Learning from experts

The best books I’ve read on data visualization are:

• “The Visual Display of Quantitative Information” by Tufte, and
• “Storytelling with Data” by Cole Nussbaumer Knaflic

Between those two books, you get a balance: the ideal (from Tufte) and the PowerPoint world (from Knaflic).

Tufte

Edward R. Tufte is professor emeritus of statistics, political science, and computer science at Yale.

And he’s a man upset by pie charts.

the only worse design than a pie chart is several of them.

– Edward R. Tufte

Tufte’s rules:

• 🦑Maximize the data-ink ratio – Most of your chart should be data. Erase the parts that convey nothing.
• 🍫Avoid chart junk – Avoid needless colors, shading, tickmarks, and gridlines: remove distractions from data.

Knaflic

Cole Nussbaumer Knaflic was the former manager of Google’s People Analytics team. She knows her way around a PowerPoint¹.

Knaflic’s rules:

• 📈Choose an appropriate visual display – Know your context, then pick your chart
• 🏰Tell a story – Charts ought to communicate something; you should know what that something is.

Tufte’s principles: a case study

Take a look at this chart from the Wikimedia Foundation’s 2023–2024 budget projections:

This is a default Google Sheets chart for this data.

What I’m able to glean from this chart:

• “Building analytics & ML Services”: the biggest, > 25%
• “Features and functionality” and “Supporting volunteers”: smaller, ~20% each
• “Fundraising,” “Protecting access,” and “General & Admin”: smaller still, ~10% each
• The colors seem meaningless

Now consider what we glean from the table used to generate it:

This table gives us more information than the chart in a similar amount of space.

From the table, we learn:

• Exact percentages—no need to guess
• Exact dollar amounts
• The total budget as a dollar figure

And we get rid of the meaningless colors.

OK—confession time: there were percentages on the original pie chart. I edited them out.

But I did that to prove a point. The pie chart is doing less work than the numbers. The slices of the pie add almost nothing to the numbers.

Tufte would prefer the table above to the pie chart above because the table:

• Shows the data
• Maximizes the data-ink ratio
• Avoids chartjunk (like the meaningless colors)

And in this case, I agree: I like the table.

But choosing data visualizations is difficult.

How to choose a chart: visual perception accuracy ranked

If you intend to communicate, then people must be able to read your chart.

In 1985, Cleveland & McGill conducted what is still the most cited experiment on what charts are easiest to read. Their purpose was simple: rank standard charts by the number of errors people make while reading them.

Here’s their ranking of charts—from easiest to hardest to read:

Icon	Name	Example
	Position along a common scale	scatterplots, bar charts, sparklines
	Positions along nonaligned scales	stacked bar charts
	Length, direction, angle	pie charts, donut charts
	Area	bubble charts, treemaps
	Volume, curvature	3d charts
	Shading, color saturation	Heatmaps

But I love heatmaps! You can use heatmaps. Just understand that only a sophisticated audience can interpret heatmaps correctly.

Know your audience, then use something like the data viz catalog to choose your chart.

Tell a story

Charts are communication.

At their worst, as Tufte said, they’re little more than “devices for showing the obvious to the ignorant.”

But at their best, they’re a powerful way to intuitively communicate a lot of data in a small space.

Why care about air quality?

Acute exposure to air pollution makes you acutely dumber.

This was the conclusion of MIT researchers back in 2022 when they looked at the effect of air quality on chess.

They combed through 30,000 chess moves, evaluating them with the Stockfish chess engine, comparing 121 players against themselves under different air quality conditions (which they monitored with foobot sensors).

The researchers concluded that an increase as small as 10 µg/m³ PM2.5 causes a 2.1% increased likelihood of player error.

Measuring air quality

Reference particle mass counters cost thousands of dollars. And even so-called low-cost air sensors like the ubiquitous PurpleAir will set you back $200.

But the same sensors used inside the PurpleAir, the PlanTower PMS5003, can be found for as little as $15 on AliExpress.

The Plantower sensor, however, is far from a reference device. But studies suggest it’s directionally correct. And, with after-the-factory calibration, it can match readings from more expensive reference meters¹.

The Plantower PMS5003 features in AirGradient’s DIY printed circuit board (PCB), which combines air quality, temperature, and CO2 sensors with a cheap ESP8266 for internet. I ordered ten of these PCBs from PCBWay back in 2021 for about $30.

I’ve since modified AirGradient’s example code to support pushing data to Home Assistant via MQTT. From there, sensor data gets sucked up by Prometheus, so I can monitor it via Grafana.

This system gives me a full view of my indoor air quality. And it’s a needlessly complicated way of reminding me to change out my air filter 😬.