Discussion of Bad Visualization Redesigns
From CS294-10 Visualization Fa08
Lecture on Sep 24, 2008
Readings
Same readings as previous lecture. Finish reading whatever you missed from last time. You are still responsible for a comment for today even though the readings are the same.
- Visual information seeking: Tight coupling of dynamic query filters with starfield displays, Ahlberg & Shneiderman. (html)
- Visual exploration of time-series data, Hochheiser & Schneiderman. (html) (pdf)
- Postmortem of an example, Bertin (pdf)
- The visual design and control of the trellis display. Becker, Cleveland and Shyu. (ps)
Optional Readings
- Table lens, Rao and Card, (acm)
- Human guided search: Survey and recent results, Klau, Lesch, Marks & Mitzenmacher. (pdf)
- Design and evaluation of incremental data structures and algorithms for dynamic query interfaces. Tanin, Beigel & Schneiderman (citeseer)
Demonstrations
Ketrina Yim - Sep 24, 2008 09:07:22 pm
I found it interesting that Nick found the Berkeley campus maps to be a bad visualization. After all the points he brought up about how the map was ineffective at giving directions, I'm inclined to agree that the maps have room for improvement. Though I never thought about that before. As an undergrad at Cal, I saw the maps all the time. Sometimes when I had classes in a previously unvisited building, I had to use the maps, too, and the only thing that really bothered me was the lack of a "You Are Here" mark. It just goes to show that a visualization's problems can go unnoticed to someone who has been exposed to it for a long time (e.g. the creator) and fresh eyes are often needed to bring about improvements.
Maxwell Pretzlav - Sep 25, 2008 10:54:27 pm
I particularly liked Scott Murray's redesign of the encryption diagram. The use of different key designs and showing how some keys transfer over the network and others don't seemed to me to very successfully demonstrate a fairly complex process. Keeping the symbols and colors used simple, and the text short and to the point helped drive the explanation home. If anything, I think the different key designs could be emphasized more to make sure the viewer notices the visual cue (and adding a + sign to the equations, as mentioned in class, wouldn't detract from the design either).
Razvan Carbunescu - Sep 28, 2008 10:12:42 pm
As I was looking over the redesigns I still couldn't help but think that for the visualization by Sarah Van Wart about the army strengths of US, China, Russia a log scale for the 2 charts which she expands would be appropriate. While the scale might be hard to read at first it would allow for all the data to be present on one chart and not make the reader look at 2 separate scales. I do like the choice of redundantly encoding the size of the population/reserves as numbers on top of the charts though.
Dmason - Sep 29, 2008 02:02:33 pm
I was particularly interested in Ketrina's redesign of the biological diagram. There was much discussion on the redundancy of the dorsal and ventral views of the Caudal Ramus, which I think connects to other discussions about encoding information redundantly, for instance encoding one datum as the size and color of points in a color plot. It is important to identify to the audience what the redundancy is, otherwise they will assume there is some form of correlation and this will distract from absorbing the rest of the visualization's material. In the case of Ketrina's visualization, I think she handled this extremely well by re-iterating the "Ventral/Dorsal View" descriptor at the top of both panels. By isolating the small appendage and reinforcing the separation with an additional scale bar, these possible confusions are deftly avoided.
Sarah Van Wart - Sep 29, 2008 02:31:50 pm
Kudos to everyone -- I really enjoyed seeing the variety of subjects that were chosen for redesigns. It was interesting to see and hear how each person though about his / her re-design as well as the reactions of the rest of the class. I was especially interested in the re-designs that involved teaching / explaining processes (Calvin and Scott). Finding the balance among graphics, text, arrows, colors, etc. seems really difficult to do, and I thought you all did really nice jobs.
In response to Razvan: Hmmm. You know, I don't think I've really ever seen charts and graphs displayed in terms of a log scale in non-science / math related contexts, but perhaps you are right. I'm wondering, though -- lots of non-technical people read the Wall Street Journal -- might a log scale effect the casual reader's ability to understand the concepts? Perhaps this is an audience-specific question. Not really sure...just thinking aloud.
And just as an aside: a few interesting links got funneled through some of the iSchool mailing lists that I thought I'd share (you all may already know about these):
- There's a World Bank API Available here (in case anyone's interested in interacting with any WB indicator data or development photos: http://developer.worldbank.org/
- There's also an interesting project sponsored by IBM that allows you to create visualizations online by feeding in your own data: http://services.alphaworks.ibm.com/manyeyes/home
Seth Horrigan - Sep 29, 2008 03:11:52 pm
The "mobility of the image" emphasized by Bertin is key. As he points out, with the advent of electronics and computer generate graphs or images, this entire process is drastically improved. While before, one had to color each bar by hand, and had to manually code the numeric or other quantitative data into graphical form, now one can simply allow software to handle the task. Rearranging the various items, changing colors, adjusting scales, and grouping like items is not trivial, but now the process is more about the improvement in conveying information and less about the physical act of rearranging and redrawing pieces of paper. Also, computerized systems allow for quicker, easier experimentation. One can simply change the graph type to a line graph, scatterplot, or bar graph. Perhaps all this reduces is the need for expertise that would allow an expert to, a priori, identify the best method of displaying the information, but I would like to imagine that instead it allows novices and experts to compare various methods and choose the best among them.
Tools like SpotFire and Tableau are integral because simply knowing that such experimentation and organization is possible is not the same as actually having the software already built, but since it is built, we can use it. Also, since we have not seen it in class yet, I believe the Many Eyes (http://services.alphaworks.ibm.com/manyeyes/browse/visualizations?sort=rating&order=desc) visualization toolkit is an amazing tool. The only downside is that it cannot be used without releasing your data to the public, so it cannot be used with arbitrary data.
Witton Chou - Sep 30, 2008 01:13:04 am
I enjoyed looking at all the redesigns to the bad visualizations. I was surprised at how many problems exist with all of these published visualizations. It really makes me wonder what kind of editing process these publications go through for such tremendous problems to slip by. With that said, I thought that these redesigns really helped with understanding and focusing on the key aspects of the visualizations.
Bertin's piece realizes the importance of converting raw data into visual indicators. By utilizing the shading of boxes rather than raw numbers and careful grouping of rows, Bertin is able to show trends in the data that would otherwise be difficult to extract. He takes advantage of the human brain's ability to recognize patterns, changes, and trends to show the patterns of occupancy throughout the year. The idea of manipulating and moving data around as needed to convey a point is very important in visualization - not all details of the data are represented in the reconstructions, but the important ideas, which are the basis of the data gathering are present.
Michael So - Sep 30, 2008 04:47:50 pm
The time box idea for exploring time series data is a pretty good idea for interaction. I like how it's simple. All you need to do is drag the mouse to make a triangle, which a lot of computer users (I think) intuitively know how to do.
I also want to comment on the Cellphones visualization. I think it is a good example of dynamic querying where the results of each of your selection updates the display (pretty much) instantly. I also like how each cell phone is depicted by a picture of how the cell phone actually looks. This is better than having each entity being depicted as some dot or rectangle as is done in HomeFinder and FilmFinder respectively. But I guess to be fair, there are significantly less cell phones than movies, so there is more space for the Cellphone visualization to put in detailed images for each entity.
Calvin Ardi - Sep 30, 2008 06:00:33 pm
@Sarah Thanks! It didn't hit me until someone said it in class that what Scott and I were trying to do was reconstruct a process as opposed a data chart/graph. That said, there certainly is room for improvement after going through a "critique" of sorts; for example the red/green arrows should either be different and more distinct colors and perhaps separated a bit. There were a lot of different things going on, but that's somewhat due to the nature of BitTorrent itself, in my opinion.
In general, it's interesting to see all the sorts of visualizations that everyone redesigned. Tufte mentioned along the lines that most of the visual artists that design these sort of things don't necessarily have a background in statistics, so in their effort to make the data look interesting and aesthetically appealing, they end up losing the point of the visualization in the first place.
The websites presented above are some neat demos; the cellphonefinder is pretty neat if you're looking for a phone with certain qualities (price, features that are needed). It would be more useful, I think, if they were to include links by third-party reviewers, and perhaps an aggregation of ratings for the phone (e.g., site A gave 3/5 stars, site B gave a 75%). I also liked zipdecode as it highlights the areas that a particular zipcode prefix covers as you're typing it out. As someone mentioned in class earlier, it would help if there were outlines of states or counties as you're zooming in (or at least a "slider to control the level of zoom); hitting 94704 zooms me to where Berkeley is, but it's a highlighted dot surrounded by other nameless dots.
Scott Murray - Sep 30, 2008 06:32:27 pm
@Maxwell and Sarah: Thanks for the feedback! Maxwell, I agree that the refinements you pointed out (and that were mentioned in class) would improve the design further.
I found it interesting that Calvin and I—who were both trying to illustrate processes, not quantitative data—ended up using many of the same elements (arrows, numbering, explanatory text) to depict sequence and action.
Simon Tan - Oct 01, 2008 03:48:43 am
The GGobi "tours" feature is certainly a novel way at attempting to break the barrier that Bertin had declared "impassible" (from the second lecture) - that is, it apparently can create visualizations that encode an arbitrarily high number of variables. Watching the video of it in action was sort of a confusing experience, however - with all the variables encoded as shapes/colors/etc. and having the entire display "wiggle" around incessantly on its tour, it was more disconcerting than revealing for me. Perhaps it requires getting used to in order to be able to extract useful revelations from watching the "tours".
I found an interesting interactive visualization while using Mint.com, the online money management web application. Mint.com attempts to give you a view of your spending trends via a Flash visualization that I can only describe as a hierarchical pie chart. You select pieces of the pie (categories of spending) to drill down and see the division among sub-categories. For example, here I am trying to find out how exactly I divide my money in Food & Dining:
This works decently well, but it is hard for me to peg into a category of interactive visualization. It's not Brushing/Linking or a Dynamic Querying system (although it does have a Dynamic Query feature: a timeline so you can adjust what time period of spending you see). I would consider it more of a zoomable or fractal visualization, which makes me wonder: Would a treemap be a better way of presenting this information?
Chris - Oct 01, 2008 10:59:41 am
The MERL paper on using a human, aided with an appropriate visualization, to aid a computer in solving a problem interesting, especially in the context of the new hot topic of human computing, with such initiatives as Peek-a-Boom and ReCaptcha.
One weakness of that paper is that they compare the human-aided computer to something of a strawman. The algorithms whihch are being aided are quite primitive (brute force, steepest descent, etc). A lot of research has gone into fast approximation algorithms for NP-hard problems, so using one of those would be a more fair comparison. Additionally, while a human can help the computer exit obviously suboptimal solutions, the human will probably not be well-suited to find an optimal solution, rather, the human will effectively be an approximation algorithm.
Matt Gedigian - Oct 01, 2008 01:16:46 pm
The Tanin article on incremental data structures was good. If we are dealing with the rich datasets that Tufte has tried to instill a love for, it can be challenging to deliver the 0.1 second response times that users demand. One trick they use is to compute a maximum hit set based on the control that is currently being adjusted. The idea is that the user can only be adjusting one control at a time, so we can precompute the range of queries that are possible when all other attributes are held constant.
However, this assumption breaks down if we allow multitouch interaction. In such scenarios, it would be possible for a user to simultaneous adjust multiple attributes. This raises two questions: are such interactions possible? and would they be useful/desirable?
Can people simultaneously adjust controls for multiple attributes? Yes. There are numerous instances of people simultaneously adjusting analog sliders. e.g.:
- adjusting gas pedal and clutch while driving
- mixing audio levels
- teleporting stuff
Is it useful for visualizations? While people probably aren't going to be hitting random combinations of checkboxes at once, but it seems desirable to adjust multiple sliders simultaneously. This seems like it would be useful when there is a tradeoff between different attributes. To use the home finders domain, it would be nice to simultaneously adjust some price and distance ranges since those both affect the value (utility) to the user.
David Poll - Oct 01, 2008 01:55:09 pm
I, too, loved seeing everyone's visualization redesigns. There were certainly some impressive changes that made SO much difference in terms of being able to take in the data and making a point. Lately I've been fixated on the presidential race, and I'm loving a particular site: RealClearPolitics.com. I've become obsessed with watching the polls and trying to explain them, and the interactive visualization on this site is one that I really enjoy. Here's a screenshot:
For the actual interactivity, take a look at the link above :). My problem, however, is that RCP is a site where there is a ton of polling data from various institutions, and there is simply no way to see all of this information together. I'd like to be able to look at the race for a senate seat, for example, alongside the national presidential polls, or the "generic democrat vs. generic republican" data alongside the "Obama vs. McCain" data in order to see more interesting trends over time, such as how Obama is doing versus the expectations based on the generic vote or how a Senate race is affected by the presidential one.
This is a visualization I'm looking at for my Assignment 3, and I'm hoping to snatch up some of the RCP data in order to allow some of this comparison. My graph might not look exactly the same as the one there, but it'll hopefully enable some new analysis.
Yuta Morimoto 19:21, 1 October 2008
In discussion session, there were many so great redesign of visualization. Especially, Nick's works most impressed me. So far, I did not think of the effect of changing color of a line lighter. Just to do it, impression of the entire figure looks better drastically. Initial bad example seems to be used primary color excessively. Intuitively, painting primary color draws attention, which is good to facilitate rapid understanding of the map. But, I think that too much attention maybe lost attention for other objects.
Nicholas Kong - Oct 01, 2008 05:42:55 pm
I agree with the above comments about process visualization; it's definitely a difficult problem, and I think Scott and Calvin did well with their redesigns! I'm wondering, though, if effectively visualizing a process may be data specific. Specifically, I enjoy folding origami. However, diagrams documenting more complicated steps, usually involving more than one fold at a time, can be very difficult to read and end up quite unclear. There's a decision that needs to be made about what creases to show and how best to display the detail/overview of the model, and in the end many diagrams are quite ineffective. I'd be quite interested in attempting to create a better visualization of that and perhaps discovering whether there are some common elements that work across all process diagrams.
Calvin Ardi - Oct 20, 2008 09:10:33 pm
Like the cell phone chooser, I ran across a (future) site that will do this sort of thing with cars. Although still in beta, CarZen aims to help users search for cars based on their specifications. The original article was posted on TechCrunch.
