Practical Guidelines for Visualization
Today the analytics company Geckoboard posted the following visualization on r/datascience. It was the trending post on the subreddit for several hours. The principles it outlines are useful best practices for charting data. I’ve outlined the key ideas it describes above the image.
- Present the Facts
- Bar charts are good for comparisons, but should always begin at zero.
- Line charts are good for showing trends, but the aspect ratio can distort trends dramatically.
- Avoid Pie charts. Visually comparing angles is difficult.
- Visually comparing one dimension is easier than two or three. Use lengths, not areas or volumes.
- Less is More
- Use the minimal number of decimal points possible. Three significant figures is almost always adequate.
- Eliminate colored backgrounds, shading, and dark grid lines. Draw attention to the data.
- 3D plots distort data. Do not use them.
- Use color if it conveys information, not for decoration.
- Keep it Simple
- Every chart and axis need a title.
- Do not plot two datasets on the same chart using a secondary axis, unless you are exploring the relationship between the sets.
- Align numbers to the right to aid visual comparison.
- If there are only one or two values, show numbers, not charts.
For more in-depth discussion of the principles described above, see Geckoboard's "original post".