Tufte Takeaways

firebus's picture

My company bought me a ticket to the Edward Tufte one-day course. It feels like something that someone in my position should have attended a long time ago. Not that I do that much data visualisation, but I do a little, and Tufte has had a lot of influence on the UI/X world which is parallel to my work.

Tufte is a weird constant in that space. For some reason I put him in the same bucket as Jakob Neilsen (if Jakob Neilsen had managed to stay relevant he'd be the Tufte of UX). Somehow Tufte's courses are like religious revivals, and that ubiquitous chart of Napoleon's army in Russia is its holy relic. There should be a Dilbert comic about it.

Overall, it was so-so. Like reading a Seth Godin book, this was a large-print experience with lots of repetition. The class wasn't targeted at visualization professionals, but at the layperson (which I appreciated!), and Tufte spent a lot of time talking about how presentations should be structured. Any kind of presentation. He spent a lot of time dissecting concrete examples of visualizations, but the meat of the talk was a framework for giving presentations.

However, Tufte is more of a narrative, long-form thinker than a framework builder, and I'm not sure his message really fits in a one-day format.

The course supports other parts of Tufte's business. As part of the course, you get hardcover copies of 4 of his books. The class costs $380. The 4 books list at $155. I'm sure that Tufte gets the books at a steep discount, and getting autographs from Tufte is a big draw for many attendees (if the lines at "Office Hours" are any indication), but I think Tufte is craftily using his popular course to keep the sales numbers increasing for his books.

Edward Tufte has turned to sculpture recently, and has an exhibition of sculptural interpretations of Feynman diagrams. The interstitional movies that were projected during breaks show him forging hot steel to create the works, works for sale at Sothebys, and work in an outdoor scuplture garden surrounded by hungry sheep.

Okay, here are the major things I learned:

* There's a continuum of UI awesomeness. At the terrible horrible very bad end is a voice mail phone tree where you can only make one yes/no choice at each node. Slightly better than that is a powerpoint slide where you get spoon fed bullet points. At the awesome end is a giant flat surface (like multiple monitors, or the ESPN website) that's filled with various areas of text, tables, and visualizations of data. The less hierarchy, clicking, time-separation of display the better. The big flat sheet is interactive and dynamic because everyone can find and focus on what they care about.

* So instead of using powerpoint, you should distribute (on paper or on screens) all the data you want to discuss and spend the first 5-10 minutes having everyone read. Even slow readers can ingest information faster when reading than they can by powerpoint. Then you can have a discussion about what was read. Then you can answer questions. Since I suck at powerpoint, this is attractive to me.

* There's a similar continuum of visualization awesomeness. At the good end, text, table, and graphics all exist on the same page, nicely laid out. Text is used to annotate images. Images (e.g. sparklines) are used to annotate text.

At the bad end, there are gratuitious boxes around all the nouns, and figures are in a separate section at the back of the book, with footnotes, and you have to flip around. In the worst case you have a "Book Operating System", where you need a set of instructions to tell you how to use the text and figures.

You can do the same thing with a touchscreen interface when you try to build another interface of virtual tabs, sliders, etc. on top of the touchscreen interface that's already there.

* If you put boxes around nouns (e.g. in an org chart) the boxes create negative space gutters that draw the eye away from the information. Tufte hates gratuitous boxes and drop shadows even more than he hates powerpoint.

* Tufte talked a little bit about credibility, overlapping with the content of "Statistics as Principled Argument" (about the rhetoric of statistics) which I've been reading. He looks for open data (so you know the presenter isn't cherry picking) and has a short list of common statistical bad behavior to avoid/look out for. Tufte would also like you to cite your references. Pretty basic stuff.

* Much of the second half of the class was spent on ranting about how terrible doctors are at making data-based treatment decisions. A lot of good arguments, but not very relevant. Tufte would like you to give your doctor a written agenda at the beginning of a consult, which is not a bad idea.

* Tufte has a problem with big data, specifically with data mining for correlations. It has a nice set up:

** If you have an alarm system, you're no doubt aware that almost every alarm is a false alarm. "All the money is in the false alarms". If there were no false alarms, you'd pay a lot less for armed response, because they'd only need one guy.

** The medical establishment is no longer recommending PSA (prostate cancer) screening for all men because the number of false positives was so high, and the cost (death, permanent injury) from the follow-up biopsies was so high, that it outweighed the benefit to men who really did have aggressive cancer. All the money was in the false alarms.

** Data mining works the same way. You're always going to find correlations. If you then make interventions based on those correlations, you're probably going to do more harm than good.

** Tufte does like targeted big data efforts - when you are asking a specific question and answering it with a large data stream instead of just fishing for bricks.

** There are a lot of people in the predictive analytics space who would vehemently disagree with this characterization of the practice, and I've heard a lot of anecdotes that counter Tufte's feelings on this.


Powered by Drupal - Design by Artinet - Amazon Affiliate