Picturing causality - The serendipitous semiotics of causal graphs

Abstract

Bayesian nets (BNs) appeared in the 1980s as a solution to computational and representational problems encountered in knowledge representation 1 of uncertain information. Shortly afterwards, BNs became an important part of the AI mainstream. During the 1990s, a lively discussion emerged regarding the causal semantics of Bayesian nets, challenging almost a century of statistical orthodoxy regarding inference of causal relations from observational data, and many refer to BNs now as causal graphs. However, the discussion of causal graphs as a data visualization tool has been limited. We argue here that causal graphs together with their causal semantics for seeing and setting, have the potential to be as powerful and generic a data visualization tool as line graphs or pie charts. © Springer-Verlag Berlin Heidelberg 2005.