Graphs That Work

We've talked about how to select the right graph/data visualization before, and we've talked about the importance of storytelling with graphs. Seth Godin now gives us a few rules for creating graphs that work, and they're worth considering.

The good folks at Junk Charts have commented on Godin's rules -- you may enjoy their perspective. (The rules are mostly good, they say.) And Andrew Gelman is the one that brought Godin's rules to their attention -- the comments on Gelman's blog make sense.

Summarizing the rules and discussion and adding a couple thoughts of my own:

Rule 1. Don't let popular spreadsheets be in charge of the way you look

His argument: the default settings in Microsoft Excel and Powerpoint tend to dominate presentations, and that gets boring. The counter-argument: default standards can be good, like reading from left to right or publishing books using a standard font and font size. My thought: If you're telling a story, don't let the blandness of default settings interfere with the message (especially when the defaults in Excel are pretty awful), but don't let creativity run amuck and make it too hard to see the story behind the strange graphics. Be compelling, and use the style necessary to make your point clear.

Rule 2. Tell a story

Godin says:

There are only four reasons I can imagine you would want to show someone a graph (not a chart, or an infogram or a diagram, but a graph of numbers):

1. Things are going great, look!
2. Things are a disaster, help!
3. Nothing much is happening.
4. We need to work together to figure out what the data means.

We've been pounding the data-as-storytelling thought a lot here -- that's why we do community indicators. Godin suggests discarding graphs that fit category 3 as pointless and 4 as a work session and not a presentation. However, "nothing much is happening" is a compelling story, especially after significant community effort/resources are poured into a problem. Failure to bend the trend line is sometimes the most important story we tell -- what we're doing isn't working, and we need to try something else.

Rule 3. Follow some simple rules

This is where he says to use known conventions -- time on the x-axis, increasing left to right. He suggests that the data/y-axis be structured so that upward-moving trend lines are good. We tried that, once upon a time. Graphing the "employment rate" rather than the unemployment rate was just silly, and meant that viewers couldn't see the story while they were trying to figure out why we were being cute. That said, following patterns in how data are displayed can help the real story line pop out -- the less time trying to decipher the message, the better.

Rule 4. Break some other rules

Somehow, your graph has to be memorable -- the story has to stick. That means mixing things up once in a while. Hard to argue with that.

So what rules would you add? What do you use as your guidelines as you present your information?