The Purpose of Visualization

From CS294-10 Visualization Fa07

Jump to: navigation, search

Lecture on Aug 27, 2007

Slides

Contents

[edit] Readings

  • Chapter 1: Graphical Excellence, In The Visual Display of Quantitative Information. Tufte.
  • Chapter 1: Information Visualization, In Readings in Information Visualization. Card, et al. (pdf)
  • Decision to launch the Challenger, In Visual Explanations. Tufte. (pdf)

[edit] Maneesh Agrawala - Aug 28, 2007 12:50:19 am

Tufte explains how the decision to launch the Challenger could have been averted, had the Morton Thiokol scientists done a better job of presenting the information about the impact of temperature on O-ring damage.

More recently Tufte has analyzed the Columbia disaster. He found that once again a few scientists were aware that the shuttle may have been damaged during liftoff, but were unable to present their information to the key decision-makers in a way that would affect their decision.

[edit] Kueda - Aug 28, 2007 09:48:33 pm

Regarding Card et al.'s immense taxonomy of data, information, and visualization, how frequently are notions like "knowledge crystallization" and techniques like "Cost-of-Knowledge Characteristic Functions" actually used in designing and assessing visualizations? This chapter seemed sort of like it was over-modeling the domain a bit. Like, why differentiate between "scientific visualization" of physically based data and "information visualization" of non-physical data? Surely scientists deal with abstract data as well.

[edit] David Jacobs - Aug 29, 2007 12:47:22 am

I find it interesting that in his analysis of the Challenger report, Tufte suggests that removing information can actually improve an argument by better focusing the viewer's attention -- here specifically the data about nozzle blow-by (in the first page or two of the section). This leads me to ask if it's possible whether selecting which information to present (or how to present it) can misrepresent the data's meaning, even if accurate.

[edit] Ariel Rokem - Aug 29, 2007 09:45:00 am

Card et al. seem to be re-doing two mistakes made before wrt to the Challenger. First, they claim that the Thiokol engineers were not able to predict that there might be a temperature-related problem with the O-rings. This is clearly not the case. The problem is that their 13 charts could not convince the people actually making the decisions. Second, they show one of the rocket diagrams as though that were the diagram used in making the decision. This is also (according to Tufte) not the case. The diagrams were concocted post-hoc, in order to justify the decision made. One principle that comes out of Tufte's analysis of this incident, in light of Steven's description of scales of measurement, is that you want to use the scale that allows you the most operations. So - if you can choose a nominal (according to date) or a ratio scale (according to temperatures), that allows you to intuit the functional relation between temperature and incidence of problems, you should choose the ratio scale. Notice that Tufte *makes up* the ratio scale. His damage index, which is the variable that allows you to organize the data in a 2D scatter plot (in contrast to just counting the number of incidents per temperature), is the hard nut to crack. That is - what exactly should the value of the damage index be for any incidence of this or that kind of damage? And hind-sight is 20/20.

[edit] Robin Held - Aug 29, 2007 10:13:25 am

I completely agree with Tufte's claim that the figure containing the outlines of the booster rockets was distracting and inefficient. The first thing that grabbed my attention was the detail put into the illustration of the rockets, because I figured they might be trying to convey some information about their failure based on their construction. However, the original author chose to essentially use symbols to indicate the type of o-ring failure on the rockets. It was completely unnecessary to include diagrams of each rocket. A table or chart would have much more effectively conveyed the pertinent information without distracting the observer and wasting his/her time.

[edit] Teale Fristoe - Aug 29, 2007 06:52:11 pm

I agree with Robin. Even before Tufte commented on them, I couldn't understand why the authors included the rocket illustrations: they contribute no information, take up a lot of space, and confuse the actual important information. But after thinking about it, I can speculate on the psychology behind including them, and how I would make a similar mistake. As a writer of interfaces, I know that it can be painful to part with something you put a lot of work into just because it isn't ideal. This is especially true when it's so much easier to use existing material than it is to come up with something from scratch for a very particular purpose. I think that will prove to be the most difficult aspect of this course for me. It's easy enough to recognize how successfully a visualization portrays certain information, but coming up with general visualizations that are effective in a wide range of specific cases is quite a challenge indeed, and I hope that some of the course focuses on tackling this problem.

[edit] Secret name name - Aug 29, 2007 08:30:50 pm

It strikes me that Card's "knowledge crystallization" might have psychological implications. Consider social networking applications; the user inputs and maintains detailed information about their self, activities, and desires within a massive relational database. This process might very well be called "ego crystallization." As a crystallized data point, a user's self-representation may be quantitatively compared to others'. In a typical social network a datapoint functions as a node in a graph. A user, by mentally or computationally running a query across different parameters (favorite bands, books) or, more subtly, by observing trends (memes, graphical/textual eccentricities) of his node's neighbors (friends) engages in a shaping of her ego. These computational tools allow users extreme clarity and precision as they assess their ego's relationship to those of their neighbors. This is a continuous process - the ego must be honed in the face of shifting trends, RL events, and, well, just cuz you wouldn't want to get too dated. It seems to me social networks' real power lies not in their potential to "connect" people but rather in their role as a tool for the crystallization of the self.

[edit] James O'Shea - Aug 31, 2007 10:43:17 am

I think David Jacobs brings up a good point above, mentioning how sometimes removing information can help improve an argument or clarify things. In the case of the challenger disaster, I think the opposite idea applies as well: the chart-makers also did not provide enough information. Although omitting the blow-by evidence may have helped elucidate the relationship between temperature and erosion (by focusing the observers attention), Tufte also notes that the chart-makers only included evidence from 2 of the previous 24 shuttle launches. The link between low temperatures and O-ring damage was much easier to see once data from all of the previous shuttle launches were included in a single graphic, as Tufte shows in his Post-Challenger Evaluation data matrix on pages 44 and 45. This case illustrates that it may be important to omit unnecessary information when presenting an argument, but even this focus of attention may fail to reveal the truth without the proper context.

[edit] Amanda Alvarez - Sep 02, 2007 07:01:10 pm

In the slides we saw how more and more information is being generated every year; there was also a quote from Simon about how information consumes attention. Above, David and Jamie mentioned the impact of information omission or addition, and providing appropriate context for that information. I think that part of this context is created by the recipient's (viewer's) characteristics. In light of this, and thinking back to the title of this class, I think the purpose of visualization will increasingly be to act as an attentional tuner.

If visualization acts as an attentional tuner, we need to take into account not only the careful selection of the information to be presented, but also the following: the recipient's goal states, prior knowledge, preferred format for accessing information, etc. Given the increasing amount of information to be sifted through, visualization should possibly be tailored to individual viewers, to maximize their information extraction. (This tailoring should be done so as to strike a balance between removal (which focuses attention) and addition of information (which dilutes attention); moreover, this 'tailoring' should not be used to reinforce preconceived notions about the data.)

'Knowledge crystallization' seems to be a manifestation of a mental attentional shift. The ways in which visualization amplifies cognition (listed in Table 1.3 in Card et al.) can all be construed as attentional shifts. Card et al. define visualization as using perception to enhance cognition. This process must then necessarily take account of the recipient's cognitive state.

[edit] Hazel Onsrud - Sep 03, 2007 09:09:30 am

“It should be noted that while we are emphasizing visualization, the general case is for perceptualization. It is just as possible to design systems for information sonification or tactilization of data as for multiple perceptualizations. Indeed, there are advantages in doing so. But vision, the sense with by far the largest bandwidth, is the obvious place to start, and it would take us too far afield to cover all the senses here.” Quoted from Pg 7, Card et. al, Readings in Information Visualization

Browsing the web for various information about our many senses, their trickery and potential, I came across a BBC website which offers a fun quiz and many explanations for how misleading our visual sense can be (among others). http://www.bbc.co.uk/science/humanbody/body/interactives/senseschallenge/ The complications that could result from such perceptions solidified my conclusion that it would be best to enhance our visualizations with attempts at sonification or tactilization etc…making use of as many of our senses as possible in order to best convey the desired perception. I would be interested in “seeing” further explorations of these types of combined techniques for the vast majority of society, rather than basing further advances perceptualization on our visual sense alone. At present, brail is adapted from our previous text, and music is often read, yet what if we as a society (most likely lead by those who have the strongest senses in these areas) could change the way we predominantly “see” the world to how we “hear”, “taste,” “smell,” “feel” or “sixth sense” it. It seems to me that poets, musicians, artists etc. often seem to try and capture our reality’s essence in a variety of manners…but most seem to be connected to visual representations….or perhaps that is simply my biased (extremely visual) viewpoint.

Although Card et. al ‘s logic on pg 16 concerning how information visualization enhances cognition makes sense to me, in general, I wonder if some would challenge their (Card et al.) assumption that vision has the largest bandwidth. Although we are all trained, through our educational system and general society to respond to various visual systems/visualizations, perhaps we are under utilizing the rest of our senses, some with great potential. Science fiction writers have certainly explored this topic…I would like to know what psychologists, biologists, and others with knowledge in this area may think.

[edit] Mark Howison - Sep 03, 2007 09:26:37 am

Although Card et. al. mention Norman's work on artifacts and Scaife and Rogers' external cognition, they fail to mention an equally relevant branch of cognitive psychology called distributed cognition. Like external cognition, distributed cognition focuses on relations between individual cognition and the external world, with emphasis on how people coordinate internal and external representations. For instance, in an article entitled "How a Cockpit Remembers its Speed," Edwin Hutchins analyzes how the activity of preparing a commercial airplane for landing requires cognition about the planes wing configurations that is distributed among the flight crew, various instruments available to them, and, most importantly, a binder full of index cards that list minimum maneuvering speeds given different positions of the wings' flaps. The coordination among representations in this scenario serves both to offload the need for pilots to memorize the content of the index cards, but also to create redundancy through multiple instruments and verbal protocols between pilot and copilot. Distributed cognition is relevant for information visualization, which never takes place in a vacuum, but instead in the context of some activity requiring coordination between individuals' cognition and an external artifact, namely the visualization. For instance, Thiokol engineers mis-coordinated their knowledge of how cold temperatures affected the O-rings with the 13 charts that all failed to highlight this dimension of their analysis. (Card et. al. do however reference Hutchins' book Cognition in the Wild, which contains an extended analysis on, among other things, the use of nautical charts in Navy navigation protocols.)

The example of mental vs. paper-mediated arithmetic in the Card et. al. article illustrates the use of external representations to offload computation. However, there is an even better arithmetic-related example of "information visualization" in what Stigler identified as the "mental abacus." Basically, he found that Chinese students use of abaci in learning how to perform arithmetic results in different mental representations for arithmetic that allow for faster mental calculation that those of American adults. Moreover, qualitative data collected from the Chinese students suggests that they mentally visualize an abacus when performing mental arithmetic. This study provides an example of how visualizations can be appropriated as internal mental representations that provide for rich imagistic reasoning. In Card et. al.'s language, the abacus has amplified the students' cognition of arithmetic, or in Norman's language, the abacus provides an external aid that enhances cognitive ability.

Here are the full references to the authors mentioned above, if you are interested in reading more:

  • Hutchins, E. (1995). How a cockpit remembers its speeds. Cognitive Science, 19, 265-288.
  • Stigler, J.W. (1984). "Mental abacus": The effect of abacus training on Chinese children's mental calculation. Cognitive Psychology, 16(2), 145-176.

[edit] Robert Carroll - Sep 04, 2007 12:11:24 am

Although it seems rather intuitive, Card et. al. don't really quantify their claim that the visual system has a greater bandwidth then the other senses. I came across this article <[1]> which tries to estimate the bandwidth used by 3 of the senses, figuring taste and smell are obviously very low, in order to put a theoretical upper limit on the computer network requirements of a single person. By the most conservative estimates presented, the visual system can process 2 orders of magnitude more data than the next most demanding sense. With that in mind is seems clear that visual presentation of information has by far the most potential, at least in terms of the sheer quantity.

[edit] James Andrews - Sep 04, 2007 06:00:19 am

Robert -- I like the idea of having a rigorous, scientific definition of information bandwidth, but is it really meaningful to think of 'visual bandwidth' in terms of amount of data we would possibly want to send at an eye? Although in some sense we are 'processing' all that data, we certainly do not 'fully' process an image at a rate of 24 frames per second; in fact we can spend hours considering and understanding a single complex, interesting image. So instead of thinking about the time needed to minimally perceive something, I think we need to think about the time needed to achieve some higher level of understanding. And in that context, it really doesn't make sense to compare 'visual bandwidth' to, for example, 'ear bandwidth,' without specifying "for what?" and "to what audience?" since that bandwidth depends on both. For example, if you want to convey musical concepts to a novice musician, it might be much faster to do so with the help of a few sound files. But to an expert musician, it might be faster to simply present the notes visually.

[edit] Kristal Sauer - Sep 05, 2007 08:59:47 pm

The idea that visualizations can actually be used to compensate for our own mental weaknesses is an exciting one. For me, this totally expanded my idea of the realm in which visualizations are useful. Before I read the "Using Vision to Think" excerpt, I considered visualizations to be an incredibly important and effective method for facilitating communication and comprehension of ideas. But, this paper suggests that they can be so much more than that -- they can be, in fact, a set of TOOL to help people overcome their own limitations. I personally have trouble really digesting information that I only hear; it helps a great deal to see the information too. A tool that would employ natural language processing to translate speech to a visual form would really help me in lectures or when listening to radio programs, for example. Imagine a device that could somehow translate words to a graphical representation (ie, draw a map given a person's verbal directions). This more active role of visualizations could be very beneficial in myriad applications.

[edit] Charlotte Wickham - Sep 10, 2007 08:36:47 am

Card et al. acknowledge that visualization can be an essential component in all parts of the "knowledge crystallization" process, from data collection, through data investigation and to the presentation of results. I find Tufte tends to focus only the final result. The engineers working on the Challenger knew the message they wanted to communicate "the colder the temperature the more likely the o-rings would fail". Tufte then focusses on how this message and the data that supports it would have been best communicated visually and I agree with his conclusions. However, I know from experience it is finding this message (or an interesting feature of data) that is the hard part for the investigator. Visualization is often of great help here but it is more of a process than a result - each visualization suggesting a new subset of the data, a new view or a new hypothesis. How do Tufte's principles apply to this more procedural view of visualization? All visualizations in the process need to be clear and true to the data but it is just as important that they are easy to create and change to allow rapid exploration. Card et al. provide the foundation for this type of process. A change at any level of their reference model (Figure 1.23 in their paper) provides a new visualization.

[edit] N8agrin - Sep 16, 2007 10:39:36 am

I appreciate Tufte's take on graphical excellence, though his examples remind me of one unspoken aspect of Information Visualization, essentially the need to understand the context of the information in the image before the underlying message of the image can be revealed. Tufte's example of the train schedule from Paris to Lyon would simply remain a confusing set of lines if not for the implicit understanding that what each town on the Y-axis represents is a stop along a train's particular route.

Card et al. discuss the interesting concepts of 'external cognition', the concept that humans have the ability to augment their cognitive processes by using external tools, like pen and paper, and that visualizations get their power from this ability. Their examples provide a clear insight into the power of visualizations and explore some basic concepts, like the ability for a visualization to reveal patterns in complex data, shed light on otherwise surprising and hidden results or simply aid in monitoring large quantities of data. I'm particularly drawn to Card's notion of 'Information Chromatography', which to me seems to mean the separation of information into its data components using visualization. Finally they attempt to synthesize some basic premises of data that allow it to be visualized. This includes the mapping of data to Nominal, Ordinal, and Quantitative, each of which lends itself to a certain types of visualizations.

Tufte's take on the Challenger disaster and his redesign of the data regarding O-ring damage in cold weather is particularly startling. I think in this case, he shows how simple visualizations can be more powerful than complex diagrams or drawings, and as others note, it's hard to imagine that, when faced with this sort of powerful message, the Challenger would have been launched.



[add comment]
Personal tools