It's only better if you -- by reading up on convention beforehand -- know which way an edge points; does decreasing thickness indicate an incident node, or the opposite? Not clear on its own.

Agreed; my initial interpretation was thinking of the tapered edges as comet tails or motion blur, which is the opposite direction to what they want to depict.

I also thought the thick end meant it was receiving flow as opposed to originating it. It's also hard for me to look at the graphs and not see motion, particularly the curved ones - years of reading comics. But reading graphs is an expert activity, and I'm open to the idea that with practice this method could become better.

It's an arrowhead. It goes in the direction it points.

For example, (A -> B) is drawn:

  A @===---- B

They test a few different schemes, but it is sort of embarrassing that in their final summary graph (figure 9), they use the arrowheads again to indicate which representation performs best.

> It's only better if you -- by reading up on convention beforehand -- know which way an edge points

Actually, if I'm reading the paper properly, no it's not. The authors appear to have found that both the light-to-dark and the dark-to-light styles performed significantly better than arrows, and that the tapering effect was better still.

This seems reasonable to me: even if you're unsure about the convention at first, you probably only have to find one obvious edge to establish which way things "point".

Umm... that's great for printed materials but how are you going to make that work on a white(black)board or for that matter with a pencil?

Not sure I've done many directed graphs in print but I've done quite a few with a marker/pencil/chalk.

While I understand how arrows on the end of the edge is a bad idea, especially if you have a lot of edges on one node; what about having arrows blended in the edge itself?

Something like this (beware the ASCII art): o---->----o

I often do this in hand-drawn graphs for my own consumption, when the line is very long (sometimes using three arrowheads): o-->---------->----------->o

BTW: paper mentions the problem of visual clutter of the arrowhead itself.

I really like the gradient use on the edges. That, along with the taper, allows a reader to easily see the direction without having to follow it till it hits another node.

Interesting, I feel for "trained" users who use directed graphs for work this is a valuable technique. But even the researchers admitted it is not obvious which way the flow goes. So I don't think it would work well for advertising or public presentations.

The research paper makes interesting reading, and how often can you say that!

I have a graph theory based application in the Android market, "Edgy," and I doubt that I would switch to this. I think once you get used to it, it is perhaps clearer, but I think there are many opportunities for styling and weights that may increase clarity without having to do something that is not expected. If this catches on, of course I'll probably go that way, but as for now, I haven't even seen many, if any, graphs use this.

They appear to have skipped curved arrows, which would reduce the muntedness of the graph considerably. Do please correct me if I skimmed poorly.

I much perfer having gradients on the edges if you are going to do something nonstandard