Registered: 1142997573 Posts: 44
Reply with quote #1
Analysts occasionally request circular graphs, sometimes spiraling, but I haven't found much analytic value. I'd be interested in opinions and research pointers. I'm aware of two papers.
Graphical Tests for Power Comparison of Competing Designs [PDF] Heike Hofmann, Lendie Follett, Mahbubul Majumder, and Dianne Cook which compares reader discernment power for Cartesian and circular graphs. The Cartesian versions did much better, though stacked bars may not be the best fit for circular displays. Kaleidomaps: a new technique for the visualization of multivariate time-series data [PDF]
Kim Bale, Paul Chapman, Nick Barraclough, Jon Purdy, Nizamettin Aydin, Paul Dark
They found that the circular charts did better because the concentric circles stand out more than the equivalent parallel horizontal bands in the lower plots. That appears to be true, but I expect the circular version to be worse for almost every other kind of pattern. And the non-perceptual color scheme is not helping. Below is an example from what appears to be a vanity Wikipedia page, but it's a typical request. I suspect most patterns would show up better with a trellis of bar charts, one for each period. The spiral does have one feature going for it, which is the continuity, but is it enough to make up for the perceptual difficulties? (Inner bands get squeezed, turning your head to read (including upside-down), comparing lengths at different angles, ...) Any other experiences? Or other solutions to the issue of continuity across periods?
Registered: 1135986598 Posts: 845
Reply with quote #2
I've reviewed many circular time-series displays over the years, including several variations of spiral displays, and have yet to find one that works better in general than a set of small multiples--one per period, such as year, arranged one above the other. Circular displays with multiple rings or a continuous spiral of time-series values suffer from a number of problems. One of the primary problems is that the period of time in each ring or level of the spiral has less space than the period that proceeded it. By the time you get to the center, a the period of time there has only a fraction of the space that was available to the outermost period of time. The result is that a particular patter in the outer ring would not look the same as one on the inner ring, which complicates comparisons. More fundamentally, it is harder to track a pattern that is arranged in a circle than one that is arranged linearly. Regarding the one downside of the linear arrangement--the fact that the end of one period (e.g., a year) and the beginning of the next are disconnected--this can be overcome by adding a feature that I once discussed with Robert Kosara when we were debating the merits of linear vs circular time-series displays. Imagine that you could dynamically change the breakpoint between graphs. For example, perhaps by moving a slider control you could change the start and end or each graph from January through December of a year to February of one year through January of the next. By changing the breakpoint in this way, you could view the connection between the points in time that were previously disconnected. Another use of a similar control could be to dynamically change the period of time contained in a single graph, such as from year to quarter, quarter to month, and so on. If you moved down into shorter periods of time, you would lose the ability to see as long a period on the screen as a whole, but you would be able to see patterns of change more clearly and perhaps spot patterns that weren't obvious when viewing periods of other lengths. __________________ Stephen Few
Registered: 1142997573 Posts: 44
Reply with quote #3
Thanks, Steve, for the confirmation on lack of utility and for the ideas on continuity.