<-- Advertise here.

directed_edges.jpg
Networks are often visualized using points and interconnecting lines, with triangular arrowheads at one or both ends to show any directionality between the different points. Although such a standard arrow representation seems intuitive, it can lead to problems in dense graphs that contain many incoming or outgoing relationships. Furthermore, since the arrowheads often have approximately the same size and aspect ratio as the small circles they connect, the graph as a whole might be perceived as cluttered with so much visual detail to the point of being distracting.

Several alternatives of depicting lineair directionality do exist. For instance, lines can vary their hue, luminance or width, along their length. But are these depictions significantly better than just using standard arrows? Which technique performs the best? These questions formed the main focus of the scientific research described in the recent paper "A User Study on Visualizing Directed Edges in Graphs" (PDF) [win.tue.nl] by Danny Holten and Jarke J. van Wijk at the Eindhoven University of Technology.

Their experiments consisted of testing the different visual techniques (or combinations of the techniques), on which participants performed specific tasks in which they had to answer whether or not there were directed connection from one point to another in a randomly generated graph. Response times and accuracy were measured and analyzed. The different techniques tested were: "arrow", "light-to-dark", "dark-to-light", "green-to-red", "curved", and "tapered".

So, you might ask, what can information designers learn from the results from this research?

. A "standard arrow" representation should be avoided whenever possible. Although this representation is straightforward and intuitive to most users, the performance of this popular representation is quite low, which is probably due to the use of arrowheads that cause occlusion problems and visual clutter.

. A "tapered representation" in which the width of an edge is gradually varied along its length - wide at the start and narrow at the end - proved to be the best representation in terms of performance.

. For an "intensity-based" representation (which is still a better choice than the standard arrow representation), a dark-to-light representation has a performance advantage over a light-to-dark representation.

. There does not seem to be a clear performance benefit associated with combining 2 techniques together. Therefore, the researchers recommend to use a single technique instead.

A graphical summary of the relative performance of the different techniques can be found below.

MORE

directed_edges2.jpg
Graphical depiction of performing Tukey's HSD test on the combination of all tasks and layouts. An edge is present for each statistically significant difference. Edge direction indicates that the origin representation performs significantly better than the destination representation.

15 COMMENTS

The evaluation occurs on a very trivial graph...In practice, I doubt that tapered edges remains efficient on bigger graphs (>10,000 edges) on the contrary of curved edges.

It would be very interesting to see this study applied to a large dataset like http://lumberjaph.net/github-explorer.html .

Thu 10 Jun 2010 at 9:47 PM
Seb

I love the use of curved arrows to present the results.

Thu 10 Jun 2010 at 10:31 PM
J

I prefer arrows.
Cluttering can be avoided by smart placing of the arrowheads (they don't always need to be on the node) and sizing of the nodes.
Like in the graphical summary you present yourself.
An arrow is direction marker #1 in almost every culture.

Fri 11 Jun 2010 at 1:44 AM
Trumpetto

Very nice and interesting work. As usual, by Jarke J. van Wijk, who is one of the most inspirational persons in InfoVis I've ever known :)

Fri 11 Jun 2010 at 1:58 AM

If they conclude that the Arrow representation should be avoided, why did they used it in the graphical depiction of the results?

Fri 11 Jun 2010 at 6:44 AM

I doubt this because of basic oddities like suggesting NOT dual-coding. But I'll be /trying/ some of their recommendations for the next revision of my diagramming standards document.

And for infosthetics.com, please stop underlining things that aren't links. The U tag is deprecated anyway, but it's confusing and wildly unsemantic. Use the em tag (emphasis), and it'll italicise, unconfusingly.

Fri 11 Jun 2010 at 1:39 PM

@Steven: Thnkx for the critique. I have changed the underline to emphasis in the post. I realize it might appear (more) confusingly in RSS feeds and such.

Fri 11 Jun 2010 at 6:54 PM

I always thought that using arrows was the best technique. I will not think the same anymore.

Although, may I asked why the graphical summary uses arrows while this article says: "A "standard arrow" representation should be avoided whenever possible." ; )

Sat 12 Jun 2010 at 12:57 AM

Might it be that the different visualizations serve different goals? The arrow heads highlight the destination more so than the origin. The tapered highlights the origin rather than the destination. The designer may have a preference for what they are trying to emphasize. From what I understand above the test was on identifying if a directed relationship existed between 2 points, a different goal than the 2 I mention.

Sat 12 Jun 2010 at 1:08 AM
Taylor

Actually this post covers specifically the topic that's giving me some headaches at the moment.
I created a visualization of some worldwide capital flows, which you can find here:
http://www.flickr.com/photos/etnobal/4699622964/
At the moment I just add up the flows between two countries although I have seperate data for both directions. In my opinion the direction really matters for intepreting the graphic.
The problem is I can think of no practical way of displaying in which direction a line heads in a static picture. I'm not sure any of the ways depicted in this post would be practical. Does anyone have a suggestion on how to solve the problem?

Mon 14 Jun 2010 at 11:14 PM
Philipp

Great post. Love van Wijk's work.

I think the obvious one that has been overlooked is animation. I know, this is about static display. There is such a huge benefit to using interactive techniques, I'm surprised anyone is still focusing on static representations.

I also don't understand why they used the two least effective methods (curves and arrows) to display the results. Was it tongue in cheek? Funny enough, but for a paper? Hmm.

Tue 15 Jun 2010 at 4:33 AM

These are great questions to ask for static visualizations, but dynamic presentations of data can also employ motion to indicate directionality between nodes. See this project:

http://alignedleft.com/blog/2008/12/visualizing-network-relationships/

Tue 15 Jun 2010 at 8:47 AM

Well, obviously animation does help. I don't think thats an issue, as there are many ways to illustrate directions (for example a gradient line that's moving from one node to the other).
The real challenge I see is depicting direction and motion in static pictures. I think the problem is, many people expect things they see on the internet to be created just for the screen. There are lots of graphics that are intended for publishing in printed magazines or books, especially in academics. Unfortunately for those purposes motion is not an option.

Tue 15 Jun 2010 at 9:45 AM
Philipp

its interesting to see that various types of arrows can make that big of a difference within a presentation of data. This just goes to show that people have different preferences in the way they see things.

Thu 17 Jun 2010 at 3:25 AM

St├ęphane and Kim, they did address their choice in the paper: "(although 'arrow' was found to be a poor candidate, we nevertheless use this intuitive representation in [the performance figure], since occlusion and visual clutter will not be a problem in this case due to the limited number of nodes and edges)"

Mon 07 Feb 2011 at 9:13 AM
Till
ADD A COMMENT
Commenting has been temporarily disabled.