4
$\begingroup$

The Chord Diagram can be used to show directed relationships among a group of entities. This is a circular chart, much like a Pie Chart. What would be its linear equivalent, in the same way that a single-column Stacked Bar Chart is the linear equivalent to the Pie Chart?

enter image description here

$\endgroup$
  • 1
    $\begingroup$ Your question suggests there will be exactly one; there may well be several. One possibility is to have all entities arrayed down a pair of vertical lines (each one present on both lines). Then the directed lines connecting them can go from the points on the left vertical to the right one. $\endgroup$ – Glen_b -Reinstate Monica Jul 27 '15 at 0:10
  • $\begingroup$ Gleb_b I like your idea. Is there a name for this? And I'm looking for the most "canonical" solution, but I agree that there are many ways to depict the same information in a linear fashion. $\endgroup$ – Ismael Ghalimi Jul 27 '15 at 0:13
  • 1
    $\begingroup$ There is a name, yes (it probably has several names), and when/if I think of it, I'll mention it (and at that point it would be worth posting an actual answer). Imagine something like this but with the points labelled (and no need for an axis). No doubt someone else will jump in to supply what my feeble grey matter has forgotten. It's commonly used for paired data but there's nothing stopping it being used on relationships where many lines come from or go to each point. $\endgroup$ – Glen_b -Reinstate Monica Jul 27 '15 at 0:14
  • 1
    $\begingroup$ In McNeil, D. (1992), "On Graphing Paired Data," The American Statistician, 46:2, pp. 307-311, Don calls it a "tilted line segment plot". While the article specifically discusses paired data the name would serve equally well for the more general case. See also Nick Cox's paper here: "stata-journal.com/sjpdf.html?articlenum=gr0005" (which doesn't name it but describes it with the same terms). There are problems if there are many lines, as both authors mention. $\endgroup$ – Glen_b -Reinstate Monica Jul 27 '15 at 0:27
2
$\begingroup$

There may well be several displays that match your criteria.

One possibility is to have all entities arrayed down a pair of vertical lines (each one present on both lines). Then the directed lines connecting them can go from the points on the left vertical to the right one, something like this:

enter image description here

If it's obvious that the source is on the left, and the target on the right, the arrows are not really needed; line segments would suffice.

(It could be done horizontally of course but it would not change the idea.)

In McNeil, D. (1992), "On Graphing Paired Data," The American Statistician, 46:2, pp. 307-311, Don calls it a "tilted line segment plot". While the article specifically discusses paired data the name would serve equally well for the more general case. See also Nick Cox's paper ("Speaking Stata: Graphing agreement and disagreement", The Stata Journal (2004), 4: 3, pp. 329–349) here (which doesn't name it but describes it with the same terms). There are problems if there are many lines, as both authors mention.

Versions of this with width of the lines to represent volume have been used.

Compare the chord diagram here with the plot similar to this (but in this case arranged horizontally) here

$\endgroup$
  • 1
    $\begingroup$ More at Cox, N. J. 2009. Speaking Stata: Paired, parallel, or profile plots for changes, correlations, and other comparisons. Stata Journal 9 (4): 621-639. Key words include profile plot, parallel coordinates plot, parallel line plot, pair-link diagram, bumps charts, barometer charts stata-journal.com/sjpdf.html?articlenum=gr0041 $\endgroup$ – Nick Cox Mar 28 '16 at 12:32
4
$\begingroup$

Being a pedant, I would say the example slopegraph and Sankey diagrams are not equivalent to the chord diagram. A chord diagram only has one dimension, in terms of the degree around the circle in a polar coordinate system. (You could argue the width is another dimension, but in Wilkinson's Grammar of Graphics this is not a dimension in a coordinate system, but a mapped aesthetic. I know pedantic.)

So if you change the polar coordinate system to a linear one, just simply imagine unwrapping the circle to a straight line. I've seen these called Arc Diagrams. Xiao and Chun (2009) in Visualizing migration flows using Kriskograms give multiple examples. Here they show how arcing flows above and below the line can help disambiguate some of the flow patterns.

enter image description here

In practice such graphs tend to be very busy, and so are difficult to disentangle all the lines. I tend to like the linear diagrams better than the circular ones, but any complicated data tends to be very hard to see patterns.

This example is fairly simple to imagine unwrapping to a linear coordinate system, but sometimes it is easier to think about these types of graphs in terms of a network of nodes and edges as oppossed to a rectilinear or polar coordinate system. This makes it more clear you can shuffle the location of nodes to however is best to visualize the connections - which may not be any regular geometric pattern.

$\endgroup$
  • $\begingroup$ This is really cool. Definitely an acquired taste, but very clever. $\endgroup$ – Ismael Ghalimi Mar 29 '16 at 0:42
0
$\begingroup$

I would argue that the Sankey diagram can be considered a linear version of the chord diagram.

$\endgroup$
  • 2
    $\begingroup$ It would be useful for CV'rs if you would elaborate further on the similarities you've observed...beyond simply stating the fact. $\endgroup$ – Mike Hunter Mar 28 '16 at 11:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.