The Purpose of Visualization
From CS294-10 Visualization Sp11
Lecture on Jan 19, 2011
- Chapter 1: Information Visualization, In Readings in Information Visualization. Card, et al. (pdf)
- Decision to launch the Challenger, In Visual Explanations. Tufte. (pdf)
- See also a critique of Tufte's argument.
- Graphs in Statistical Analysis. F. J. Anscombe. The American Statistician, Vol. 27, No. 1 (Feb., 1973), pp. 17-21 (jstor)
Maneesh Agrawala - Jan 08, 2011 03:40:07 pm
What do you think of Card et al.'s distinction between "scientific visualization" that deals with physical scientific data (e.g., air flow over an airplane wing) and often has a natural spatial mapping versus "information visualization" that deals with abstract data (e.g., stock prices or an online social graph) and usually requires designers to assign the spatial mapping? However, Tamara Munzner, has pointed out that "information visualization is not unscientific, nor is scientific visualization uninformative".
To what extent do you agree or disagree with this distinction? Do you see it as helpful or hurtful to the study of visualization?
Card et al. also mention Larkin and Simon's study of people solving physics problems with and without the use of diagrams. If you are interested, I highly recommend reading their paper Why a Diagram is (Sometimes) Worth Ten Thousand Words.
Maneesh Agrawala - Jan 19, 2011 04:17:27 pm
A new comment here.
Julian Limon - Jan 20, 2011 11:30:51 pm
I think I agree with Munzner's view. Although the distinction between "scientific visualization" and "information visualization" might be helpful to tell apart conference papers and academic departments, I believe both are ends of the same discipline. One can imagine that the way in which we represent air flow over an airplane wing are also an abstraction of a more comprehensive reality. Characteristics like color, space, focus, and detail apply to both disciplines. Thus, I believe that the distinction between the two is just about the application and not about the theory behind it.
Sally Ahn - Jan 21, 2011 01:56:59 am
I think Munzner's statement may have been intended to clarify the terminology of Card et al.'s "scientific visualization" and "information visualization," and not necessarily to oppose this distinction in the field of visualization. I think Munzner is pointing out that "information" in "information visualization" refers to the abstract nature of the raw data and not implying that the visualization itself is somehow more informative. This distinction shows that information visualization may require additional work in finding the appropriate visual mapping for non-physical data, but I think it may be hurtful to the study of visualization to emphasize the spatial mapping of scientific data with its physical properties, because such a mapping may not be the best form of visualization. As quoted in the reading, "The purpose of visualization is insight, not pictures." Scientific visualization may have a more obvious mapping to a particular "picture", but the challenge of finding the picture that best facilitates cognition may be just as difficult as it is for information visualization.
Michael Hsueh - Jan 21, 2011 03:47:07 pm
The distinction should be seen as a tool rather than an absolute rule. There are useful visualizations that do not fall cleanly into either category. Take for example a map that superimposes a color gradient and colors regions based on the most frequently occurring type of crime in the respective areas. Another example is a digital model of a vehicle that colors each part depending on its aggregate failure rates over the past decade. Though such visualizations may tend towards one of the two classifications, they critically depend on simultaneously incorporating both tangible, geospatial information and more abstract data. The results are visualizations that are far more useful than if solely limited to either type of data. Card's text does not insist on contradicting Munzner's statement regarding scientific and information visualization. His distinction is not necessarily harmful if it is taken as a tool to reason about what can be achieved using either type of visualization (as he does in the text) rather than to rigidly classify any given visualization.
Side note: I just stumbled upon this. For fans of TED talks, just posted yesterday: Visualizing the medical data explosion
Matthew Can - Jan 22, 2011 03:03:23 am
I agree with Michael's analysis that a taxonomy of visualization is just a tool. It can help us guide our thoughts on the subject, but we should not let it prematurely constrict them.
On another note, this reading discusses techniques to increase the information density of a visualization. Should it always be a goal of a visualization to present the viewer with as much information as is reasonably possible, without degrading the quality of the visualization? Consider Facebook, for example, which only shows users a small but relevant subset of news items in the news feed. This is one way to deal with the problem of massive amounts of data. Think about how this compares to an interactive visualization that lets the user sift through orders of magnitude more news data.
Saung Li - Jan 22, 2011 06:58:07 am
I partially agree that there is a distinction between scientific visualization of physical/spatial data and information visualization, which tends to deal more with the abstract. However, there is probably a lot of physical, scientific data that has no obvious spatial mapping. I do not think such distinction actually helps with people's understanding of data or observations, whether scientific or not. I understand that there are economic boom and bust cycles by looking at a graph just as much as I understand that there are electrons swirling around an atom by looking at a picture. Since atoms cannot be seen, they feel as abstract as economic cycles until I see visualizations of them. The point is that data is converted to something we can visualize and understand better, so I do not think such a distinction is necessary.
Card et al mentions a lot about using "computer-supported" visualization, but we do not always have to use computers to help us visualize things. Using paper and pencil to do multiplication, for example, is not computer-supported, yet it represents an important visualization process that helps us understand and perform math more efficiently.
Siamak Faridani - Jan 23, 2011 03:58:56 pm
I disagree that there is should be a distinction between "Information Visualization" and "Scientific Visualization". As many scientific visualizations are about showing abstract concepts. Take the very example that was provided in the text (flow visualization). Flow visualization, which I assume is the flow velocity visualization, is itself the visualization of a computed, average, velocity field around objects. It does not mean that there is an underlying movement similar to what has been visualized. It is the interpretation, that abstracts most of the underlying molecular movements and shows the average of their movement. Another example would be visualizion of stress values inside rigid objects. Stress itself is a tensor (has direction, value and a normal surface on which the force is acting) the mapping from visualization to the actual underlying phenomena is not as direct as we think (the stress values that are typically being visualized are von-misses stresses that are calculated from the actual stress values but can help engineers study failure modes easily). I personally do not see a difference between these examples and visualizing netflix movies, facebook network or tweets. What we see are just shadows of the actual information.
Manas Mittal - Jan 23, 2011 07:24:13 pm
I think Card et al. distinction between scientific and information visualization is a useful heuristic that a designer can use when designing a visualization. I don't quite see the value in formalizing this distinction though.
Sometimes scientific data may not be easily mapped (eg. rate of a reaction), and sometimes doing a simplistic mapping adds to the confusion. For example, consider the case of atomic orbitals (high school chemistry). We instinctively think that atomic orbitals "ought" to have a natural physical mapping. However, they don't and what we often see in diagrams is the mapping of the ψ function which is unrelated to anything physical (ψ squared is indicative of the electron probability but does not look like the pretty ψ shapes). Orbital Image URL:http://upload.wikimedia.org/wikipedia/commons/2/2d/Neon_orbitals.JPG
Karl He - Jan 24, 2011 12:06:54 am
I believe the distinction between scientific and information visualization is unnecessary. Scientific visualization is simply a subset of information visualization that happens to have a pre-existing structure of some sorts. This makes it easier to design visualizations for them.
For example, to show the elevation in a geographical area, a common visualization method is simply to take the map representation of the area and apply a color overlay to convey the information. This is simply using a property of the data to create a base, so that it is easier to understand. Some more abstract data such as web traffic could be mapped over time, and ad clickthrough could be mapped over position on the page, for example. The strive to visualize data better does not depend on what the data is.
Sergeyk - Jan 24, 2011 04:24:38 pm
The distinction between scientific visualization and information visualization seems false to me. It is motivated by the fact that science is performed with measurements of the world; as such, these measurements should be re-mappable onto the world. But there are many cases where scientific data does not easily conform to a a natural spatial mapping. For example, dendritic connections between neurons would be hopelessly complex if presented as they are measured. To lower the entropy of this information, we need to present it in a simplified, and spatially incorrect way--to assign our own spatial mapping. Coming at the distinction from another direction, we can consider the data collected in social sciences--for example, reaction times ito different visual stimuli with an additional variable of some factor of the participant's environment. The multivariate experiment is not spatial or temporal in nature: the point is in the relationship between experimental conditions. Census data, on the other hand, could very informatively be mapped onto a spatial map.
The point is that everything is information, even scientific measurements, and more often than not, presentation of information is aided by re-mapping it into a different projection, even if it does not agree with the original spatial layout of the measurements performed. The distinction seems hurtful to the study of visualization, as it may serve to close some avenues of thought.