Tuesday, August 7, 2012

Edward Tufte Excerpt on the Design of Causal Diagrams

Design of causal diagrams: Barr art chart, Lombardi diagrams, evolutionary trees, Feynman diagrams, timelines

by Edward Tufte
  It is easy to draw a linking line or an arrow of implied causality, but very hard to make credible causal inferences.   Linking lines, arrows, and influence trees bring with them many implicit but powerful assumptions. Suppose we take the arrows seriously--how are we to evaluate the evidential quality of influence diagrams? The answer is clear, at least to the open and skeptical mind: by the usual standards for evidence of causality. A good start on what it takes to make causal inferences is found in A. Bradford Hill's classic paper on making causal inferences (in Hill's case, about the link between smoking and lung cancer), posted here. The principles of making sound causal inferences favor no particular ideology or point of view, except that of wanting to find out what is actually going on.   Below is a discussion of the famous art chart of Alfred Barr. The analysis applies to evolutionary trees, Lombardi diagrams, and a good many timelines. The last two paragraphs are particularly important with regard to the assumptions involved in linking lines.   The material below is from a draft of my book, Beautiful Evidence.\    

For more on the topic, visit Edward Tufte's website at:


Beautiful Evidence by Edward Tufte.

No comments:

Post a Comment