PARENT SESSION
Symposium 9: Structural Equations, Path Analysis, and Other Causal Models
Organized by: S Scheiner
Tuesday, August 5. 1:30 PM to 5:00 PM, SITCC Chatham Ballroom C.

Translating from the language of causality to the language of probability (and back ...).

Shipley, Bill^*,1, ¹ Département de biologie, Sherbrooke, Québec, Canada

ABSTRACT- Ecologists invoke the notion of causality to express an asymmetric relationship between a cause (C) and its effect (E): C→E. Probability theory, the official language of statistics, cannot express such asymmetrical relationships and instead describes only symmetrical relationships between random variables. This has led to a century of confusion between the relationship between causality and statistics, and has resulted in ecologists consistently mistranslating their causal claim (C→E) as E=f(C). Here, I give a user-friendly introduction to an alternative mathematical language that can express asymmetrical relationships (the theory of directed graphs) and describe the link between directed graphs and probability distributions. I then present a new, general and robust statistical test (d-sep test) of path models that allows one to test causal claims without physical manipulation, and also how the same logic can be used to explore observational data and generate causal hypotheses. I conclude by describing the relationship of these methods to structural equation models.

Key words: statistical methods, causal analysis, path analysis, structural equation models