Data Visualization vs Art

Data visualization is the art of creating images which display a set of data in the way which makes the most important tends and patterns easy and intuitive to understand.

For instance

Temperature (C)	Quad Occupancy
0	0
3	0
5	0
6	1
7	0.5
9	2.3
10	5.1
12	6.7
13	6.1
15	8.6

Looking at the table on the left it is clear the Quad in occupied on average by more students at warmer temperatures but the shape of the relationship is hard to grasp just looking at the table. By plotting it as a scatter-plot ('visualizing') on the right it is clear no-one wants to be in the quad below 5 degrees then the number of students in the quad increases roughly linearly as the temperature increases and there is still no signs of the Quad getting full even once on average 8.6 students are sitting/working out there. Recently there is a fad for more and more fancy graphics to display larger and larger amounts of data at once. The problem with this is that increasingly data visualizations are more about the art and prettiness of the plot than about conveying information. Here are a few examples of modern data visualizations I've snatched from the internet:

Offender 1

Creator: Pehr Hovey, Source: http://pehrhovey.net/blog/2009/03/3d-data-visualization-dewey-calendar/

 First off just looking at the image it is hard to know what the data you are looking at actually is because there is no title or axis labels. If you are familiar with libraries you might recognize the legend on the right is referring to the Dewey decimal system so we are looking at book borrowing/usage from a library. If you are used to the American month:day date system you might also guess that each column is representing a day and the columns are arranged as a calendar for one month. From the source page I found out that the height in this case represents the "number of transactions" for each book category. Once you have spent a few moments deciphering what it is you are looking at you are still no closer to actually obtaining understanding because the 3D arrangement causes the columns at the back to appear smaller because of perspective making it extremely difficult to compare their height to the columns in the foreground. Similarly, the bars in the foreground are not parallel because the bases of the columns are not on a flat plane but on a curved surface, again making it extremely difficult to compare their heights (were there more total transactions on 01-28 or on 01-30?). So really you can't tell anything about the trends or patterns of book usage from this figure at all other than that Arts (700) books are used much more overall than the other categories - but how much more is unclear. The same information could have been imparted using a simple 2D barchart of the average number of transactions for each type of book each day with error bars indicating the variability between days - this would have made it easy to tell exactly how much more the Arts books are used compared to the other books and revealed which type of book has the most variability in usage and would have taken very little time to both create and interpret.

 Offender 2

Source: GE Data Visualization http://visualization.geblogs.com/

This graph at least tells us what the data is it is displaying unfortunately that is about all you can learn from the figure. Once again the height of the graph, which is supposed to be the information we are interested in, is very hard to compare because the plot is in 3D so the bars farther away are smaller than the bars in the foreground. Additionally, there is no scale for the height so for all we know these could be tiny gas generators like they use for backups in hospitals. An additional sin is that the labels around the circumference of the plot as unreadable except for the quarter in the foreground. Also what is the point of having a barplot arranged in a circle other than make one half of the plot invisible? Pretty much all we can learn from this plot is that there are quite a few turbines in the world (exact number is not evident in the figure) of differing capacities - this could be more effectively conveyed with 4 numbers: the total number of turbines, number of countries they are found in, and the range of output they can produce.

 Offender 3

Source: http://seeingcomplexity.wordpress.com/2011/03/13/why-visualize-data-we-dont-know-yet/

Another example of bad data visualization: some terms on the left and different countries/continents on the right, the area of the orange/brown arcs indicate population size. I don't know what the grey lines are supposed to represent (nor their colour/width) but it doesn't matter anyway since it is impossible to trace the path of almost all of the lines so impossible to know what is connected to what anyway. I guess I should give them some points for resisting the urge to unnecessarily use 3D graphics and at least having most of their labels be readable - except those ones at the bottom which are some of the few you can actually trace the links for. So yeah different places have different numbers of people (how much more or less is hard to tell since people are in general very bad at estimating areas) and are connected to different abstract concepts in different ways, feeling enlightened yet?

 Offender 4

Source: http://blogs.uoregon.edu/teamingwithmicrobes/media/dataviz/

Another data visualization offender. Network visualizations are 90% of the time a complete waste of space because the only information they convey is "its fucking complicated". This network represents different bacterial species living in a corn seed (dots around the outside) and their co-occurrence in the same corn seed (lines). The colour of the line indicates whether they co-occur more  (blue) or less (red) often than expected by chance. The size of the dot indicates how frequently the bacteria was found in corn seeds. We can get little information from this, common bacteria co-occur with many other species but not more often than expected (usually) while rare bacteria tend to co-occur more than expected (some of the time) with many other species, unfortunately the dots aren't labelled so we don't know which species these are nor does this arrangement of the network reveal whether the rare bacteria form multiple distinct groups which co-occur or whether they all are just out-competed by the common bacteria. Also because the figure is so messy it is hard to know if any of the patterns we see are statistically significant or the strength of those patterns (how much more likely are rare bacteria to co-occur more often than expected with other species than common bacteria?). Looking at this figure could help you form hypotheses which would be interesting to test but it doesn't reveal any answers or impart any knowledge in itself.

My Hero

Source: http://enthusiasm.cozy.org/archives/WorldEd.png

Finally an example of effective data visualization. Notice the helpful axis labels explaining not only what the axis is but giving the units and values as well, the exquisite legend elucidating the meaning of the size and colour of each data point, the title saying what the graph is as well as the date the data is from. Unfortunately this low-resolution version I found makes the data labels hard to read but each country is represented and labelled. The trend is blissfully obvious (high education is associated with high GDP) without requiring you to squint or guess what is going on. Subpatterns and variability is easy to see as well - Africa is stuck down in the bottom-left while Europe is mostly in the top-right with the Americas, and Asia in between. Just as much data is represented here as in most of the other figures I've discussed (each country its population, GDP, education level) but the plot is much clearer and informative (and simpler to create) than the artistic but meaningless 3D images or the messy unreadable networks.