Archive for the 'Multivariate' Category

02
Jan
12

An Unintelligible Language

Gentle readers, our first map of the new year is one that I am finally getting to eleven months after it was brought to my attention by a reader, Matthew. It concerns a favorite subject of mine, American English dialects, and was produced by hobbyist Richard Aschmann.

Click to visit Mr. Aschmann's page on North American English dialects.


The style of this work will be familiar to those with an interest in language mapping, with boundary lines delineating different pronunciations and vocabularies. Here’s another one from the Telsur Project at the University of Pennsylvania:

Click to visit Telsur project page

While Mr. Aschmann’s work is of a conventional type, it is also by far the most complex I’ve ever seen, and therein we find the problem. There is simply too much going on in this one map to be comprehensible.

One of the primary things a map reader is going to want to do is look for spatial patterns. After all, this is quite probably the entire point of having a thematic map — showing a relationship between what happens and where it happens. If you there isn’t one, then you might as make a table, instead. Now, in the case of Mr. Aschmann’s map, there’s certainly a connection between where people live and the sorts of speech patterns that come up. The problem here, though, is that this pattern is nearly impossible to discern.

To be able to see how dialects change over space requires that you look at a certain region, determine its characteristics, then look at a second region and do the same, then a third, and so on, comparing them all along the way. Your eyes sweep across the map, and each time you take a quick read and compare with what you’ve already seen. But this only works if that read can indeed be quick. With Mr. Aschmann’s map, figuring out what’s going on in any one location is a significant chore. There are so many possible symbol types, sorting through the legend is a challenge. Just figuring out which set of lines your target area falls within can be difficult, given how many layers crop up. Even if a reader is interested only in looking up data on a single place, and not making comparisons or seeing patterns, the density makes it nearly too much trouble to be worth checking. Once you’ve successfully figured out what’s going on with one region, you can move on to the next region to compare. But by the time you’ve waded through the decoding process a second time, you’ve already forgotten what the first region means. Comparison, and therefore pattern recognition, is nearly impossible, because your brain simply can’t hold that much complexity at a time or absorb it fast enough.

Compare this with a simpler map of rainfall, below. Here, it’s easy for you to quickly spot the distribution. The color pattern is simple, and you need only look for one data set, instead of twenty. There are a couple of other reasons that this map is a bit simpler to read, as well, having to do with the symbology type, but the great majority of the difference is simply in complexity.

Grabbed from Wikimedia Commons

I understand well the urge to include multiple data sets on a map, and longtime readers may recall seeing an overly complex, multivariate map of my own on this site. The more complexity you can show, the richer the story and the more versatile the product. The map quickly begins to be more than the sum of its parts. Putting two thematic layers on a map gives you three data sets — one each for the layers, plus allowing you to visualize the relationship between the two layers. One plus one equals three. But all of this is worthless if it becomes so complex as to be unclear. A map with one clear data set is worth more than a map with fifteen data sets you can’t read. Good mapmaking is about making space intelligible — otherwise, why make a map?

This map needs to be split into a series, each of which tells its portion of the story clearly. The topic it is attempting to portray is deep and rich and complex, and any single map that attempts to encompass so much is likely to end up like Mr. Aschmann’s: uselessly dense. Not every subject can be condensed into a single visual statement, and there is no shame in breaking it down into a series of simpler points in order to clarify.

Before I leave off, I’ll also mention one other thing. This map, like so many others, is going to be even less intelligible to the millions of people out there with color vision impairments. If you happen to have standard color vision and would like to see what I’m talking about, check out Color Oracle by Bernhard Jenny.

I’ve been trying of late to focus more on major items in my critiques, rather than dealing with too many nitpicky details, in order to not repeat too many points from earlier posts. Thus, I leave discussion of the rest (such as the quality of the labeling) to you, dear readers.

One Nice Thing: Mr. Aschmann has done a valiant job of trying to ensure that everything is layered clearly, which is no small task given how many data sets are crammed in. No one data set actually obscures another. There’s still far too much going on to be useful, but it’s not impossible to pull some information out of it if you’re willing to sit down and work at it.

16
Jun
10

On the Abuse of Chernoff Faces

Good day, gentle readers. It turns out, not surprisingly, that no one entered my little redesign competition, and so I’ve no news to report on that front, sadly. Bravely I shall soldier on, however, perhaps to give it another try when circumstances have changed.

Rather than discussing a specific map this time around, I want to take aim at an entire symbology: Chernoff faces.

A face map by Eugene Turner, 1977

You’ve probably run into this sort of multivariate symbology before — using faces to convey data. It’s an intriguing idea. As Herman Chernoff proposed in 1973, we can leverage the power of humans to recognize faces to easily communicate information. The face becomes a gateway for people to see patterns in the data.

When Eugene Turner made the above map, he knew that faces carry emotions. As he said on his website, “It is probably one of the most interesting maps I’ve created because the expressions evoke an emotional association with the data. Some people don’t like that.” A face symbology can give people empathy with the numbers — high unemployment is sad, high urban stresses cause anger. Turner could just as easily have made a multivariate symbol map which used abstract geometric figures rather than faces — say, a cross, with each of the four arms changing length according to the data. The map would convey the same information to the reader, but the emotional content — so much of this map’s power to influence readers — would be lost.

Here’s one problem: if you’re using faces, you’re using emotions, so you’d better be prepared to make emotional statements about your data. Empathy can be a powerful force for the narrative you’re trying to convey, but it’s also hard to escape.

By Aaron Rothberg, 2007. From: http://aaronrb204.blogspot.com/

This is a student map, from the looks of the website it comes from (not from my university, however), and while I try to avoid bringing up the work of students, this one happens to be a good example of this problem of reading faces. According to this map, it’s sad when people over 50 are executed, but it’s pretty happy when people under 40 are. That’s going to rub a lot of people the wrong way, I would think. This map also suffers from an issue that Turner’s map does: the skin color of the faces. In each map, the fewer white people in an area, the darker the face gets, towards a skin color presumably suggestive of African-Americans. But there are plenty of non-white people who don’t have dark skin. It sets up an easy and dangerous racial spectrum that runs between white and black.

Here’s another Chernoff example symbology, one taken from an ESRI conference paper:

From Spinelli and Zhou, 2004, linked above.

One of their example faces, assembled.

The eyebrows are pretty emotionally charged, and are here linked to how many women are in the workforce. Using their symbology, if you have an area where there’s high unemployment and a lot of women working, you get angry-looking faces. On the other hand, if there’s not a lot of women working, and high unemployment, the faces look sad and depressed. Is this at all sensible or appropriate? More tense emotional states seem to be on display the more women there are in the workforce.

There’s another issue here, besides the dangers of conveying unintentional emotional messages, and that is the simple problem of a nonsensical mismatch between the data and the way its being conveyed. Do places with higher crime have denizens with bigger ears? Does divorce make your nose bigger? Look, it’s not always possible, or even advisable, to make strong visual connections between the symbol and the data (or, if we want to draw on my limited knowledge of fancy semiotics terms: to reduce the gap between the sign vehicle and the referent), but faces seem to me to pose a special case. Perhaps it’s the deliberateness of choice — again, the author didn’t go for something abstract or geometrical, they went for a human face. The reader is not expecting something as out-of-the-blue as “their nose gets bigger when there are more divorced people.” This kind of nonsensical connection breaks the very humanity that the symbol is going for. We know people get angry or sad when there’s high unemployment, and we can relate to that, but their hairline doesn’t recede as their income drops. Why use a face, in the first place, then? These sort of mismatches bother me, but I’m still working out an articulate explanation as to why — perhaps you all can help me with that.

One more example:

One by Daniel Dorling, 1995

A place with a lot of young voters has a big fat nose? And your eyes get bigger, I suppose, if you’re likely to be in a service occupation. Again, people might disagree with me on whether or not this is a problem. I think not paying enough attention to the emotional content of a face is a bigger issue than these sorts of lesser mismatches between eye shapes and % service employees. But the deliberateness, the unusualness, of employing a face suggests to me that I should be looking for a connection, and I am frustrated not to find it.

Chernoff faces can let you bring a lot of power to bear on social data, by showing how people feel right on the map. But it’s easy to squander or misuse their great potential by treating them as simply something cute, amusing, or attention-getting. They require some thought to use.

One Nice Thing: I applaud all of these people for giving Chernoff faces a try — they are a challenge to employ, and not just for the reasons I discussed; they’re also time-consuming to actually draw in most cases. And everyone here used them for social data, which seems the right idea to me. Data about people, shown with faces of people. Much better than say, geological data, which Chernoff used as one of his initial examples.




Follow

Get every new post delivered to your Inbox.

Join 69 other followers