Wednesday, October 04, 2006

Edward Tufte Lecture Notes

Edward Tufte Lecture 10/2/06: SG Notes
Presenting Data and Information

Use high-density data
· Allow audience to read. People can read 2-4 times faster than you can talk.

Minimize graphic flair
· Allows more space for explanatory material.
· Eliminate boxes on org charts when 2-D location of text already describes relationships inside organization
· Eliminate chartjunk (e.g., explosions; cf FEMA chart)

· Explain what’s going on

Show causality
· Differentiate links/relationships
· Use whatever evidence is important
· Multiple levels of information in one exhibit
· Allow users to tell their own little stories

Presentation Methods
1. Don’t get it original, get it right
2. Model: NYT/WSJ
o News stories
o Sports page
o Weather page
o Stock page
3. This is a solved problem
4. NYT/WSJ are:
o High density
o Conventionally designed—designs that have worked
o Despite complexity of data, millions of people can read the sports page, get lots of information, so don’t underestimate the capability of your audience to understand

Two main problems in Information Display:
1. Multivariate problems
Subproblem: how to display in 2-D/ on paper
2. Information Resolution

Put your name on your work. Work is not done by organizations, it is done by people. Putting your name on it is a sign of pride in craftsmanship, and allows users to direct questions/criticisms back to the author.

Label data directly. Avoid legends & distant labels.

2-D workaround example: Euclid, 1st edition, 1570
· Use of 3-D popups, still functional in book copy almost 500 years later

The Grand Principles of Analytic Design
1. Use comparisons
Rulers, scale, normal ranges, etc
2. Show causality
3. Show many dimensions/levels of data
4. Completely integrate words, numbers and images
“The evidence is indifferent to the mode of production”
5. Document everything and tell people about it
This adds credibility and reasons to believe
Also enables further study
Show all of your data—don’t cherry pick
6. Content counts most of all
Serious presentations stand or fall on the quality, relevance and content of
7. Adjacent in space, not stacked in time
Don’t make users turn the page to see next result. It is easier to compare
adjacent things
8. Use small multiples (cf Galileo sunspot diagrams)
9. Put everything on a universal grid
How big is it?
Where is it?
What is its context?
These are the commonalities among all humans, regardless of time, culture

Low resolution of screens vs. paper
· Computer screen uses ~1/500th of human eye-brain capacity
· 1 glance = ~150Mb data
· 16 bit color (up to 20 bit for artists/designers)
· Paper 10-12 times better than screen for resolution

Aside: p.162 of Galileo’s Demonstrations features the only direct reference he made in print to the movement of the earth: “dal meuimento anno della Terra”. (“The annual movement of the Earth”) The church asked him not to repeat this.

You know what you’re talking about
You aren’t cherry picking

Give lots of handouts
· Lets people use their own cognitive style. People not paying attention to the speaker in order to read the handout is a sign the audience is awake

Respect the audience. Contempt leaks through.
To clarify, add detail
· Sounds paradoxical, but is true

To see best variation, scale average slope of data to 45°
· Avoids ceiling & floor effects
· Most useful for cyclic data; allows best detection of variation
· You want lumpy, not spiky or flat graphs

Use sparklines: intense, simple, word-sized graphics inline with text
· Software: Sparkline—shareware office plug-in
· Other software:
Stats program: SAS, SPSS, “anything that costs more than $500”
Design program: Illustrator
Page layout software

Resolution of words=Resolution of sparkline graphics

Usability tests: do people use something similar

John Tukey: “Better to be approximately right than exactly wrong”

Project management:
· Best to do wallchart
· Make it large, updated regularly
· Readable 2-3’ away

When reading a PPT presentation, ask:
· What’s the story?
· Does it establish credibility
· What’s the scope? Two problems:
o Overreaching
o Irrelevant Domain

PowerPoint Flaws:
· Low Resolution: 4 sheets of paper=50-250 PPT slides
· Chops all info into slide-sized pieces
· Encourages fake hierarchies (bullets)
· Encourages clipped jargon, non-English
· Discourages non-findings
· Encourages pitching, rather than reporting

Pitching out corrupts within (cf Iraq evidence)

Solution: use technical reports, built in word processor, instead of decks

Technical Report:
11”x17” paper, folded once, written on both sides

Start with 200 word intro:
What the problem is
Who cares (what’s the relevance)
What your solution is

Model is article in Science or Nature

Meant to be read through
High resolution data dump
Followed by discussion
If someone derails talk, at least you got through the 200 word summary

One handout means
50-250 slides of average density
1000-2000 words or
500-1000 sparklines

General presentations:
· Worry about content, rather than production values
· Practice, practice, practice
· Video of self reveals: verbal & gestural tics, placeholders, other annoying quirks
· Show up early
1 time in 20, this heads off a serious problem (double booking, etc)
1 time in 20, allows a quiet opportunity to prepare, talk to/greet audience
20 times in 20, you are not late

Start with 200 word, 3 part summary from report
Never apologize (Don’t waste time talking about you, talk about the data)

For presenting complicated data/data display:
PGP=Particular, General, Particular
Give an example, then orient to whole of data, then give another specific example

Give everybody a piece of paper

Don’t condescend to your audience. Instead of “Knowing your audience,” which leads to underestimation: Respect your audience, Know your content.

Use body language

Finish early.

Key readings:
The Visual Display of Quantitative Information, Introduction and Ch. 1
Envisioning Information, Introduction and Ch. 2
Beautiful Evidence, Introduction and pp. 12-45

No comments: