DataslopeEthics and Accessibility
Charts can mislead and exclude. Both are avoidable.
A chart is a public act. It shapes what people believe and what they decide to do. With that power comes a small set of responsibilities — to be honest, and to be inclusive. This page is about those two responsibilities.
Why visualization ethics is its own topic
A misleading chart can spread further and stick longer than a misleading paragraph. Charts feel evidential — "the data says so" — even when the chart is showing only a slice, or using an encoding designed to inflate a difference.
Most misleading charts are not the work of malicious actors. They are the result of honest people not noticing what their chart is doing. The fix is awareness.
The classic ways charts mislead
1. Truncated y-axis on bar charts
Bar length encodes magnitude relative to zero. Cutting the axis turns a 5% difference into a 10× visual gap.
The data is identical. The second chart screams "B is twice A" — which it manifestly isn't.
2. Cherry-picked time ranges
Show only the months that support your story; omit the months that don't. The chart looks technically correct; the conclusion is wrong.
The fix: always show enough history to put a trend in context. If a series has been wandering up and down for 10 years, a 3-month "uptrend" is noise.
3. Dual y-axes
Two unrelated series on the same chart with two y-axes can be scaled to look correlated — even when they aren't. Most visualization style guides discourage dual axes entirely.
4. Misleading color choices
- Using red-vs-green for "good vs bad" excludes color-blind viewers (and culturally varies).
- Using rainbow color scales creates false structure.
- Inverting an intuitive color direction (red for "good"?) makes the chart read backwards.
5. Inappropriate aggregation
Showing an average can hide a bimodal distribution. Showing a median can hide the magnitude of outliers. Always ask whether your chosen summary is honest about the variation.
6. Map projections that distort area
We covered this in the geographic chapter. Mercator's exaggeration of polar regions has shaped centuries of cultural geography. Choose your projection deliberately.
The Tufte "data-ink" principle
Edward Tufte's famous principle:
Above all else, show the data.
He coined the data-ink ratio: the proportion of a chart's ink that actually encodes data, versus the ink used for decoration (heavy gridlines, 3-D shading, borders, logos). Maximize the ratio. Strip away chart junk.
The simple_white template this course uses is essentially a
Tufte-style design choice: minimal gridlines, no background fill,
no decorative borders. The data carries the chart.
Accessibility: who can read your chart?
A great chart can fail not because it's wrong but because parts of the audience cannot perceive it.
Color blindness
About 8% of men have some form of color vision deficiency. Red-green confusion is the most common. Mitigations:
- Use a colorblind-safe palette (
Viridis,Cividis,Plasma,Plotly's default qualitative). - Double-encode important categories with shape and/or label, not just color.
- For sequential scales, prefer luminance-varying palettes that survive grayscale.
Low contrast
Light gray text on a white background may look elegant but is hard to read for low-vision users. Default to dark text on light backgrounds (or vice versa) with high contrast.
Screen readers
A chart embedded in a webpage should have a meaningful alt text — a 1-3 sentence summary of the chart's takeaway, not just "bar chart." Most chart-export workflows make this easy if you remember to do it.
Font size
Chart text that's readable on your 27" monitor may be illegible on a phone or in a slide. Default to ~12-14pt for axis labels and ~16-18pt for titles. Plotly templates already do this reasonably.
Interactivity for keyboard users
A purely click-and-drag chart can be unreachable for keyboard-only users. For dashboards and public-facing charts, ensure the critical information is also in static labels, not only in hover tooltips.
A simple ethics checklist
Before publishing any chart, ask:
- Does my y-axis start at zero (for bars)? If not, do I have a good reason and a clear label?
- Is my chosen time range long enough to put any trend in context?
- Could my color choices fail for colorblind viewers?
- Am I summarizing in a way that hides real variation?
- Does the title fairly summarize the takeaway?
- Have I added a data source and any necessary caveats?
- Would a knowledgeable reader, given the underlying data, agree with the conclusion my chart implies?
Check your understanding
Why is truncating the y-axis on a bar chart usually misleading?
Plotly disallows it.
It makes the chart harder to render.
Bar length encodes magnitude relative to zero — truncating the axis exaggerates small differences and produces a visual cue that does not match the underlying data.
It uses more colors.
Which of these does NOT improve a chart's accessibility?
Using a colorblind-safe palette.
Adding meaningful alt text.
Using high-contrast text.
Adding a 3-D effect to the bars.
What does the data-ink ratio, from Edward Tufte, refer to?
The cost of printer ink to publish a chart.
The amount of color in a chart.
The proportion of a chart's visual elements that actually encode data, versus those used for decoration (heavy gridlines, 3-D shading, borders).
The amount of black ink in a chart.
A trend chart shows 3 months of "uptrend" but the underlying series wanders up and down for 10 years. What is the most ethical way to present this?
Show only the 3 months — it's the most recent.
Show only the 10-year average.
Show enough history (e.g., the full 10 years) so the reader can judge whether the 3-month trend is meaningful or just noise.
Show the chart without axis labels.
The simplest "is my chart honest?" test is:
"Does it look pretty?"
"Did Plotly render it without errors?"
"If a knowledgeable reader saw the underlying data, would they reach a different conclusion than my chart implies?"
"Is the chart interactive?"