In the most recent episode of Tableau on Tableau, Dustin Smith and I briefly mentioned what we refer to as the “Andy Cotgreave” method – which is reflected in a lot of the British Andy’s visualization work. I brought up Andy’s tendency to defy convention in the interest of making visuals which will start conversations and draw his audience in. There are pros and cons to this approach.
On one hand, and as more comprehensively described by Robert Kosara, a data graphic designed purely for presentation purposes can be more memorable if it contains creative or pictographic elements. This can make an audience remember and care about your point even without a complete understanding of the data.
Alternatively, breaking visual best-practices can distract from the real conversation if you’re in a room full of data viz experts, and it can create real barriers to entry for your audience. If they can’t quickly understand your view, it doesn’t matter how great it looks.
Perhaps this topic is what made me take the approach I took for this weeks’ Makeover Monday exercise. First, take a look at the source viz, published on cnn.com.
I looked at this viz and read the article, and these were my initial impressions:
- Easy to see regional trends. I can tell that a majority of the “bad” states are in the South.
- I can flip through using the arrows to see the same map displayed for different elements of the model described in the article.
- It uses a consistent color palette, clear and concise headings and legends, and avoids unnecessary embellishments.
- The article describes a bunch of additional points that aren’t readily viewable in the data. What use is a graphic if I need to read an article to understand the data set?
- Real rankings are not available – I can see the top 13 states but I can’t tell which one is 1 and which one is 10, etc.
- Though the model is a composite of four factors, I can’t easily check which factors have effected a particular state’s ranking.
In short, this viz could use a fair amount more detail so the article can be more easily understood. As is, I’m likely to just skip the graphic and read the details.
For my effort, I found myself gravitating toward being able to show each of the four factors. Something I found interesting when exploring the data set was the fact that a state might have a particularly high ranking in one category, but rank low in others. I wanted to be able to call that out in relation to other states. I tried text tables and bump charts, and it’s difficult to convey both the individual values and total aggregates in a single, consumable chart.
Eventually I found myself exploring the idea of a radar chart.
A quick aside on radar charts: They aren’t a great practice. They display data at odd angles where numbers can’t easily be compared, and they connect data points that often shouldn’t be correlated. Additionally they apply position to measurements where none exist, possible creating bias in a consumer when they see one number positioned in a more prominent location than others. Usually a bar chart is better. Here, Alberto Cairo writes on the topic (and links to Stephen Few). These guys can explain it a lot better than I can. I just know it’s a no-no.
Nevertheless, I found myself wanting to attempt this chart type, partially because I’ve never made one before, and partially because I thought it would be a unique and interesting approach. My justification: combined with other graphs this can create a catchy visual that will draw a user in and create a memorable image in their mind. If we were simply analyzing data, it wouldn’t be a good idea, but we have an opportunity to present an information graphic with interactivity where the discrepancies across the data set can be made more clear.
Here’s what I ended up with (click for live viz):
I intentionally published this with a selection because I think that’s where the viz is most powerful. You can see where the selected state ranks and how it compares to average / other top states. You get unusual oblong shapes when clicking through the data, which helps you remember elements related to specific rankings even if you can’t remember the hard numbers.
Another thing I experimented with during this exercise was layout. This is a very wide viz but I also tried out a tall and skinny layout:
The portrait mode allows you to see the entire visual on one page rather than having to scroll. It also features the two visuals more prominently at the top of the page and lets you look at the crosstab for more detail. In the end it felt a little unwieldy to me because of how tall it is, and I thought since the viz was interactive anyway that it wasn’t too big of a deal to ask a user to scroll through the table. But even without interactivity, I think the left-to-right eye movement required to read the landscape version made for a more compact and readable view.
I’m still not sure that the radar is the best approach. In fact, I’m relatively certain that it’s not. But it was a fun exercise and I’m curious to hear what other people think.