I saw Edwards Tufte’s seminar yesterday. I am not going to
summarize his more well known ideas since his seminars have been well
documented elsewhere. You can find
one here and another here. Tufte’s work and thoughts on
information display and Powerpoint are excellent, and I’m glad that I went. I’m
going to add a few general comments that are perhaps different from the
hundreds of seminar summaries published elsewhere, and I’ll also talk about some areas where I
disagree with Tufte.
Firstly, Tufte practices what he preaches…partly. For most of the
presentation, the projector was off and he only used it to show full-sized
images or videos. Other than that, it was just him talking and no PowerPoint
Phluff. Not even the usual title slide
when you walk in. I suppose we already know the speaker and topic, so a title
slide would have been superfluous—makes me think about which presentations of
mine don’t need a title slide.
Strangely, Tufte offered little evidence to support many of the design claims that he was making. His theories and ideas seemed sound, but there was no data presented that gave scientific weight to his advice on proper data design. In his seminar, he advocates that you view a presenter skeptically, making sure that they are a "detective" without bias rather than an "advocate" of their ideas. Tufte certainly sounds like an advocate in much of what he preaches.
Tufte brought several first edition books that were hundreds
of years old to demonstrate data layout by the world’s geniuses. He talks about
these examples in his books, but it’s amazing to see original books by Newton and Galileo in
person. I felt sorry for the person wearing archival gloves who had to slowly
and delicately carry the books around, pausing at each row so the audience
could see the book up close.
Accommodation of presentation design to cognitive styles is
a theme of Tufte’s, which I found interesting given the current discussion
of this in the blogsphere and in my own blog. Tufte argues that eliminating
bullet points from presentations accommodates a diversity of cognitive styles.
While it’s true that many people expect to see bulletpoint summaries of what
the speaker is saying, and a few probably assimilate information best in that style,
the bullet point mode is not optimal for the cognitive styles of most people.
By avoiding bullet points, more people in the audience are better served.
The PowerPoint structure entices people to reduce their
syntactic discipline and causal structure in our thoughts and arguments, so in
a way it affects the cognitive process of the presenter as well as the
audience. Tufte’s The Cognitive Style of PowerPoint should be read for detailed
thought on this topic. My final thought on this is that there must be a way to
effectively use PowerPoint that eliminates Tufte’s complaints, but Tufte
doesn’t offer such a solution–he recommends that a Word report be distributed at a meeting
instead. To that end, Cliff Atkinson is being more realistic in his work and
advice on trying to work within the PowerPoint paradigm, and Garr Reynolds at PresentationZen gives practical advice on improving PowerPoint presentations consistent with Tufte’s theories.
Tufte spent considerable time talking about sparklines,
his concept for tiny graphics inline with text that will be a significant part
of his next book to be published, Beautiful Evidence. You can find some discussion of sparrklines at Agile Testing and Anil Dash.
While the concept is
innovative, Tufte over-values the amount of information that these can provide.
As an example, he considers the plotting of a mutual funds price embedded
inline with text (the chart is the height of an uppercase letter and the length
around five letters). He argues that this tiny chart consists of over 250 data
points with a resolution of at least two significant digits (or, say, over 7
bits of resolution). Therefore, there are 1750 bits of information in the space
taken up by four letters. This seems like a reach, since certainly the reader
isn’t assimilating that many bits of information. People do a considerable amount of data reduction in their
perceptual space, and the amount of bits of information obtained from such a
small graph is considerably smaller than the actual resolution of the graph (think
of principal components).
The fact that people can discriminate two charts/sparklines
that differ in one data point doesn’t mean that they will interpret the two
charts any differently—the information to the reader is the same in both. This
is the classic difference between discrimination and identification: two images
might be discriminable but not differently identified. In my field of hearing,
a one-second vowel sampled at 44.1 khHz with 16 bit resolution doesn’t mean that
there is over 705,000 bits of information in that one-second sound sample. A vowel
with a formants at 710 Hz and 1100 Hz will be identified as an /a/, as will a vowel
with formants at 850 Hz and 1300 Hz even though the two data points are considerable
different. The two vowels can be discriminated but are not differentially
identified. I believe that the analytical value of sparklines is not nearly
as large as Tufte calculates.
Tufte the Artist
Tufte spent a small amount of time showing us pictures of
giant sculptures in his garden. While interesting, I’m not sure what the point
of this was.