# Data and Image Models

Lecture on Jan 25, 2010

## Contents

• The eyes have it, Schneiderman. (html)
• The structure of the information visualization design space. Card & Mackinlay. (ieee)(Google Scholar sources)
• Chapter 1: Graphical Excellence, In The Visual Display of Quantitative Information. Tufte.
• Chapter 2: Graphical Integrity, In The Visual Display of Quantitative Information. Tufte.
• Chapter 3: Sources of Graphical Integrity, In The Visual Display of Quantitative Information. Tufte.

• On the theory of scales of measurement. Stevens. (jstor) (duke.edu)

## Jon Barron - Jan 21, 2010 01:14:19 pm

Shneiderman: I found the most interesting part of this paper to be the analysis of the problems and opportunities that come with different dimensionalities of the data being visualized. This was also the least-dated section, as most other topics (overview->details mantra, vocabulary/concepts for interfaces, boolean operators for filtering) seem obvious in 2010, while we still don't have good 3 or 4d visualization methods.

Card & Mackinlay: I appreciate the effort towards creating a formalism for discussing visualizations, but I don't really see the merit for their system. I do appreciate the irony in their practically unreadable charts.

Tufte: Awesome text. My favorite explanation of good visualizations: "Graphical excellence is that which gives to the viewer the greatest number of ideas in the shortest time with the least ink in the smallest space".

I found the results from the study in Chapter 2 on perceived area of the circle very interesting (perceived area = (actual area)^x, x = 0.8 +/- .3). What's curious about this is that we would see nearly exactly this result if half of the population took the radius (area^0.5) as the perceived area, and the other half took the actual area (area^1.0) as the perceived area, which I would find to be a very illuminating and satisfying result, especially since the bulk of this chapter is on the difference between taking volume, area, or length as "size". A histogram of each x from the study would be helpful here (I appreciate the irony).

## Jeffrey Patzer - Jan 21, 2010 03:33:20 pm

Shneiderman: What I found to be most important from this article was Shneiderman's insistence that successful commercial products need to deal with multiple data types. Although it may be "easy" to find the best way to create a visualization of one data type, I think the real challenge comes in trying to find a way to successfully employ multiple types. Discovering and implementing the best interactions between the multiple data types may be the where one needs to put in the most creativity.

Card & Mackinlay: What I found to be the most important point is this article was the concept of Overloading. Since this article was dealing with how one maps data into an XYZ space, the idea of how one properly overloads the XYZ space with multiple datasets is quite intriguing. You have to be aware of occlusions or how one is sensing spatial depth. This relation is extremely important if one is looking to build a 3 or 4D display (something touched on in Shneiderman).

Tufte: I found Tufte analysis of the Napoleon Army size visual to be extremely informative. I have seen that visual many times before, but his analysis brought a much deeper understanding for why the visual works so well. Tufte states that the leveraging of multiple types of dimensions to embed as much information into the visual in an interesting way is what makes the visual so useful. By keeping in mind the different ways that one can embed information into a small space, one can maximize the usefulness of the visual.

## Chetan - Jan 24, 2010 02:53:21 pm

Shneiderman -- It was interesting to see a breakdown of the different interaction elements and how they are used by the user. However in detailing all the different elements of the visual displays, the approach seemed quite ad-hoc in listing the known interaction schemes and commenting on the pros and cons of each rather than a truly first principled approach to interaction in general (and maybe this is too much to ask for).

Card and Mackinlay -- I found their formalism to break down a visualization to be interesting although I'm not clear how it can be effectively used. As of now it just describes a mapping from the variables of the dataset to the dimensions of the visualization.

It would be interesting to see different visualizations for the same dataset and then see how the formalism can explain why one visualization succeeds while another fails.

## Sara Alspaugh - Jan 24, 2010 08:51:19 pm

Shneiderman -- One thing that occurred to me in reading on his seven data types and the examples of each is that it seems that each is presented in such a way that it is often assumed to be more or less static - the user is able to interact with it and change it in that way, but what I mean is that I would have been interested to hear more about how animation or motion/change over time within the visualization can enhance a visualization or a user's understanding of a set of data and also how that would fit into the taxonomy.

Card and Mackinlay -- I did not think their tables were helpful and I actually found them unpleasant to look at and try to process. Kind of ironic for a paper from visualization experts.

Tufte -- I liked chapter 2 best. It seems a lot of his rules for creating visualizations are common sense in a way but as he says, visualization is a powerful tool, and his visual examples of good and bad/misleading visualizations really send his points home.

## Timothy Wheeler - Jan 24, 2010 08:59:36 pm

I find the scheme proposed by Card and Mackinlay (C&M) to be a useful means of organizing and comparing "points" in the "design space". However, the application of this scheme is limited by their coarse parameterization of the design space. The information encoded in the tables is rather abstract (e.g., this visualization represents quantitative data in the X position using colored points), which implies that any “new designs” we discover by looking at the tables will be equally abstract (e.g., “Maybe we should try recoding the data first”). Perhaps, this is where Tufte’s finer analysis of (certain regions of) the design space should be used to eliminate the design possibilities that are ineffective or misleading. Although the tabular scheme of C&M may help explore new types of designs, the examples presented by Tufte (Chapter 2) suggest that practitioners are still struggling to make existing designs work.

## Stephen Chu - Jan 25, 2010 01:01:46 am

Schneiderman

Before this class, I didn't know anything about the study of Visualization. I only knew that it existed. After reading this paper, I am beginning to understand the importance of developing advanced visualization techniques to handle the plethora of data. As we know now, Shneiderman sure was right to emphasize search engines that can handle multiple data types.

Card and Mackinlay

I have never looked at a visualization and actively classified the types of variables. I plan to use the matrix technique in deconstructing future examined visuals. I love Fig. 9: The Mac File System and the similar market visualization shown in class. It really helped me understand more clearly market cap and market share.

Tufte

According to Tufte, one goal of data visualizations is to present many numbers in a small space. Initially, I found this counter intuitive. How do we know when to stop packing information into the visualization? Of course we need to filter the data, but how do we figure out the boundary?

While reading the chapter on Graphical Integrity, I wondered if having a lie factor > 1.05 or < 0.95 ever justified? Is it ever okay to veer from data variation? For example, what if one scientist created an exaggerated visualization to protest the launching of the Challenger?

Chapter 3 made me wonder how we change American culture so that statistics is more celebrated, like it is in Japan. What is it about the Japanese culture and people that make them emphasize statistics and relational visualizations more than we do? To compete with the attractiveness of purely artistic visualizations, Tufte says that we have to pick the right numbers. The right numbers will make the visualization interesting. However, this requires hiring graphic designers who have a statistical and substantive background. How do we push publications such as the New York Times to hire these graphic artists?

## Mason Smith - Jan 24, 2010 10:12:31 pm

Tufte: In chapter 2, Tufte's 3rd tenet of data integrity, 'Show data variation, not design variation,' strikes me as the most fundamental one, at least in the case of incidental (rather than purposeful) lying graphics. In almost all cases, the graphics were improved merely by removing superfluous 'design' components and employing standard techniques (line/bar charts).

Card: 'Set' data strikes me as conspicuously absent from the N-O-Q data classification. I suppose that it could be represented as nominal (with a table of 1's and 0's representing set membership, for instance), but the sets have distinct operations themselves that don't fall under the other three categories.

## Pauloppenheim - Jan 25, 2010 03:27:10 am

Schneiderman - in general reading this was like a blast from the past, and everything about the writing was so certain that this was gonna be the way the world would be. Turns out it's half right, half wrong, and half so everyday common-sense as to be unimportant (i mean of *course* we're gonna zoom in to see detail!). I realize this is 3 halves, sadly.

rest of reading - i haven't gotten my tufte in the mail, and it's too late for the others. Tomorrow I'll tack on more.

## Arpad Kovacs - Jan 25, 2010 11:42:25 am

#### Shneiderman

Some of Shneiderman's principles, such as the Visual Information Seeking Mantra, seem to be so pervasive in visualization today that they appear to be common sense; I am curious how novel these ideas were in 1996 when the paper was published. What I find more remarkable are his proposed ideas that have not yet been widely adopted, such as history, filter-flow boolean expressions, and accomodation of multiple datatypes.

I am surprised that 3-dimensional visualizations are still quite rare today, other than a few specific applications such as scientific/medical imaging and CAD. It seems that people have been thinking about 3D trees, networks, and desktops since over a decade ago, and yet even with our modern hardware and software tools we cannot easily generate 3D visualizations for everyday use.

#### Card & Mackinlay

This attempt to formalize the various categories of visualization by mapping distinctions to variables seems intriguing, but ultimately does not assist in actually evaluating how effective a visualization is. I am curious whether any additional visualization categories have been discovered since this paper was published.

Also, I am not convinced that GIS should be a separate category. It seems to just be a specific application of Informational Landscapes and Spaces, in which geographical coordinates happen to be the variables laid out onto the XY plane.

#### Tufte

I found Tufte's chapter on graphical integrity, and in particular his discussion on visual area and consistency in the number of dimensions to be very valuable, since these misleading area graphics seem to be everywhere today. Also I found it interesting that Japanese publishers tend to print more sophisticated visualizations than their American counterparts; I wonder if the findings of that study are more or less true today.

## Mila Schultz - Jan 25, 2010 11:48:32 am

Card & Mackinlay: C&M's proposed structure of the information visualization design space falls short in a few areas. I think they unnecessarily focus on either numerical/quantitative data. Even in the case of textual data, their examples show simply derived quantitative data. Their distinction between spatial data and geographic data, QX and QXlon biases their examples towards emphasizing the geographic aspect of geographic data. In some cases, the geographic part of the data is not the most important, and it seems that if one were beginning their visualization by looking at C&M's structure, geographic data could be emphasized unnecessarily. Furthermore, C&M do not have distinct metrics for evaluating the success of visualizations in each category. The visualizations are possibly more alike than different despite being in separate categories. That being said, their insight into different types of datasets is good; the problem is that the field of types of data sets and accompanying visualizations is more nuanced than one would guess from simply reading their papaer.

## Aaron Hong - Jan 25, 2010 12:21:10 pm

Shneiderman's article gives an interesting overview of the different things we face in Visualization, from data-types to tasks (seven general ones in each). Also he gives us the Mantra with which to work with: "Overview first, zoom and filter, then details-on-demand." Just having some structure in which to lay the study of visualization is necessary and reading through the different data-types and tasks sparks interest for me in particular areas, 3d being one of them. So far 3d-visualization is a field that still needs much work and especially since we are starting to explore novel interfaces that are multi-touch, 3d-dimensional, or even using haptics.

Also he has a charge at the end. He exhorts us to look into the computer (the cause and solution to our problem) and also to look into integration of these many data-types and tasks. As simple as Google is... it is still pretty much a one dimensional visualization (list with attributes), so I am eager to work on and see what is to come.

## Yotam Mann - Jan 25, 2010 11:30:48 am

The Schneiderman provided a good overview of the many components and interactions we have with visualizations; though, I was not familiar with most of his examples, I was able to think of some examples of my own, like zooming into detail on google maps, or some of the 3-dimentional interactive visualizations that nytimes.com sometimes has. It's interesting that he was able to predict the rise of some of these things.

Card & Mackinlay gave some interesting, and probably useful terminology when dealing with visualizations, especially interactive ones, since most of the terms described finding and sorting data (Q, O, N). I found the example pictures mostly incomprehensible.

## Priyanka Reddy - Jan 25, 2010 12:38:16 pm

I enjoyed reading Tufte's book - it was interesting to see how graphs have evolved over the years. I also agree with Arpad's comment on graphic integrity - after reading that, I see those misleading graphics everywhere.

Also, I really liked Marey's graphical train schedule in Tufte's book on page 31 - it's a very elegant depiction. At first glance, it looks like random lines. But, once you take a closer look at it, the graph is incredibly intuitive, includes all the information you need and makes planning a trip as simple as finding a set of connecting line segments from your departure city to destination city.

I find this graph much more intuitive than the tabular schedule used by Caltrain (http://lh5.ggpht.com/_BxgsIucHMMI/SfNAQybqXvI/AAAAAAAABZo/JxirdMZouzg/Caltrain_Weekend_Holiday_Schedule_03-02-2009.gif), which conveys almost the same information. Exact timings, as seen in the Caltrain schedule, can easily be added to Marey's graph if increased to the same size. Given that, I would say Marey's graph actually can convey more information than the Caltrain schedule since those trains take many different paths while Caltrain stays on one linear path; and the northbound and southbound schedules are combined into one image.

## Jiamin Bai - Jan 25, 2010 12:39:19 pm

"Overview first, zoom and filter, then details-on-demand". Yes, a simple mantra, but the wisdom of it is often underestimated. More often than not, we find visualizations (like powerpoint presentations) crammed with data and information. However, the presentations that I find easiest to understand and follow are those that only present the crux of the matter and leave distraction information aside.

## Lita Cho - Jan 27, 2010 12:51:13 pm

I really enjoyed Tufte's reading this week. I personally had no idea that many people believed that that statistical graphics are dull. On the contrary, I find them the most useful and interesting part of any textbook which enhances my learning along with the text. I usually take more time trying to interpret the statistical graphics then rereading the text. I thought the most interesting part of the reading was how textbooks use more relational graphics without using a time-series or a map compared to newspapers or news magazines. Most of those relationships were key to my understanding of the material. The fact that scientists and statisticians are not the ones creating the graphics for news is really alarming, since that is how most people make opinions of the issues happening today. However, that doesn't really surprise me, since the news has to be pumped out so fast. Quality goes down the drain when speed is the ultimate priority.

## Stephen Jayanathan - Jan 27, 2010 06:01:46 pm

Schneiderman

I found the 'Advanced Filtering' section to be a great suggestion of how visualizations and software might be useful to other types of engineers. Frequently engineers have several properties they need to design around (cost, weight, strength, etc.) and an advanced filtering scheme could make this decision process easier. Similar to the Ashby charts, a dynamic query could help a user rapidly navigate a very large database to choose a couple of suitable options.

Card and Mackinlay

My favorite part of this reading was that the system for classifying different visualizations was very difficult to understand. Unless you remember what each of the twelve columns and 27 different symbols are, this system seems like it might be impossible to use on an everyday basis.

Tufte

Six principles for graphical integrity have been laid out. I'm going to put these on an index card and see how difficult it is to obey all of them when designing my visualization for HW2. My guess is that right now, they are easier said than done. I was also made aware of the difference between the sophistication of the text and graphics in common publications. I have always enjoyed seeing the graphics in the NY Times and been able to understand them, but I couldn't say the same for all of their articles. What do you think--for how many people is the opposite case true? I give the Times my approval to kick it up a notch with their visuals, and try to use more multivariable visuals. Challenge us.

## Akshay Kannan - Jan 28, 2010 08:45:59 am

I found Tufte's analysis of Time-series plots particularly interesting. The huge volumes of information that can be conveyed in a simple graphic, such as the periodic pattern portrayed in the outgoing mail of the US House of Representatives, is truly amazing. Tufte even reinforces this with a side-by-side comparison of the same article. Another visual comparison that really caught my eye was the graph for Nobel prize winning countries on page 60. While it took me a while to realize how these two graphs could differ so dramatically with the same dataset, I noticed the uneven time-scaling used by the graph on the left, producing the significantly lower number shown in the figure. The graph on the right, showing the correct data, caught my eye as well, especially due to the design decision to extend the line beyond the graph itself. While this may seem like an untraditional design decision, one that may even potentially detract from the professional appearance of the graph, I found it to be a very powerful visualization technique. Not only does this emphasize the huge growth in Nobel prize winners in the US, but also in doing so, this graph allows other countries to be easily visible within the graph scale as well, allowing the reader to make comparisons between those countries.

## Boaz Avital - Jan 29, 2010 03:44:06 pm

Card & Mackinlay use their semiotics to map out an array of different visualization styles without commenting on the qualities and strengths of any particular style. However, based on their mention that the styles outlined are the emerging trends in visualization, the assumption is that they are all effective in their own way. I'd like to comment on one particular method, the three-dimensional Information Space. For an information space visualization - like the NYSE example they gave - to be useful, it must almost necessarily be interactive. However, I don't believe there are many things a 3d interactive information space can accomplish that a 2d interactive information landscape cannot, especially for abstract (not scientific) data. This is partially because of Schneiderman's excellent principle for information-seeking: overview, zoom and filter, details on demand. First, a 3d space is very difficult to overview effectively. Without motion, a person only sees in 2 dimensions, so for a complete three dimensional overview already requires interaction or information liable to be hidden from the user. Once the user has located the information, he can zoom in to a filtered view. At this point the desired data is probably displayed two-dimensionally from which the user can demand details. If not, then the view requires yet more interaction and the steps start over.

So, the only phase of a visualization that stands to gain from being a space as opposed to a landscape is the overview. However, even this is questionable. An effective overview of a large dataset must find an intuitive way to group data. It's difficult offer clear groups 3d space where not everything may be visible, and it's likely that any effective grouping found - be it by hue, position, size, or something else - would be as or more useful in a 2d landscape. Therefore for abstract data, where there is no real-world analogue upon which to map in 3 dimensions, I feel that an information landscape is often preferable to an information space.

## Danielle Christianson - Jan 31, 2010 10:08:56 am

Card & MacKinley: I'm not sure what to think about Card & MacKinley's break down of design space. It seems a little over analyzed yet perhaps useful for coding? I would have liked more explanation of the utility. Their division of automated vs. controlled processing got me thinking a bit more about differences between the two, especially concerning hue. While cognitive processes recognize difference in color "automatically," I wonder if the meaning of the hue is controlled processing. This might complicate their schema.

Tufte: Tufte's guiding principles at the end of chapters 1&2 are definitely useful for data graphical design. The last of graphical integrity list "keeping the data in context" ties in nicely with Schneiderman's concepts of keeping the overview in mind while zooming into detail. However in general, I was disappointed in these 3 chapters. Tufte uses the same good examples over and over again. Also, I think Tufte totally missed a major component of the ill-use of statistical graphics in the US -- poor science and math education starting in the 60s and the general disinterest in science (starting about the same time). I suspect (although have no empirical evidence; however, Unscientific America [1] by Chris Mooney and Sheril Kirshenbaum discuss similar issues) that this influences the use of statical graphics during the time period of his empirical investigation of relational graphics in addition to the use today. Additionally, I am skeptical that the general public will suddenly think statistics are not boring if every statistical graphic follows Tufte's advice -- I think the problem is more complex: largely the competition for a viewers' attention given the current bombardment of media. Finally, Tufte's confrontational tone was really irritating, and I found myself constantly trying to removing my initial reaction to his negative writing style from my analysis of his content.

## Jaeyoung Choi - Jan 31, 2010 05:11:37 pm

Tufte

Many ideas introduced in this book is freshly new and inspiring, but what I really like about the book is that it also organizes and give names to things that you already knew with your feeling but not quite organized within your brain. It's just like what Norman said, "It is things that make us smart". About the graphical integrity, I considered myself to be not so naive that I'd never be fooled by distorting graphics. Well, now I'm not so sure, I might have been fooled many time without knowing. This order form could be one example. It's easy to miss that there are shipping charge and some 'extra fees' as these are written in tiny tiny letters, more often hiding somewhere in the corner. One question pops in my mind. Distorting the graphic which represents the data, or manipulating the data itself, which is worse thing to do? Well, I think they are at the same level as both of them share the same purpose: fooling people.

One question - In the graphs showing relation between inflation and unemployment rate on p48, why are dots connected between each consecutive years? If the point was to show that there isn't inverse relationship between the two variables, aren't those lines unnecessary? It proves that they aren't inversely proportoinal, but then, it doesn't show what relation they have between either.

## Ebby Amirebrahimi - Feb 01, 2010 12:51:09 am

Shneiderman: I thought he gave a good overview of visualization. While the mantra may be simple and overlooked, it's amazing how often it's contradicted. So often we see visualizations that are overly complicated and try to present too much information. The mantra applies to both static and dynamic visualizations as well. It specifies a control pattern for dynamic visualizations, where static visualizations would simply be a zoomed-in instance to convey the most relevant part of the data set.

I also like how Shneiderman lists the different types of visualizations, i.e. 1-d, 2-d, 3-d, multidimensional, etc.

Card and Mackinlay: The authors seemed to formalize several concepts in data visualization and did well in presenting examples of the different types of visualizations mentioned by Shneirderman.

Tufte: I enjoyed reading a formal introduction to visualization. They do well in showing its use and defining its objectives. The principles for graphical integrity will be invaluable in creating visualization in the future and I look forward to attempting to implement them.

## Paul Ivanov - Feb 01, 2010 12:56:19 am

Tufte:

I really enjoyed all of the graphs in Chapter 1. Most were clearly well thought out, told a coherent detailed story, except for the 'Phillips curve' plots on page 48. All of them are at different scales both along the abcissa and the ordinate, that's it's hard to make any sort of conclusion from looking at them. I realize that real economic data doesn't always look very nice, but I couldn't help but feel like someone's trying to pull a fast one in using this graph to conclude that the inverse relationship between inflation and unmeployment rates doesn't hold. These plots, to me, don't exactly meet the grade for "graphical excellence."

It was good to read Chapter 2 in detail, as much of the material I previously saw covered by Marty Banks in the annual talk he gives to first year Vision Science students.

Now, maybe it's just my commie roots talking, and call me a pinko if you'd like, but I have beef with conclusions about the lack of graphical sophistication in Pravda. Table 1 in Chapter 3 of Tufte's VDoQI (p. 83) clearly lists Pravda as having by far the smallest number of graphics in the sample - just 54, with almost every other paper in the table having more than twice that number. Here's a relational graphic I made to back me up:

Indeed, in my pursuit of Truth (if you pardon the Russian pun), I found out that something is off with at least one of the percentages. The most graphically sophisticated paper listed, Akahata, is said to have 9.3% of it's statistical graphics meet the criteria of sophistication. But with 202 graphics in the sample, that would mean that 202*(0.093) = 18.786 graphics in the sample met the criteria? Clearly, it should either be 18, or 19, since we're dealing with discrete categories: a graphic either IS relational, or it ISN'T. But 19 bumps us up to ~9.406%, and 18 down to ~8.911%. What gives?

## Prahalika Reddy - Feb 01, 2010 03:28:56 am

In Tufte, it was interesting to see where visualizations started off and to see how much they've evolved since then. The basic principles still seem the same though.

When describing time-space graphics, Tufte provides three examples of graphics that depict many dimensions but are still understandable. While I agree that the graphics shown do achieve their goal of complexity with simplistic designs, a couple of the graphics themselves are not very useful as a whole.

The first one is of Napoleon's march on Russia, that we also saw in class. I definitely agree that this was one of the best visualizations in the book since it manages to convey so many different aspects with such a simple graphic.

The next one is the levels of three air pollutants in southern California. This graphic manages to show the relative levels of the pollutants in a 2D space and is not very cluttered. However, because of this design, it's hard to compare the levels of pollutants in one location over time and among the different types of pollutants, as well as what the level of pollution actually is. In addition, because it's displayed in the multi-dimension format, the images could be distorted.

The last time-space graphic is one of the life cycle of a Japanese beetle. This visualization is interesting in that it shows the physical location of the beetle with relation to its life cycle, but there are other things that aren't as good. Without reading the text surrounding the graphic, I found very hard to understand what this image actually showed. It seems poorly designed and not at all obvious what exactly is happening to the beetle as the months go on. Especially confusing is what happens after July, where one beetle somehow turns into several different ones, and then somehow converges back into one beetle for the remaining months.

Overall, while these graphics show many different dimensions in an uncluttered fashion, it seems to be at the expense of conveying clear and accurate information.

## Kerstin Keller - Feb 01, 2010 10:02:17 am

Tufte:

It was definately very interesting to see how graphics have evolved over time. I still think it is amazing how a big junk of data, if ordered and displayed properly can tell stories with just one glance.

Before reading Tufte, I have never really paid attention to how often area is used to display one dimensional data. I suspect that the designer of those graphics must be well aware of that "rule", however it seems to be tempting to purposely use a second dimension, to make number look larger/smaller than they actually are. I found the Lie Factor to be an interesting concept I haven't thought of before. The Lie factors for the individual graphics often surprised me, they were often higher than what I guessed they would be.

Schneiderman:

The concept "overview first, zoom and filter, details on demand" somehow reminded me of "Google Maps"/"Google Earth". There you get an overview of a town or a route you want to take (position, length of route), you can zoom in to see more clearly, you can filter for restaurants, hotels, ... on the way and retrieve detail for those if you need them. I think this concept is a very powerful concept, which makes it easier to deal with big amount of data.

## Ryangreenberg - Feb 01, 2010 11:37:50 am

One of the themes throughout the Tufte reading that I thought was interesting was how often he argued against visualizations. When discussing time-plot charts he wrote, "Why waste the power of data graphics on simple linear changes?" Later in chapter one he added, "small, noncomparative, highly labeled data sets usually belong in a table." This point is neatly underlined by the frequent appearance of tables in the text.

There is also a fascinating tension between Tufte's perspective, focused on the explanatory and communicative power of data graphics, and that of publishers who exploiting the appeal of visual information. The quoted art director at the NY Times said that "graphics are intended more to lure the reader's attention away form the advertising than to explain the news in any detail" (p80). The fact that data graphics can be powerful analysis tools may be irrelevant to some publications, simply because they view them as a tool to introduce readers to a story, rather than to stand on their own. One gathers from Tufte's criteria for excellent data graphics that they require a far greater investment to produce, both in terms of creation and rigorous data collection, than existing graphs. This said, I have noticed that the NY Times recently has been producing far more data graphics, especially interactive ones in the online edition, which suggests a changed understanding of the purpose of data graphics in publications.

## Shimul - Feb 01, 2010 12:07:21 pm

The visual desgin guidelines from "The eyes have it, Schneiderm." summarize very aptly the basic principles of making visualizations, i.e., overview, zoom-and-filter and details-on-demand. However, it is hard to relate this mantra with static images. The examples it brings to my mind include mainly interactive applets that allow users to zoom in and out and/or fine tune by other measures. A perfect example is google maps which is a great visulization tool and uses the matra effectively. That being said, I would like to see how this rule applies to single and static images.

Tufte, Chapters 1 and 2: There are some images, including "Towards Full Employment and Price Stability" amd "Thermal Conductivity of the Elements" that first look intimidating but are actually readable. It seems that numerical data is often displayed as graphs and as such have an esoteric/scholarly feel to it. Although graphs do seem to be the right way to show numerical/fixed-point data, they have to bear with certain negative stereotypes that go with them. An image that broke this sterotype, while still maintaining the integrity of the data it wants to represent is the "Commission Payments to Travel Agents" (Page 54, Tufte) which uses bar charts coupled with icons on the x-axis, making the diagram look less intimidating overall. Another good example is "Fuel Economy Standards for Autos" (57, Tufte).

Tufte, Chapter 3: I liked E.B. White's comment, "No one can write decently who is distrustful of the reader's intelligence, or whose attitude is patronizing" and its relevance with respect to statistical graphics.

## Jonyen - Feb 07, 2010 10:08:46 pm

Shneiderman: I guess the visual information seeking mantra was the most interesting thing that I got out of it. Couldn't help but think about that as I was skimming through the article. "Overview first, zoom and filter, then details-on-demand". That's pretty much how I processed the whole article, looking at the headings and then reading the details when one of the headings caught my eye. I guess the problem is that this doesn't help to understand how to make content more attractive for reading.

## Zev Winkelman - Feb 10, 2010 12:31:01 pm

The attempt to formally define visualization concepts by Card, Mackinlay and others, is a necessary and helpful step in advancing the science. It also seems to be converging on a few core concepts and classifications such as Nominal, Ordered, and Quantitative data. I was pleased to see a discussion of "Themescapes" because it is a slight departure from the quant heavy data to visualizing large amounts of text data instead. This topic is not covered much in the other readings and I hope we will have a chance to explore it in greater detail.