Visualization Principles
Encoding data so the eye reads truth
Charts encode data into position, length, color and shape. Choosing the right encoding — and resisting pie-chart-itis — determines whether a plot communicates or misleads.
What you will learn
- Match data types to chart types
- Apply the honesty rules (no truncated axes, proportional areas)
- Explain why position dominates color and area
What is this?
Visualization maps data attributes to visual channels: position is read most accurately, then length, angle, area, color, and shape worst. A bar chart uses length, a scatter uses position, a pie uses angle+area. Honest charts keep axes un-truncated, areas proportional and ink devoted to data — the practices behind 'showing the true shape of the data'.
Why do you need it?
- Exploratory plotting finds the patterns statistics confirm — faster iteration.
- A misleading axis (y starting at non-zero) is a deliberate or accidental lie.
- Wrong chart types hide distribution, correlation and outliers.
- Reporting charts drive decisions; their honesty is a product requirement.
Mental model
Reading a map legend
Chart encodings
Maps teach you where tall things are by position and color scale; charts do the same for numbers. A map that drew mountains in proportion to area would look nothing like reality.
How it works
- Position (x, y) is the highest-accuracy channel — use it for the main signal.
- Length (bars, lines) ranks second; area and angle (pies/bubbles) are hardest to read.
- Color works well for categories (hue) and magnitudes (light-to-dark sequence), not absolute reading.
- Honesty rules: start bars at 0, keep aspect ratios unexaggerated, and show a value-per-ink ratio.
- Choose chart by question: trend → line, distribution → histogram, composition → stacked bars, relationship → scatter.
Question → chart selection.
In production
- An experiment report plots CIs as error bars starting at zero, killing the 'looks better' illusion.
- A churn analysis uses a box plot per cohort rather than means — the outlier readers can see the variance.
- A data-org style guide bans pie charts and 3D theatrics, enforcing small multiples instead.
Best practices
- Put the most important variable on position, not color.
- Use bars when comparing named categories; lines only for continuous data.
- Label axes and units; add annotations for the one takeaway.
- Embrace small multiples over one crowded megachart.
Common mistakes
- Truncated y-axes that exaggerate differences.
- Pie charts with 7 slices — angles are unreadable.
- 3D charts and shadows that distort proportional reading.
- Rainbow color scales implying ranking where none exists.
Debugging tips
- A chart showing 'no difference' while tables differ? Check the axis range and smoothing.
- If a plot looks empty, verify you plotted the right subset (mask bug).
- For categorical jitter: a scatter of overlapping dots needs alpha or a violin.
Interview checkpoint
Quick quiz
Human perception reads which channel most accurately?
Bar charts should start their y-axis at zero.