories by revenue. Best chart?
Pie chart
3D bar chart
Horizontal bar chart sorted by value
Histogram
A scatter plot of 100,000 points looks like a black blob. Best fix?
Increase the figure size
Plot fewer columns
Lower the opacity (or use a 2D density / hexbin plot) so the structure becomes visible
Change the color palette
A reporter shows two bar charts. The first starts the y-axis at 80; the second starts at 0. Both use the same data. Which is more honest?
The y=80 chart — it shows detail
Both are equally honest
The y=0 chart, in most cases — starting at zero preserves correct visual proportions, especially for additive quantities. A truncated axis can dramatically exaggerate small changes.
The y=80 chart — it is more zoomed in