|
What are the biggest pitfalls in data visualization?
|
I see so many misleading charts and graphs these days, especially in business presentations and news articles. Some are intentionally deceptive, but I think most are just made by people who don't understand data visualization pitfalls.
Things like truncated y-axes, 3D pie charts, and inappropriate chart types seem to be everywhere. What are the worst offenders you've seen?
And what are some data visualization pitfalls that even experienced analysts sometimes fall into? I'm particularly interested in how visualization choices can create statistical illusions in data.
Truncated y-axes are my biggest pet peeve. News outlets love to use these to make small changes look huge. A stock might go from 100 to 101, but if the y-axis starts at 99, it looks like a massive spike.
3D pie charts are terrible too. The perspective distortion makes it impossible to accurately compare slices. And don't get me started on pie charts with too many slices - anything more than 5 or 6 becomes unreadable.
Another bad one: using area to represent quantity in bubble charts, but making the diameter proportional to the value instead of the area. Since area = πr², doubling the diameter quadruples the area, creating misleading visual comparisons.
Choosing the wrong chart type is a common data visualization pitfall. Like using a line chart for categorical data, or a bar chart for time series data when a line would be better.
Also, overcomplicating charts with too much decoration - 3D effects, gradients, shadows, etc. These don't add information and often make the data harder to read.
Color choice matters too. Using non-sequential color schemes for sequential data, or using red-green color combinations that colorblind people can't distinguish. And using too many colors - humans can only distinguish so many shades accurately.
One subtle pitfall is what Edward Tufte calls chartjunk" - all the unnecessary elements that don't convey data. Gridlines that are too prominent, backgrounds that distract, legends that could be labeled directly on the chart.
Also, not starting bar charts at zero. This can create statistical illusions in data by exaggerating differences. If one bar is 50 and another is 100, starting at zero shows the second is twice as tall. Starting at 40 makes it look 6 times taller!
The best visualizations are simple, clear, and let the data speak for itself. Remove everything that doesn't help understanding. Use the simplest chart type that shows what you need to show.
|
