Data and Statistics: 2011

I had the privilege of hearing Edward Tufte speak on February 7th in San Diego. Here are my notes from the session:

Be driven by content and data, not display method.
Label linking lines if there are relationships
Boxes are not necessary
Goal is to zero out the interface; should be all content

Be sure to include:

Story
Credibility

Also:

Let viewer scan and scroll
Minimize linking/stacking
Use resolution of real life
Graphics can be anywhere
Sparklines are an excellent tool. Built into Excel 2010.
Order by content, not alpha
Eliminate legends/codes where possible
Design and present for highest denominator

Making Reports

1. Find good examples and copy

Google Images
Text, tables, maps
Public Library of Science templates
Look at the Journal of Nature

2. Use performance data

Tables w little data
Graphics w lots of data
Avoid lines/stripes/grids
Trebuchet or gil sans font

3. Find supergraphic to use as opener

There is no such thing as info overload, only lousy design

4. Intellectual model should be like Google News, NY Times, etc.

Think sports section as model
Use executive summaries for micro/macro (w less jargon)

5. Credibility: Use proper language, be clear/concise, be responsive.

Challenges: flatland (dimension) and resolution

Name your work (individually, not by team/dept)

Design principles 1) Causality; 2) comparison; 3) multivariate complexity; 4) integrate words, numbers, images, diagrams; 5) thoroughly describe evidence; 6) content counts most of all (quality, relevance, integrity); 7) do important things adjacent in space

I asked Tufte how one knows which data to look at in their multivariate analysis. His answer: "you have an idea" ... Too many analysts, he said, don't have ideas and therefore look at everything.

I asked him education measures in particular. He said that when he consulted for IBM they used to say that if there was no practical application for something it must be for education.

Read, present, then allow time for questions and answers

Many square inches, many pixels per square inch = analysis (ie multiple monitors)

Small multiple technique (a la sunspots)

Presenting data