?

Log in

No account? Create an account
Previous Entry Share Next Entry
Visual toy: Horizon Charts
device
jducoeur
This is a Data Visualization burble, so most of you probably want to just skip along here. But if you find the Tufte books interesting, you might want to check this out.

Tonytip (my co-Architect at work) just pointed out a couple of fascinating articles about Horizon Charts, a semi-recent (2008) visualization concept that I hadn't come across before. This article (PDF) is an excellent introduction to the concept, and explains how the author went from being a skeptic to gradually grokking the logic behind them. And this article (which leads to another PDF) is a delight -- a proper scientific study that examines how well this visualization works. (Answer: better than the study's authors originally expected.)

The short version is that a horizon chart is a visualization technique for comparing a large number of time series next to each other. You would use it in circumstances where the obvious thing to do is stack a lot of line charts -- the first article does a fine job of working through each step of how you go from simple line charts to this new model, which uses color and contour in some brilliant ways to provide a very glanceable notion of the large-scale trends, and draws the eye quickly to interesting outliers.

Neat stuff, and it's always great to have a new tool for my kit...

  • 1
I though about this, and I like the concept.

I am not sure I have EVER had the need for 50 linear time line graphs however. I am not sure that I can think of an application where it would really be informative. Looking at 50 stocks for instance just doesn't seem like a reasonable use of anyone's time. What trends are you trying to tease out that wouldn't benefit by pre-processing?

Can't go into details, but I can actually think of several ways in my own products where this could potentially be useful, at least as a first cut. I think you'd need to accompany it with good drill-down tools, but there are a variety of dashboards, for example, where this might be a good glanceable way to start out, to spot the anomalies quickly.

(Granted, I suspect that those wouldn't usually be comparing 50 lines. But 20 is entirely plausible.)

Also, it provides an interesting alternative for some data-visualization problems we've had -- in particular, trying to show charts that contain a few huge spikes. Naively displaying those tend to suck: the spike washes out all detail in the rest of the data. Log graphs are sometimes a conceptually good way to address that, but most users don't understand them. We've struggled with this from time to time, and I don't think this model is right for every case, but sometimes it might be the right tool...

  • 1