On Dashboards and Pink Days

I love my wife, and one of the reasons I love her so much is because she and I can entertain ourselves in ways that must look inane and boring from a distance. Last night for example, we sat on the couch watching the late news cast of a local CBS affiliate and making fun of it the whole way. When the weather came on, she didn't let up, calling out their awful graphics, including a 7-day 'outlook' of colored bars without a legend that included several "blue" days, a "yellow" day, a "green" day, and a "pink" day. She had me laughing so hard I was in tears by the time the sports came on.

The point here is that there are lots of bad graphics out there. Bad graphics are those that don't mean anything to your audience. When you're in the audience, it's really easy to spot them. When it's your job to make them - especially if you are intimately familiar with the data behind them - it's not so easy. Good graphics stand alone. So here are a couple of tips on building dashboards:

1. Legends and Titles. Don't just have them, have good ones. They should be clear and explain the meaning of and differences between colors, symbols, etc.
2. Data Points. Don't try and cram too much different data into a single chart or graph. Have one purpose per graph - for example, don't display moving averages and sums in the same area. But do have a high density of data.
   
3. Scale. When placing graphs next to each other, trying to make the layout symmetrical can often skew the interpretation of the graph. For instance, one bar graph with peak values in the hundreds of thousands next to another bar graph with peak values in the tens of thousands will probably confuse your audience, even if there is a y-axis legend explaining this. Similarly, if data is important, showing it to scale next to other data that occurs at different orders of magnitude can hide that data or make it seem irrelevant.

While this sounds sort of obvious, chances are whatever you're using, from gnuplot scripts to a commercial SIM package, probably sabotages you in one or more of these ways out of the box. The best way to fix any of this is to ask the people you're showing it to - in a discreet fashion, lest they pretend to know what's going on to save face - whether or not the graphs you give them are useful and summarize the data they actually care about. Otherwise, someone like my wife will tease you mercilessly. You'll have it coming.