Using Space Effectively: 2D

From CS 294-10 Visualization Sp10

Jump to: navigation, search

Lecture on Feb 24, 2010




  • Multi-Scale Banking to 45 Degrees. Heer & Agrawala. (pdf)
  • Pad++: A zooming graphical interface for exploring alternate interface physics, Bederson & Hollan (acm)
  • Chapter 11: The Cartogram: Value-by-Area Mapping. In Cartography: Thematic Map Design. Dent (pdf)

Optional Readings

  • Generalized fisheye views, Furnas. (acm)
  • Hyperdimensional data analysis using parallel coordinates, Wegman (jstor)
  • A framework for unifying presentation space, Carpendale & Montagnese. (acm)
  • Nomography
  • Cartogram central

Jeffrey Patzer - Feb 24, 2010 12:37:09 am

Pad++ I think is an extremely interesting program. The reason I find it so interesting is that the issues they discuss and ideas they recommend, are essentially what modern browsers have become. We have a way to visually look at pages most visited, history, when moving tabs around, and so on. The zoom feature is implemented by either zooming through the OS (at least if you have a mac) or by increasing the text via the browser. The visual layouts they suggest are essentially what have become the norm for browser interaction. What I do have a slight issue with is the idea that by providing visualizations of your data (folders, pictures, web pages, w/e), it does not necessarily help me filter the information any better. Rather I find it more difficult to locate a most commonly visited website via picture rather than typing their url myself. It would be interesting to see if a more effective way than zooming could be found for interacting with large sets of visual data.

Arpad Kovacs - Feb 24, 2010 10:40:04 am

Even after reading Dent's chapter, I still have mixed feelings about cartograms. On the one hand, contiguous cartograms can be encoded without generalization/simplification, resulting in extremely high data density. Dent also lauds contiguous cartograms for preserving shape, which he deems to be the paramount factor in cartographic design. However, he concedes that the user must have familiarity with the source data/map, or I think that alternatively the unmodified map could be placed side-by-side for comparison purposes.

Another issue with cartograms is the difficulty of and lack of accuracy in data retrieval. As discussed in earlier lectures, humans perceive area less accurately than position, or length. This is mitigated by Dent's observation that cartograms should focus on conveying an overall thematic distribution, rather than detailed values. However, there still remains the problem that regions with value=0 will not show up on the cartogram, so the reader must explicitly be aware of the absence of these areas.

Overall, I think that heavily distorted, but still recognizable cartograms can be very persuasive, by conveying spatial peculiarities in the preattentive phase of processing. However, compared to other visualization forms, it takes much greater serial processing effort on the part of the reader to gain more than just a cursory understanding of the overall trends.

Danielle Christianson - Feb 24, 2010 02:02:37 pm

Two thoughts on Pad++: 1) I think that developing ways to organize / track information within the mass of information that a person explores is super important in allowing for efficient use of the information. Nice work especially on the html hyperlink alternative. 2) I also appreciated the authors emphasis on a physics-based approach to new information visualizations, etc (vs. metaphor-based). I do think metaphors can be constraining but in practice not using metaphors is hard not to do.

45-degree banking: Still rolling this around in my head. Feels a little similar to the adage: statistics can be used to tell any story. And I am a little leery of assigning algorithms to determine the best representation -- I worry about reduced human thinking. I do like the idea of showing multiple aspect ratios so that the viewer can get a more complete picture. I am a bit confused about how the power spectrum is getting the *best* views...

Jon Barron - Feb 24, 2010 02:48:29 pm

Multi-scale banking: The problem of identifying good aspect ratios for line-graphs is certainly a good one, and this approach (though a bit messy) seems like an intelligent solution. I was a bit disappointed that the evaluation was not more quantitative. Looking at the output, I had a hard time convincing myself that the chosen aspect ratios were "good", or that the data deserved multiple aspect ratios. I had hoped that the evaluation would have been something like: "we presented these visualizations of the data to people in a study, and had them record every trend they could notice", and measure the precision/recall of the recorded trends. Or what if the solution was, instead of presenting multiple graphs, presenting a single graph with the largest aspect ratio, with low-frequency versions of the data rendered on top of each other? This might give users access to the equivalent information, while only requiring a single visualization.

Cartograms: I really like cartograms in theory, but this article reads like a long list of reasons why cartograms are bad. And unfortunately, this article seems to pre-date the idea of automatic generation of cartograms... I can't imagine how horrible these are to make by hand. Are there any articles on automatic generation? The choice of loss function interests me.

DavidZats - Feb 24, 2010 02:54:40 pm

These papers discuss three different issues in visualization. The first is how to scale graphs to present the most valuable information. The second is how user interfaces should be modified in order to become more effective. Finally, the third discusses the merits of cartograms.

One question I had about the reading was the extent to which user interfaces have been modified to take advantage of the ideas presented in Pad++. While the ideas in this project of avoiding metaphorical representations and instead using physics seem to be very powerful, we still use files, folders, and desktops today. Are there many examples of successful user interface modifications based on these principles that have made it into mainstream operating systems?

Another question I had was about the cartograms chapter. This chapter discusses how values can be encoded as area of objects (i.e. sizes of states in the US) to convey information to the user. It seems that this approach violates one of Tufte's principles: multiple dimensions (i.e. area) should not be used to encode uni-dimensional data (i.e. population). Additionally, we have discussed that discerning the differences between areas is much more challenging than discerning differences in position or length. Given these issues, how appropriate are cartograms as a data presentation mechanism?

Jonathan Yen - Feb 24, 2010 03:11:18 pm

MultiScale banking paper was a good read. I wonder if there's more to be considered in terms of additional data that can be added along to a line graph. We see line graphs everywhere. I wonder if there ought to be more data that can be provided in line graphs. What about the usage of area above/beneath the line graph? If we do a fill, will that help in the comprehension of the data? Also, line graphs are often associated with time--would it be more relevant to provide finer detail of data that's closer to the present time vs. information that's occurred further back in the past?

The Cartogram reading was also a good read as well. I think it's really interesting how you can take some sort of representation and generalize it somewhat to convey information yet retain the symbolism/identification of that representation. I'm reminded of another similar and interesting article that I read here: Realism in information visualization is something that could be examined further, as there's advantages and disadvantages to realism, just as there is for information representation.

Boaz Avital - Feb 26, 2010 06:14:37 pm

Banking graphs to 45 degrees to highlight trends is an obviously effective strategy, but does it lie to the viewer? If I got a $1 raise, would I want to bank my salary graph to show that as a trend for rising salaries?

For cartograms, you need to know a lot more about your audience than other visualizatoins. You need to know what kind of maps and shapes they're familiar with so you're cartogram transoformations make sense and have meaning to them. Perhaps a good interactive technique for displaying cartograms is to have them change between normal mapping and transformed mapping when you mouse over them. Also, a color encoding of each area of the map should help you match the area to its transformed counterpart.

Mason Smith - Feb 28, 2010 08:43:31 pm

Wegman: I was largely unfamiliar with parallel coordinates before this reading. Generally, they seem pretty useful, but definitely not intuitive. In Figure 7, for instance, I found it initially difficult to distinguish between a crossover set (indicating negative correlation) and the hyperbolic boundary (indicating no correlation) in the price-mileage pair. The parallel coordinate density plots seem particularly useful for dense plots; I could see them being used as an underlay, for instance, with the actual plot on top, to motivate analysis or subset selection with sliders. The histograms, on the other hand, were much more difficult for me to parse. The other issue is the number of graphs needed to display all of the adjacencies. On the other hand, a scatterplot matrix has n^2 entries, so ceil( (n+1)/2 ) parallel plots isn't actually that bad. Additionally, for particularly high-dimensional data, an auto-generated ordering might be possible (for instance, by maximizing correlation coefficients between adjacent entries).

Zev Winkelman - Mar 01, 2010 12:32:25 pm

I was impressed by the work on multi scale banking. The original study, as well as the alternate methods, and expansion of methods presented in the paper, continue to reinforce the message that visualizations can be manipulated to emphasize different aspects of the data. On the one hand this is encouraging if you are trying to demonstrate a particular point of view. On the other hand it is disheartening if you are searching for some objectively 'true' way of presenting the data. Truth, I suppose, does not have to be mutually exclusive, and, in fact, the best part of the multi scale banking algorithm is that it reveals the candidate aspect ratios based solely on the characteristics of the data. The candidates can then all be given equal weight in a small multiples display (3.3.1 Small Multiples Reports). Too bad there isn't such a politically neutral manner of creating the data in the first place.

I also really enjoyed the article on parallel coordinates. When I have worked with multivariate data in the past, I have encountered challenges creating visualizations beyond 3 (or at max 4) dimensions. It's good to have this tool in the repertoire now as well. I was also thankful that the authors pointed out that only (n+1)/2, not n!, permutations are required "so that in some permutation every axis is adjacent to every other axis" (670) - just in case I ever wanted to look at them all.

Stephen Chu - Mar 03, 2010 05:29:45 am

I agree with the comments above that one weakness of cartograms is the increased difficulty in differentiating area as opposed to position/length. Preserving the shape of the units of a cartogram does provide readers more information but it also complicates area comparisons. E.g. comparing Florida, a state with a long, "stretched-out" shape with Texas. As mentioned in class, cartograms also might not be able to handle outliers well. The Elvis concert attendance visualization is an example of an outlier (Nevada) requiring an awkward explanation for its absence on the map. Another problem with cartograms arises when the data has very high variance. Attempting to construct a contiguous cartogram with this data will be difficult and time-consuming. In this case one has to trade between the advantages of a contiguous structure with the ease/quickness of scalability and construction.

Kerstin Keller - Mar 03, 2010 10:26:17 am

I liked the "Multi-Scale Banking to 45°" paper. When I was working with Tableau for the last assignment, it often scaled the data by default in such a way that you couldn't really see any trends but just a more or less straight line. I also agree that different scales show different aspects of the data, but it's the responsibility of the designer to choose the one that is most appropriate for what he wants to say, ideally he includes multiple scale factors of the same data.

Cartograms: I initially liked the idea of cartograms, but I must say that most examples given in the paper rather confused me and I found it hard to extract the information, especially if you're unfamiliar with a map. Dent's suggestion, to include a map with the real proportions sounds good. The fact, that most cartograms have to be drawn by hand seems to be a drawback, I am not sure if it is worth putting so much effort into creating a cartogram.

Akshay Kannan - Mar 06, 2010 11:58:37 am

I found the work on multi-scale banking to be particularly fascinating. While it is clear that the aspect ratio of a visual can significantly impact the information/trends that are emphasized for the user, it is particularly difficult to determine any ideal aspect ratio, since different configurations are capable of telling different stories. 45 degree banking definitely seems like an interesting approach to this. As for cartograms, I am yet to be convinced of their effectiveness of visualizations. For example, for area-mapped cartograms of the United States, I tend to be more familiar with shapes of states rather than sizes, so a visualization approach that uses the sizes of a state to map a given data set seems somewhat ineffective. I feel that a better strategy would be either encoding in luminance, or encoding in different colors to represent different ranges. Even a simple bar graph, in conjunction with a map with corresponding colors, would be a clean method of implementing this, although it would allow users to do a mental "join" between the two visuals.

Prahalika Reddy - Mar 10, 2010 04:42:17 am

I'm not sure I understand what "banking to 45 degrees" really means. The idea about showing the greatest discriminability between two line segments makes sense, but trying to "bank trend lines to 45 degrees" doesn't.

The cartogram reading was very interesting. The concept of cartograms is pretty cool, and I think they are a very useful visualization tool. I don't quite see the need for non-contiguous cartograms however. It seems that contiguous cartograms are a better choice in most cases. I especially liked the figure that defined the reader and cartographer tasks; it's nice to see what the visualizer should do to convey the best information the user. It is very helpful if the visualizer knows what exactly the user needs and how to provide it.

Priyanka Reddy - Mar 10, 2010 08:18:54 am

I think cartograms are an interesting data visualization technique. I say interesting because I'm not sure how effective they are. If trying to show quantitative information, I think cartograms are especially ineffective. Not only are they area comparisons, which we know is harder for people to perceive, the original shape and size of each state/country/etc is different. That means that not only does the user need to know the original size and shape, but also needs to be able to mentally normalize the map in order to do any quantitative comparison. However, I think they are effective for certain qualitative visualizations. If the goal is to shock people with the results and if there's only a little bit of data to display (ie. 5 largest states), I think cartograms would work. The graph would probably go against most of Tufte's ideas, but with the trend of presentations becoming more visual and audiences expecting more unique/shocking/memorable presentations, these cartograms will continue to be used.

Shimul Sachdeva - Mar 14, 2010 08:31:05 am

Slopeless line culling addressed a common problem with data visualization - outliers. While it is true that at times, to make data more readable, it is a good idea to ignore outliers, I am not sure how its affect on the accuracy of the data scales. The concept of small multiples has been discussed in various contexts in class and the one addressed by the paper seems useful; the examples were good. Pad++ seems to be an easy tool to learn and use. When I compare it to other interactive software like Photoshop, it wins by a leap in terms of the gentleness of the learning curve. It is necessary to address the user's ease in understanding while making interactive tools. The features of Pad++ are great for zooming-in. A good parallel running on newer UI is prezi. Cartograms are an interesting way of representing information. Although, again, outliers pose an issue, but barring those, I think cartograms can convey some information more explicitly than numbers. Given that not everyone might know the details or outline of the original diagram/map, an inset entry with the original shape is good to have.

[add comment]
Personal tools