Complex Diagrams

Electoral-Vote.com is an excellent and popular site for tracking polls for president on a state by state basis. One of the nice features is a historical graph that shows the changing number of electoral votes for each candidate over the election season.

Currently it looks something like this:

(click to see the original on electoral-vote.com)

Just for fun, I decided to see what it would look like showing the candidate totals as a single shared line. (This is similar to the epic graph representing the Senate History Timeline.)

In the revised version, the vertical axis counts votes for Obama going upward, as in the original, and votes for McCain going downward. The axis labels are color coded to reflect the candidate they represent. The color of the line reflects the candidate with more votes for that time period.

Theoretically, when the votes total line is above the 271 line, Obama is ahead; when it’s below the 271 line, McCain is ahead. However, the revision isn’t perfect, and accuracy suffers relative to the original in a few ways.

• In situations where a state (such as Florida) is tied, neither candidate is credited with the votes, so the total number fluctuates. This is not handled well in the revision.

• It’s a little trickier to represent the situation when the count is very near 270 (see late April and early May in the original).
• Because the total number of electoral votes is 538, it’s impossible to use round numbers at both ends of the shared grid lines.

While not a superior solution overall, the revised version does have some merit. As a rough tool for observing trends, I think a single line is cleaner, easier to follow, and the changing colors are a little easier to tell a story with.

Zillow has posted a series of excellent diagrams which show relative changes in assorted housing markets, broken down by segment. Their diagrams are very clear and allow fairly quick access to a lot of good information. Please go explore there, and then come back here for my commentary.

By my count, there are eight different axes of information represented for each city or market:

1. lower bound of each market
2. upper bound of each market
3. overall span of each market
4. lower bound of each segment
5. upper bound of each segment
6. overall span of each segment
7. direction of change for each segment
8. magnitude of change for each segment

The first six of these axes are a natural byproduct of using the (horizontal) stacked bar chart format that they’ve chosen. Given how bad bar charts typically are (N.B. Excel), they did a great job with the shapes, colors, and generally minimalist approach. They’re really quite good.

To fully optimize the diagrams, I’d make a few minor changes.

Sorting the markets alphabetically makes it easy to find specific markets by name, but difficult to compare them. Regional groupings might be more useful. In an ideal world, the user could select a few markets by name and compare just those selected few.

The weakest point, in my opinion, is the ambiguity introduced by the arrows. I understand that they are meant to represent trends, not specific quantities. However, they can easily be made a little more clear.

Judging from the legend, the length of the shaft of the arrow represents the actual quantity, while the head encodes direction (redundantly with position and color). In some cases, such as our hometown of Seattle, the heads are ponderously large compared to the length of the shaft, making anything more than a cursory comparison impossible.

Removing the heads and using rectangular bars would make it easier to compare values at a glance. If we wanted to get more specific, the percent change per segment could be displayed within the segment oval, in red or green text.

While not quite as dramatic, I think these changes make the diagram more useful, without reducing their accessibility.

Finally, and this is a minor point, the legend states that “the top and bottom groups have been cut for better visualization.” This is entirely reasonable, but it made me curious about where they chose to trim those groupings. It would be nice to know what fraction of the actual number of homes are represented in the high- and low-end bars.