Designing Tables and Graphs to Enlighten
From Visualization Sp06
Lecture on Jan 24, 2006
Slides (9MB pdf - requires password)
Readings
- Chapter 8: Table Design, In Show Me the Numbers. Few (handout)
- 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. Tufte.
Cynbot - Jan 23, 2006 06:37:50 pm
In "Show Me the Numbers": As I will be out of town tomorrow, I will miss the discussion of this article. However, I wanted to note that the discussion of data precision and font use were very helpful for me. As a mechanical engineer, I was trained to use as many decimals as I were able to reliably measure (experimentally) in my engineering documents. Too often I let that carry over into other documents where it is unecessary. So as another student noted on this wiki, learning to transcend ones point/goal into different forums is begining to be a mantra in visual explanations. Fonts in general fasciate me. I would like to see a discussion about the importance of fonts and font design.
Jingtaow - Jan 24, 2006 11:11:54 am
Tufte made some good points about data illustration and graphics revision in the readings. His five principles looks valid in general. However, IMHO, I feel his arguments to support these principles are not tight and compelling at this stage. Showing materials that support the principles should only be considered as part of the arguments, are there any trade-offs that need additional consideration? Under what situations over-optimizing data-ink ratio might look bad ? In addition, I don't agree with Tufte's arguments in pp97 that half of an Chernoff face is better, cause showing half of a Chernoff face wont change the data-ink "ratio" and such change hurts the "visual integuity" of the original graph a lot.
Bryan - Jan 26, 2006 02:27:31 pm
As someone who has been writing web pages for years, I remember when HTML tables first gain widespread use in the mid '90s. There were very few options, so most of the tables ended up looking like the worst-case examples from "Show me the numbers," with huge, 3D borders separating headers and data alike. As HTML matured and I got better at it, I got more and more power to fix up tables, alternate colors, pick where separators were, fix alignment, etc. I remember one of the easy-to-spot things about a well-designed webpage was how much the table attributes had been customized from the horrible defaults. It was nice to see a discussion that attacked so systematically table formatting. It seems that the "default" presentation--whether it be for HTML, LaTeX, Excel, whatever--is almost always bad, and effort expended thinking about how to lay it out is never wasted.
Ryanaip - Jan 26, 2006 10:38:44 pm
I agree with Jingtao's comment. In many cases, it does seem that maximizing the data-ink ratio is a good general guideline to follow, but only to a certain extent. As a specific example, I find the redesign of the quartile plot (Visual Display of..., p. 124) to be rather difficult to read. While the proposed redesign does use less ink than the traditional box plot, it makes it more difficult to see the encoded information.
Yi-Tao - Jan 27, 2006 09:01:32 am
In regards to Cynbot's comments: There's not much to say about fonts. Even in the article, there's only a very small passage devoted to it. I think the reason is that most people don't use poor legibility fonts. Websites, essays, resumes, etc are always in one of the "fine legibility" fonts. Although I have seen people use boldface, italics, and font size to emphasize the incorrect information. I would have preferred some discussion about that in Few's paper.
In regards to Bryan's comments: I don't think the amount of customization is a good indicator of the quality of a website. I have seen websites with tons of poor customization. This one site (I think it even won an award but I can't be sure) chose brown text on brownish orange background for their tables; just looking at it gave me headaches. I think the point is that the complexity of the visualization is irrelevant as long as it's intuitive.
Noaa - Jan 27, 2006 03:51:31 pm
I think Tufte went a little overboard with his villanization of what he refers to as redundant data ink. Just because a point can be made with 4 lines instead of 6 doesn't mean that the 6 line solution is so terrible. For example, his solution on pg 124 for Tukey's box plot seemed a bit overkill to me. The very very slightly offset line and an almost imperceptible gap being used to represent the distribution of the data didn't seem like such a spectacular design decision for me. I find it hard to distinguish exactly where the offset part begins and ends without very close analysis, and it makes it difficult to compare the length of the offset to the rest of the line, as well as its position relative to it. While it works interestingly as the axis line on the graph on pg 132, I think in general Tufte takes his erasing mantra a little too far.
Raymond - Jan 29, 2006 03:11:12 pm
In Prof. Few's presentation, he mentioned that the developers themselves should take fault for allowing users to create such "horrible" visualizations. He goes on futher to say that end users need to be trained in creating effective visualizations. When he originally made this comment, I didn't think that this was actually prevelant until a couple days ago I saw a fellow classmate make a pie graph for her econ class. Not only was the pie chart in 3d (thus lowering the data:ink ratio), but you had to look at the percentages to compare each slice. Obviously the information would have been better off as a table. On a side note, I can't really think of an instance when a pie graph would be the visualization...
p.s. when will lecture slides be posted?
Maneesh - Jan 29, 2006 10:35:44 pm
Jingtaow - It's true that Tufte's principles are presented a a fairly high level and don't always hold in every situaion. It would be fun to do a more rigorous analysis of his principles. The data-ink ratio principle would be a great one to start with.
Bryan - I too am amazed at the poor defaults in most computer-based table design systems. It seems like we should be able to fix this problem.
Nchentan - Jan 30, 2006 2:09 am
In addition to Jingtaow's comment on the data ink ratio, I also disagree with Tufte on page 96 about the shaded bar chart example. I think that each of the six ways can serve some other purposes other than just indicating height. For example, the shaded region of the bar, when viewed amoung other bars, give us a better sense of comparison between data points and give a rough estimate of the magnitude of the quantity. The number itself gives an accurate value, which is difficult to obtain from the graph alone. The left and the right lines help seperating the two adjacent bar visually. The bar chart without left and right seperating lines between bars, where each bars being shaded the same way, would looks horrible. From these reasons,I would think that in many cases we would still should have all or most of the six elements in a visually good bar chart.
IvanTam - Jan 30, 2006 11:22:31 am
One concern I had about Tufte's Data-Ink Ratio rule of designing his graphs, is the accessiblity of the presented information to populations with less than perfect eyesight. While, I think Tufte's Data-Ink rule produces graphs that may be visually appealing to a certain demographic (contemporary graphic designers?), elderly or other visually-impaired demographics may find it difficult to extract information from the same graphs.
Todd - Jan 30, 2006 01:56:55 pm
I found Tufte's book to be very enjoyable, but I do think we need to carefully understand where he is coming from and what he is getting at, rather than blindly accept his principles. It would seem that he is pursuing a particular style of minimalist aesthetics, as his overriding concern. His goal is to get as much data into as little space as possible, while using as little ink and with as little ornamentation as he can get away with while still displaying the data. Some of the examples, such as the simplified box-whisker diagrams, are more an exercise in taking these ideas to their logical extreme, than they are examples of what you would actually want to do in the real world. By making his examples so vivid, he makes his point perfectly clear, whereas a more balanced example might be more ambiguous.
One complaint I have with a lot of his favorite diagrams is that, due to their stripped down nature and lack of grid lines, you really can't read ANY numbers off it. What you can clearly see are trends and the overall character of the data. I see this as useful in a very limited sense; if you want a quick feel for the data set as a whole, then these diagrams are good. You would still need a table to get any kind of quantitative data out of it.
For all the fuss Tufte makes about focusing only on the substance, and not on the style, his diagrams are so unusual that when I look at them, all I see is their unique style, drawing my attention completely away from the data they are actually presenting.
Sharena - Jan 31, 2006 12:11:55 am
I also agree that some Tufte's examples of erasing all non-data ink seem a little extreme; however, Stephen Few in Show Me the Numbers also talks about using "the minimum means necessary to do the job"(pg 138). And I think his examples very clearly show how you can use whitespace instead of grid lines to more effectively communicate information. He talks about how you can distinguish the Total column of a table with a simple line instead of a complete box, and how most often full grid lines are not needed. Here I think is where removing the non-data ink makes the tables much more effective, more so than Tufte's changing the box plot into the smaller quartile plot. Although, I did think it was interesting how Tufte changed the traditional scatter plot into showing the range and distribution of the data as well. I do not know if it always makes the data clearer, but I thought it was a very interesting way to show more information about the data by acutally taking away ink.
Brien - Jan 31, 2006 12:54:35 am
I like Bryan's comment about terrible defaults across all technologies. I wondered if their designers had any good examples to draw from, so I opened up some old books (circa mid 80s) of filter tables -- and wow! If these graphics design principles had even reached electrical engineers back in the day, somewhere between then and now they got radically lost (perhaps a y2k issue). We experienced a total downgrade.
Stephen Few says that there is rarely a good reason to orient text vertically. I think one situation in which vertical text can work is when the data are indexed by two values (like a multiplication table) and the header of the first column is long. If the header can be pushed to the left and oriented vertically (like a y-axis label), the resulting table may consume less space and be easier to read.
I agree with Sharena that having just two horizontal lines in a table -- one below the header and above the total -- is super elegant.
AaronHoover - Jan 31, 2006 12:12:19 pm
I had many of the same reactions as the previous posters to Tufte's minimalist aesthetics. The entire time I was reading about increasing the data ink ratio and reducing so-called "chartjunk," I found Tufte's reductions to be nearly inscrutable.
It was also interesting to me that Tufte anticipates the reader's discomfort with his examples but claims that the discomfort is just because the reader has been exposed to the old paradigm for years. However, it's not like his proposals require the use of new technology that wasn't previously available, so it leads me to wonder why they haven't been explored before, given that we've been communicating graphically for quite some time now. I'm not necessarily asserting that he's off the mark, but I'm also not entirely on board with his claims that his reductions/modifications drastically improve the effectiveness of the graphics with which he's working.
Mehershad - Feb 02, 2006 12:44:18 am
simplicity is the name of the day In the past whenever I have tried to do some form of visualizations I have always found it best to stick to the good ol' types of graphs and today Stephen Few just confirmed my belief. Not to say that there is never a need to use unorthodox graphs but the point of the graph should be to get certain information across to the user. As Stephen pointed out, in the business world one encounters so many 'fancy' graphs just for beauty sake. But at the same time when presenting to a customer it might best to wow them or ignore integrity. Some of the graphs which Stephen illustrated where they play with the scale to make ones product look far better than another when it is just slightly better might not be the most accurate visualization or ethically correct but in the business world it still goes.. Morality kept aside, sometimes i think we are harsh on the creators of graphs like that because they did achieve their purpose even if they did not follow the convention/theory behind creating visualizations.
