From CS 294-10 Visualization Sp10

Jump to: navigation, search


What most interested me in the dataset Totals page was the huge "Other" category. Therefore my visualization attempts to tease apart, in a bit more detail, what types of groups are giving and how the individual donations in each type compare to those of the other types. To do this, I first classified all the donors on the full details page into the 6 categories indicated in the visualization. I then aggregated all donations made by donors of the same name. This is a little problematic especially for the World Organization category because some subgroups of a large organization (e.g., the UN) were listed as separate entities. I ordered the aggregated donations in decreasing order. Another problem here is that I chose to order ties as they appeared on my alphabetized list which creates unmeaningful blocks of color at some donation levels.

From this visualization, I hope the reader can see immediately that 2 donor types are giving the majority of the donations. However, in the World Organization type, there a few large donations whereas in the Government type, the donations are more evenly spread across the spectrum. This pattern is echoed in the two mid-range donor types: Aid Organizations / NGOs have a few larger donations and Corporations have many smaller donations. The Foundations are mostly smaller donations with one outsider. And finally, Private donations is a single fairly significant amount.

The general scheme I use is a series of small multiples that highlights each donor type singly. I use a single hue against neutral gray to indicate each donor type within the small multiple so that the type *pops* preattentively. Because of the small multiples the viewer is able to compare across types. Alternatively I could have put all this information into a single pie chart and bargraph series. In this case I thought it would be harder to compare the individual types b/c each hue would be competing for attention. I chose the pie chart to show the donor type's percent of the total. In this visualization I think pie charts work best for this because 1) they immediately give the impression of a percentage as a part of the whole and 2) they are a nice contrast with the barcharts (i.e., if I had used a barchart for this comparison, there would not be the obvious separation of the two concepts). I chose a rank order bargraph to show the relationship of the individual donations in an ordered fashion -- the viewer can get a sense of how much of each amount has been given in relation to the others. This is somewhat of a combination of position and length to code this quantitative variable. (Position, at least, as Cleveland and McGill define it.) I think it is likely that the combination of hue and length/position of the bar results in a redundancy gain for comparing the donor types across the small multiples. Because the donations are heavily skewed I broke the graphs to show decreasing ranges (factors of 10) in more detail. Alternatively I could have log transformed the data but I think for a general audience this is not so intuitive. By keeping the successive scales of the individual donations in real $, the viewer easily understands the huge difference between the high and low donation amounts as well as the relative quantity of different donation amounts.

Some self criticisms: 1) The text is a bit small on the graphs; I should have added the percents or $ amounts on the pie chart labels; the break outs could maybe made clearer (at least a more visible dotted line) or better understood; and I did not keep the width of the bars constant across the break outs -- biasing the smallest scale. 2) There is a bit of wasted ink, especially in all the gray of the pie charts -- perhaps I should have had one central pie chart and the multiples all coming off of it. 3) I paid little attention to hue choice -- they probably could be better.

Other notes: 1) I did think it was odd that the totals on the Total worksheet and that on the Full Details worksheet were off by $100M -- couldn't figure out where. Additionally, I couldn't figure out what was going on with the Funded and Unfunded worksheet -- this could have been interesting if one could assess who was giving to what specific aid areas. 2) I began this investigation by first looking at the proportion of corporations that were based in the US -- there were so many corporations that I thought they could add up to a substantial amount (>25% of the total). When they didn't I focused my attention on the broader picture. Pretty amazing that the World Organizations have so much money -- would be interesting to trace their payrolls back to countries. There were far more US corporations making donations than foreign corporations - didn't run the numbers but I would guess the majority of this group is from the US. 3) I first played with the data in Excel, mostly the Totals worksheet. When decided to look more carefully at the Full Details worksheet, I assigned donor types in Excel and then transferred the data into R, where I did the real work up and generated the graphs individually. I put them all together with the individual donor callout annotations in Adobe InDesign. Looking forward to a more streamlined approach.

[add comment]
Personal tools