Visualization Design

From CS294-10 Visualization Sp11

Jump to: navigation, search

Class on Jan 26, 2011

Slides

Contents

Readings

  • Chapter 3: The Power of Representation, In Things That Make Us Smart. Norman. (pdf)
  • Chapter 4: Data-Ink and Graphical Redesign, In The Visual Display of Quantitative Information. Tufte.
  • Chapter 5: Chartjunk, In The Visual Display of Quantitative Information. Tufte.
  • Chapter 6: Data-Ink Maximization and Graphical Design, In The Visual Display of Quantitative Information.

Optional Readings

Krishna - Jan 26, 2011 12:24:52 pm

I am yet to be convinced by Tufte's arguments on data ink - especially the case where he starts tweaking the axes of a chart and almost makes them disappear using his data ink ratio argument. I think this would result in a design that is too minimalistic, he seems to have a wont for such designs. Also, I completely disagree with him on removing side-bars on box-plots or histogram bars. These side-bars, I believe, provide some contiguity that are important for perceptually organizing each bar and help distinguishing and relating them with other bars in a box plot. It could be an interesting UX study to find what the users actually prefer. Replacing the bars with single lines looked very incomplete atleast to me. Though I completely bought Tufte's arguments on chart junk and moire vibrations, I still think the latter can be put to good aesthetic use by a skillful visualization artist.

Brandon Liu - Jan 27, 2011 02:18:28 pm

Regarding the discussion of tree maps for the stock market visualization in lecture - I found the Wikipedia article on TreeMaps to be useful: http://en.wikipedia.org/wiki/Treemapping . The point I found most interesting was the tradeoff between the aspect ratio of sub-rectangles and the ordering of the rectangles. The major advantage to having square or near-square aspect ratios is that it is easier to compare the area of two rectangles. It seems to me a major limitation of the treemapping paradigm is that the size of the hierarchies should be appropriate relative to the dimensions of the diagram - for example, the Map of the Market a few dozen categories such as 'Financial' and 'Energy', but the effectiveness would break down if this number was in the hundreds.

Sally Ahn - Jan 27, 2011 02:58:51 pm

I am somewhat skeptical about some of Tufte's redesigns in chapter 6. The range-frame, for example, seems to focus on minimizing data ink and overlooking that absence of ink can also be distracting. I agree that muting non-data ink such as grids improves the legibility of graphs, but removing parts of a grid to minimize ink alters the grid system itself and thus introduces new distractions from the data. The white grid is another example that I question. I think the way the white grid cuts into the bars themselves creates potential confusion on why the bars are divided; the non-data grid interrupts the data. I would argue that muted lines of regular grids would be less distracting than such interruptions created to minimize ink.

Tanushree - Jan 27, 2011 03:31:53 pm

Some of Tufte's criticism seems to be directed, specially the "Chartjunk" section, towards the popular visualizations at the time. Using a light background grid and not using vibrating textures makes intuitive sense now. It was helpful to have the default options in Protovis draw a grid that was well spaced out and came in a light gray shade. While I may not agree with everything that Tufte said, some of his minimalist redesigns did help me see patterns that were obscured before (for example, where the grid was removed/ subdued and in the example where the boxplot is reduced to two lines and a dot). I think this might come up in the lectures on perception later - but I am wondering that so much of our perception is conditioned - We are so familiar with bar graphs and making sense of them, that it'll be hard to switch to some other design (of the kind that Tufte proposed) and even more, on a larger scale.

I enjoyed reading the assigned Tufte chapters because they help you question everything you add or not add to your visualization. For the same reason, it seems instructive to use visualization libraries such as Protovis (and others) where you need to consciously add every element or mark to the panel - as opposed to say Excel or R where the graph is automatically drawn for you, and you then think about making appropriate changes/ erasures. It might be much easier in the latter case to fall into the comfort zone of blindly accepting the status quo.

Michael Cohen - Jan 27, 2011 04:31:53 pm

I think the value of Tufte's work is in articulating principles (or rules) that lead us to a workable, minimalist design. We can choose to break the rules as appropriate for the subject matter or audience, but they give us a framework within which to justify our decisions. Because he's critiquing existing graphics, Tufte frames his rule in the negative: "erase non-data-ink, within reason". But for those of us designing from scratch it's more useful to think about the converse: "start with the minimal ink necessary to convey the data, then add only what justifiably improves the visualization". It's not a hard and fast rule, but it leads to clearer design thinking than if we had no principles about ink usage. I agree that using a tool where you build your visualization up from scratch makes the principle easier to apply.

Poezn - Jan 27, 2011 05:16:11 pm

The discussion we had in class about Wattenberg's stock market tree map revealed a very important point: The purpose of the visualization has to drive the design decisions for a visualization. Wattenberg explains that [[T]he goal was to give a quick answer to the question, "what's happening in the market?"]. From this point of view, I would argue the visualization is very effective. Without having to interact by scrolling or drilling down into the visualization, it shows the areas if and where the market is acting unusual, either as a whole, or just parts of it. When we talked about the "story" of our assignment 1 visualizations, we did the same: figuring out whether the design decisions help conveying the story.

On a side note, I was wondering in class why he did include the yellow/blue color spectrum into the visualization, when red and green have such a strong connotation to their meaning (negative/positive). The reason he did that, I guess, is to allow color blind people (10% of men, and financial analysts are predominantly men) to still be able to use the visualization.

Saung Li - Jan 27, 2011 06:56:37 pm

I found Tufte's discussion of data-ink quite useful and interesting. Had I read about this earlier, I would have removed the horizontal grid lines in my scatter plot in my assignment 1. This would make it look more pleasing to the eye and allow viewers to focus more on the data points themselves. I do not fully agree with how much Tufte removes from some of the graphs, though. In the bar chart, he goes as far as to strip the box/axes into a thin baseline, which looks simple, but seems to take away what the Y axis better conveys. It looks like the Y axis values are just floating there, and a Y axes would make it look better organized. I think even just having the tick marks is better than having nothing to pinpoint the values. The removal of the bottom left corner of the range-frame scatter plot also makes it look weird. However, in both cases it may just be a matter of getting accustomed to these changes, as I can see them as potentially being helpful.

Dan - Jan 27, 2011 09:59:18 pm

The beginning of the reading was extremeley captivating: "The real powers come from devising external aids that enhance cognitive abilities ... increased memory, thought, and reasoning!". Representation plays a huge role in our lives and increases the bandwidth of information transmission. I also really liked the dialog with Socrates: "once a thing is put in writing, the composition, whatever it might be, drifts all over the place, getting into the hands not only of those who understand it, but equally of those who have no business with it". This is very interesting because it seems to imply that there is a universal language underlying graphical representations of ideas and objects, such that anyone can understand it. That is the beauty of visualization. However, the author also discusses the fact that representations are abstractions and sometimes the distance can obscure the true meaning or actual truth of an element.

Michael Hsueh - Jan 27, 2011 9:48:07 pm

I appreciate Tufte's sense of humor and find his pull-no-punches approach entertaining. I found most insightful his condemnations of the use of widespread moire vibrations and "graphics for the sake of it" practice.

I had minor issues with some of his proposed alternatives that maximize data-ink efficiency. On page 128, he suggests replacing grid lines with background-colored lines so as to reduce high frequency distractions in the graphic. This certainly does reduce ink utilization while preserving the information encoded in the graph. Still, I believe it does not actually make the important data the graph more easily accessible, which is the point of reducing non data-ink. Take for example the shortest bar in the chart, whose top lies beneath the lowest "invisible" grid line corresponding to 5%. It is difficult to gauge how far beneath 5% the bar actually reaches, since the viewer must now use his imagination to extend the grid line across the chart. Tufte seems to qualify his recommendations by using the phrase "within reason." Perhaps he would argue that in this case, the optimization is within reason because seeing the relative position of a bar from a major grid line is not important for most viewers of the graph. But I actually think this feature can be quite important for most bar charts, and not just a limited use attribute.

Tufte's revision of box plots are also interesting. He proposes the concept of offsetting lines to indicate different quartiles of data. This works quite well in principle to minimize ink while preserving data, but I don't think it works quite as well in practice when human visual acuity is factored in. I myself found it difficult to discern the different lines (perhaps I have a faulty copy of the text), though a computer or robot hooked up to an imaging sensor certainly wouldn't. Note that this is not an issue of me not being accustomed to the *style* of presentation (Tufte addresses this and I agree it's not a valid reason to dismiss a visualization), but rather simply an inability for the human eye to easily resolve those minor differences.

By and large, I liked Tufte's criticisms of problems in visualization practices. I found some of his suggestions extreme, but mostly agree in principle.

Natalie Jones - Jan 28, 2011 12:39:40 am

Like many others who've commented here, I found a lot of Tufte's points about data-ink ratios to be thought-provoking and informative. The main question I have after reading his writings about it, is: what really constitutes "data-ink?" It might seem obvious to some people, but to me it seems debatable. If something is more understandable with a label than without, then couldn't you argue that that label is part of the data? Most lines on a page represent some kind of data, it just depends whether someone thinks it's essential or not. And that also seems to depend on what particular information (i.e. story) that graphic wants to convey.

I also agree with some of the other comments, that some of his graphics were not necessarily more understandable when stripped of some of the excess ink. I found that simply breaking up clean lines and continuity was sometimes distracting for me, even if it improved the data-ink ratio. I think the argument is really valuable on a theoretical level, though, and suspect that Tufte is emphasizing the point to drive it home. We are surrounded by gratuitous information in all formats, and we can afford to think about trimming a lot of it.

Matthew Can - Jan 28, 2011 12:04:24 am

In his book chapter on representation, Norman describes how representations and their accompanying cognitive artifacts can help aid human thought and reasoning. He stresses the importance of choosing a representation that is appropriate to the task it is intended to support. In particular, he makes a distinction between two kinds of representations: those that require us to think reflectively about information and those that allow us to experience information.

Though Norman says some tasks should (or must) be reflective, his examples clearly favor representations that make tasks experiential, especially for tasks where the goal is to understand information. One way to look at this discussion is as a generalization of what we learned in class about creating effective visualizations. We are trying to build a mapping (a representation) from data to visual variables that makes the information most easily perceived (an experiential task). I thought this was an insightful reading. I'm also curious, can anyone think of a situation where it would be appropriate for a visualization to include reflective cognitive artifacts?

Siamak Faridani - Jan 28, 2011 01:13:31 am

One of the aspects of Tufte's book that is becoming very frustrating for me is the fact that he overlooks the existance of Interactive/Dynamic data visualizations. I know many people like his book but his book is becoming very difficult for me to read. Firstly, he sometimes presents "ideas" as "hard rules". For example I am not convinced that aesthetics is not a main element in visualization, we have quotes from theoretical physicists saying that if a theory is not beautiful it is not worthwhile, why should we ignore the element of beauty in visualization? Additionally, the concept of data-ink is interesting but I am not yet convinced that it is always correct. I see how sparklines can be a result of this thinking. I imagine the data-ink value is maximized in a sparkline but how much information can we really extract from it? I am also wondering if anyone in fact uses a sparkline to represent data?

Sparkline.jpg

David Wong - Jan 29, 2011 02:26:09 pm

I liked how Brandon cited the wikipedia article for treemaps as I wanted a bit more clarification on the exact algorithm. The entry in wikipedia directly quotes: "As the aspect ratio is optimized, the order of placement becomes less predictable." I found this to be of significant concern because arrangement of the squares within a sector is very important to the overall visualization. For example, if a sector is experiencing high growth, but only in a handful of stocks, if those stocks were arranged next to each other in the treemap, we'd have a large green area. However, if they're arranged spread apart, with no squares adjacent to one another, the sector might appear less green. Overall, there is a bias that comes with arrangement that should be accounted for in the algorithm. Furthermore, if a user came to the site on two different days where one sector had the exact same stocks shown, it isn't clear whether or not the arrangement of those stocks would be preserved or not.

I agree with Poezn's post that his visualization is effective in portraying "what's going on" in the market at a glance. While you do get a sense of the aggregate, it's utility is quite limited if you wanted to get any more specific information.

Karl He - Jan 29, 2011 08:35:30 pm

In response to Matthew's comment, Norman's example of the car accident is a great example. The scene of the accident is recreated in order for the people involved to contemplate alternative actions in the situation.

Hopefully I have the definitions right, but I think it'd be similar to a hand-drawn map vs a full-blown map. A hand-drawn map would likely be based on actions you'd have to take with following the map, vs a full map which just shows everything, and you'd need to interpret which actions to take. Perhaps an even better example would be a map vs online driving directions in Google maps. While in the latter you don't need to think, you have no choice but to follow the directions, and anything preventing you from following the directions would be a dealbreaker, whereas in the case of a map, you'd just look for an alternate route. You'd have to do more thinking, but the information would be accessible from the visualization.

Julian Limon - Feb 01, 2011 12:50:43 am

Norman's example of conveying flight information using the Official Airline Guide (OAG) reminded me of Hipmunk, a startup that was launched very recently. Just like Norman suggested, Hipmunk displays flights using bar charts. It uses color to differentiate between airlines, airport codes to show layovers, the X axis for time, and the Y axis for different flights. To see an example of how a SAN-London schedule would look like, take a took at [1].

You can also see the flights sorted by arrival time, departure time, duration, price, or number of stops. However, when looking at the duration sorting, I noticed that flights are still arranged over a timespan. There is a trade-off here -- although the timespan makes comparisons between flights more difficult, it provides information about arrival and departure time that might be useful for the traveler. I suggest that a column on the right showing the total duration could be added to get the best of both worlds.



[add comment]
Personal tools