InfoVis as Seen by the World Out There

From CS294-10 Visualization Fa07

Jump to: navigation, search

Lecture on Nov 5, 2007

Slides

Contents

[edit] Readings

  • Review of Tufte's Beautiful Evidence. Few (pdf)
  • Save the Pies for Dessert. Few (pdf)

Optional Reading

  • Review of FYI Visual. Few (pdf)

[edit] David Sun - Nov 05, 2007 03:01:37 am

The main criticism of the pie chart is their poor encoding of relative rankings among data points. The area chart on p12 appears to suffer from the same problem, for example, it's impossible to rank Company D, B, and E in the years 2004 and 2005. We may be able to more accurately gauge rankings from lengths compared to areas and angles, but the area chart really doesn't do any better than pie charts. Having said that, it is more effective in eliciting temporal trends compared to using multiple disjoint pie charts. Generally it seems that as soon as we plot quantitative data against different baselines (i.different zeros) and not in parallel to each other (like in a bar chart) we have difficulties in visually ranking the data.

Review of FYI Visual: What was Glyphs original conceived for in the context of infovis?

[edit] Nate - Nov 05, 2007 11:12:41 am

I find it amusing when the "dashboards" of business intelligence tools include widgets based on gauges found on the dashboard of vehicles - as if executives can really "drive" the business based on a few simple gauges of business activity.

I find it ironic that the focus of the layout for the Smashing article on visualization is advertising, rather than the article itself.

I enjoy the contrast between Many Eyes and Swivel - I think differences stem partly from each effort's financial position and freedom to focus primarily on visualization best practices.

I am biased, as I come from a corporate finance background. My understanding is that the target market for Excel is corporate management, a field in which accuracy of numbers are critical (for accounting purposes, Sarbanes-Oxley requirements, etc.). I don't think of Excel as primarily a visualization tool - in fact, I would guess that the majority of the time there are incentives for corporate presentations to "massage" the conclusions drawn by the data.

[edit] Robin Held - Nov 05, 2007 02:51:06 pm

The periodic table of graph types was amusing. But in addition to being a funny example of poor visualization, I think it illustrates a lack of confidence in the intended audience. The designer clearly wanted to use a familiar form of visualization to teach a new topic. Here, a table would have been sufficient, if not optimal. But instead, graphical elements are added to make the information more palatable. Ironically, the result is the exact opposite. The display is cluttered and confusing. The inappropriate use of the periodic table reminds of the overly shiny graphical elements Steve mentioned later in lecture. The emphasis is all on the initial reaction from the user to the aesthetic quality of the visualization, rather than effective presentation.

[edit] Hazel Onsrud - Nov 04, 2007 01:18:31 pm

I learned about fractions by cutting up a construction paper square, rather than slicing a circular pie. Perhaps a casserole approach to the graphical pie issue would address some of Few's issues with the display, while allowing the public's love of graphics modeled after food to remain, although tweaked. For example, splitting up a square into smaller pieces really does allow one to get an instant idea of the magnitude relationship between the areas and the viewer does not have to wrestle with the properties of circles which are difficult to estimate (angles, areas).

I am sure there are many reasons why we don't use this technique often. It wouldn't work with all of the examples Few cites, and some of the inaccuracies involving area estimation would remain. In fact, if changing public practice, one ought to aim for perfection...but this friendly casserole approach did help me learn about fractions.

[edit] N8agrin - Nov 06, 2007 11:02:29 am

I appreciated Few's review of Tufte's latest work. His comments on Beautiful Evidence reflect my own general opinions on Tufte, namely the great theoretical guidelines he provides which are highly relevant, and then the sometimes bizarre and often misappropriated examples he provides. I have heard much of Tufte's discussion of Sparklines in the latest book, and as Few indicates, it may still be worth reviewing Beautiful Evidence in order to gain at the very least an insight into what Tufte's approach and technique has to say about their use.

In so far as the discussion of pie charts, Few reiterates the points Tufte and others have already made. Essentially, pie charts are difficult to interpret and nearly impossible to use as a tool for discovering visual correlations between data series. Still, his explanation is worthwhile as it is aimed at the common man, and does not attempt to explain the complex cognitive barriers which make pie charts hard to interpret. By discussing real world business products, like MS Excel, that common visualization creators will have at their reach, Few successfully explains best practices in creating visualizations, regardless of how these products might render data by default.

[edit] David Jacobs - Nov 06, 2007 04:09:03 pm

Hazel: Though I find your proposed casserole chart intriguing, I don't think it succeeds in addressing the problems associated with pie charts. First of all, casseroles vary widely in shape and size. A small sample of casseroles shows that 25% of recipes are made for circular dishes, while the remaining 75% vary widely in aspect ratio and area. I think a fundamental problem with such an approach is that there are many ways to represent the same value, because we're essentially representing a one dimensional value (percentage area) using a two dimensional space (delicious casserole). This brings additional difficulty, as an example, consider the task of determining which region is bigger, the red or blue food-colored regions of this tasty hash brown casserole. Image:Tasty_hashbrown_Casserole.jpg

In a pie chart, the two-dimensional space is reduced to being 1D as it must be proportional to representative angles. I think the solution could be found with an inherently one dimensional food metaphor. I propose a standard 3-foot (0.9144 meters) Submarine Sandwich Chart. Because submarine sandwiches have a fixed (and narrow) aspect ratio, there is an implied cut direction (perpendicular to the long axis of the sub), meaning that all sections areas will be proportional to the length of the sandwich segment. Additionally, because of the sandwich's regular structure, quantitative measures of size can be inferred by counting slices of meat, pickles, etc. Additionally, because the overall area is fixed, percentages can be inferred from individual slices, without the need for the whole sandwich to compare against (much like there is an inherent maximum for total angle in a pie chart). In the Submarine Sandwich Chart, the task of comparing two segments is much easier: Image:GiantSub.jpg

When constrained by the limitations of print media or area, Submarine Sandwich Charts could be reduced in complexity to simple colored bars. Such an approach combines the advantages of low resolution legibility as well as familiar food metaphors.

[edit] Ken-ichi - Nov 07, 2007 08:37:46 am

I thought Few did a reasonable job depicting how the field of infovis has been perceived by the blogosphere and to a lesser extent by the business community, but I would have appreciated more examples or anecdotes of encounters with people who aren't already into vis. How visually literate are decision-makers at the world's most powerful institutions? How does one measure that, and has that degree of literacy changed over time? If the world at large is still enamored of pie charts and dazzled by chart bling, how should we combat these things?

I was also somewhat confused by his list of "visulati". For the most part, these people seemsed to have reasonable but unremarkable visualizations, but with real talent for performance and rhetoric. Al Gore's line graph of CO2 and temperature is a simple, effective, but prosaic visualization. What makes it special is the way he presents it, writ large on a screen, with dramatic pauses. Same goes for the GapMinder presentation. It's really cool, but I don't think I'd find it nearly as informative if it wasn't presented by a wildly gesticulated man narrating the story of the data.

[edit] James O'Shea - Nov 07, 2007 10:07:46 am

I think one of Stephen Few's points was that a gap exists between infovis research and what is available to infovis consumers in the general public. Although this gap may be particularly wide in the domain of visualization, I think this separation exists to some extent for a great many research fields. Before starting grad school, I worked in a research lab in a hospital. There was a vibrant research community at this hospital, and the majority of the doctors were very interested in adopting new technologies and engaging in the latest research. Yet despite this wealth of science talent, I frequently heard physicians complain that they rarely could take advantage of the research because it was too difficult for a medical practitioner to use. This was usually because developing a tool for a doctor to use would be 1) time-consuming, and 2) not particularly interesting to a research scientist. More often, new technology and software was developed for specialized systems, the results were published, and then the scientists moved on to new projects. In the medical field, the buzzword invoked to fill this void was "translational" research, referring to the process of taking basic science research and applying it to specific clinical applications. To some extent, it sounds like infovis might need some translational work as well.

[edit] McD - Nov 10, 2007 08:04:34 pm

I think there is much room in the middle between aesthetically minded designers and statisticians, and Few fills the space well. I gather from the talk and his writings that he is passionate about being shown the numbers, and respect his work at fulfilling this real-world need. I think, however, that his own passion heavily colored his review of Beautiful Evidence. In my view, Tufte's books have progressed from a quantitative to a more aesthetic approach to visualization, and I think Few's expectations based on the word Evidence appearing in the title were misplaced. I have not read Beautiful Evidence, but in reading the introductions and various excerpts (Power Point, Sparklines, etc) that were published in advance I got the impression that this was a softer, less data-driven book. Few's review ignores what I think was Tufte's expressed intent, and is colored instead by Few's own convictions about data presentation. "Evidence" does not necessarily mean quantitative data, and Few's review ignores this.

(I will add, though, that I agree on the repetition point. Each of Tufte's first three books is peppered with many of the same examples, a feature I too find frustrating.)

[edit] James Andrews - Nov 12, 2007 08:13:38 pm

The pie chart paper dismisses comparisons of summed parts out of hand, which is odd to me -- I think most of the time when I've seen pie charts, comparisons between summed parts have been interesting or important. Charts of allocation (budget or disk space, for example) often have categories that the user will only sometimes want to lump together. I think it's even fairly common to see a bit of tree map representation in pie charts, where larger regions are marked with (for example) the same color, but then subdivided to show a more detailed breakdown.

[edit] Amanda Alvarez - Nov 13, 2007 12:19:19 pm

In reading the comments on Smashing's article (Data Visualization: Modern Approaches), one finds posts that express genuine surprise that such a field (information visualization) exists and is so developed, and that there are effective and cool methods beyond those the public is usually familiar with (pies and Excel's offerings). I think the narrow/uninformed view might be reinforced by a 'show me the numbers' mentality; a lot of people are number-phobic, and even a picture illustrating numbers can turn them off. I think Tufte offers an antidote to this by trying to appeal to the aesthetic sensibility. The goal of the book title, I assume, was to convey the thought that even hard evidence (numbers) can be made beautiful; indeed, it should be made beautiful, because otherwise it will fail to reach many people who do not want to confront the evidence in its native form. The goal of Few's book title (Show me the numbers) is obviously to temper the aesthetic approach so that it does not descend into chart bling. The question is, are so many horrendous charts created because of the tools on offer, or because people lack a well-developed sense of aesthetic sublety?

[edit] Christopher Volz - Nov 13, 2007 11:07:01 pm

I have to admit, I have for a long time harbored a certain ill will towards pie charts but was never able to articulate it until now. The problem has always been that comparing ratios of slices of a pie is really really hard. Per Hazel's comment I would say this, a chart of nested squares or rectangles would be vastly easier to compare ratios to than a pie chart. It seems to me that the eye can make much better estimates of the relative size of squares than angles in a circle (which the reading seemed to support)

I will defend pie charts, however, if the ratios you're comparing aren't necessarily important (that is, their exact values matter less) than their relative size. But this defense quickly falls apart if you're comparing more than a few values.



[add comment]
Personal tools