Wednesday, August 1, 2007

Notes from Ed Tufte

Sorry for being silent in the last while. Blame it on the book I'm writing (on data integration). I had to finish a chapter before I could write anything else.

After several years of getting the brochures in the mail, and watching Tufte's book on envisioning information sit happily on my shelf, I decided to go for his 1-day course in SF and learn a thing or two about effective visualization.

If you want the full experience, I recommend sitting next to someone who worked on Microsoft Powerpoint at some point in their career. I did that, and it added quite a bit to the entertainment factor of the course. Tufte's powerpoint rant starts about 5 minutes into the course and he makes his last jab in his closing remarks. But more on that in a bit.

The course was interesting, even if it mostly gets you thinking about issues relating to effective visualization. I jotted down a few notes that I'm repeating here mostly so I don't forget next time I'm preparing a presentation. Most of them are obvious, but that doesn't mean they're not often forgotten.


  • more detail in your presentation increases your credibility
  • more detail does not necessarily imply clutter (if done right)

  • annotate anything you can in a visualization. For example, annotate links (otherwise you're implying that they all mean the same)
  • don't try to be too fancy. Focus on the content not on the design. For example, tables are a very effective, yet simple presentation. Order the rows in the table according to some performance measure you're trying to emphasize.

  • don't focus on being original in your visualizations, focus on getting it right (don't innovate, steal).
  • get users out of the decoding business (i.e., remove legends where possible)



The deeper point made in the course is that principles underlying creating effective visualizations mirror the principles that underly thinking processes. Hence, for example:

  • Make and show comparisons between different aspects of the data
  • Make sure causality of effects is emphasized in the presentation
  • Build credibility -- make sure you show all the data rather than just cherry-picking what's convenient for you.
  • Enable the audience to drill down and see more data.
  • Integrate evidence from multiple sources (aha, a plug for data integration!)
  • Always give the source of your data (yes, lineage, folks!)


Following the principle that good presentation should support critical audience thinkers, Tufte also points out what audience members should keep in mind as they listen to a presentation (surprisingly, reading your email on the blackberry is not one of them)
  • What is their story?
  • Can you believe them? Do they have a any conflicts of interest affecting their perspective? What's their track record? What's their reason for bias?
  • What precisely does their argument apply to? What are its limits?
  • What do I really need to know when I leave the room?


Ok, back to the powerpoint issue. I actually found myself a bit confused throughout the main point of his powerpoint rant, but I think I get it now. His basic point is that powerpoint forces you into a very low resolution presentation mode. He argues that people can read 3 times faster than you can talk. In addition, powerpoint encourages you to leave quite a bit of detail out and summarize everything in bullets. The human brain can take in much more than what you can convey with a powerpoint presentation. Hence, you're not really using your time with your audience very effectively, since there are better methods of conveying information that make much better use of the audience's mental capabilities. For example, he argues that you should come into a meeting with a 3-4 text summary of your points, have your audience read it, and then have a discussion and answer questions.

The latter suggestion makes it pretty clear when his methods are effective and when not. For example, it's a non starter for large audiences (e.g., conference presentations). On the other hand, there are cases where we do this by default (e.g., hiring meetings don't typically involve slide presentations). So, don't dump powerpoint just yet.

Interestingly, Tufte was not following his own advice very carefully during the day. I felt that the principles he espoused could have been communicated more efficiently (but then, perhaps he assumed that some of the audience also had blackberries to attend to).

3 comments:

Nan said...

It's useful, thanks Prof. Alon.

Anonymous said...

May I Know when you're planning to finish writing your book (just to have idea for sale's date), and is it only on data integration or with something else. Thanks

Anant said...

I am being bugged (nicely, of course :) on writing a book... Pray how do you find the time?