Data visualization: approaches to creating effective figures

PSYC 81.09: Storytelling with Data

Jeremy R. Manning
Dartmouth College
Spring 2026

Why visualize data?

Our visual systems rapidly process massive amounts of information and are adept at pattern recognition. We can leverage this to convey patterns in data — but only if we figure out how to turn data into the right pictures.

A good visualization can reveal what no table of numbers ever could. A bad visualization can obscure or mislead.

Which representation is clearest?

Anscombe's quartet

These four datasets have identical summary statistics (same mean, variance, correlation, and regression line) — but look completely different when plotted!

The most important question

What message do you want your audience to take away?

Every visualization decision — color, layout, chart type, annotation — should serve that message. Start with the message, then choose the visualization.

Many people pick a chart type first, then try to fit their data into it. Instead, decide what you want to say, then find the visualization that says it most clearly.

Design principles

  • Data-to-ink ratio (Tufte): maximize the share of ink devoted to data; minimize non-data ink (gridlines, borders, decorations)
  • Clarity over cleverness: if the audience has to work to understand your figure, simplify it
  • Consistent encoding: same color = same meaning throughout your presentation
  • Accessible design: use colorblind-friendly palettes; don't rely on color alone
  • Label everything: axes, units, legends — if someone can't understand the figure without your narration, add more labels
  • Be willing to break all of the rules! Sometimes the most effective figure violates a guideline

Grammar of graphics: figures are built from layers

Based on Wickham's A layered grammar of graphics and Wilkinson's The grammar of graphics

Choosing the right visualization

If you want to... Consider...
Compare categories Bar chart, grouped bar, stacked bar
Show distributions Histogram, density plot, violin, box plot, strip plot
Reveal relationships Scatter plot, bubble chart, heatmap, pair plot
Track change over time Line plot, area chart, sparkline
Show composition Pie chart, stacked area, treemap
Display spatial data Choropleth map, bubble map
Show connections Network graph, chord diagram, Sankey diagram
Add a dimension Animation, 3D projection, vary size/color/markers/style

Bar chart: comparing values across categories

Grouped bar chart: comparing subcategories

Histogram: showing the shape of a distribution

Box plot and violin plot: revealing distribution shape

Scatter plot: revealing relationships between variables

Heatmap: showing magnitude in a matrix

Line plot: tracking change over time

Area chart: emphasizing cumulative magnitude

Pie chart: parts of a whole (use sparingly!)

Humans are bad at comparing angles and areas. Bar charts are almost always more readable. Popular in some contexts (e.g., business presentations) but generally not recommended for scientific communication.

Doughnut [chart]: a slightly better alternative to pie

Choropleth map: coloring regions by value

Color intensity encodes magnitude. Great for showing regional variation — but large, sparsely populated areas can dominate visually.

Animation: life expectancy vs. GDP over time

This famous visualization — popularized by Hans Rosling — shows how life expectancy and GDP per capita have changed across countries from 1952 to 2007. Bubble size denotes population; color denotes continent, time denotes the year.

Network graph: showing connections between entities

Nightingale's rose diagram (1858)

Nightingale's chart convinced the British army to prioritize sanitation — saving thousands of lives.

Snow's cholera map (1854)

Deaths clustered around the Broad Street pump — disproving the "miasma" theory and proving cholera was waterborne.

Minard's map of Napoleon's Russian campaign (1869)

Six variables in one image: army size, location, direction, temperature, latitude, and longitude. — "The best statistical graphic ever drawn" according to Edward Tufte.

The periodic table (1869): organizing all known matter

Mendeleev's arrangement revealed patterns in chemical properties — and predicted elements that hadn't been discovered yet.

Radar plots: visualize many dimensions at once

Each point along the circumference of a radar plot denotes a different variable (dimension). Distance from the center encodes magnitude. The shape of the resulting polygon reveals patterns across dimensions.

Source: Owen et al., (2024)

Questions? Want to chat more?

📧 Email me
💬 Join our Slack
💁 Come to office hours
  • Thursday (X-hour): Continuing to discuss data visualization OR open office hours— vote!
  • Friday: Workshop data story ideas + Assignment 2 release