Probabilistic causal systems and the conditional probability approach to causal analysis

Abstract

In this paper we introduce an alternative approach to the analysis of causal systems that involve probabilistic causation, the conditional probability (CP) approach. We show how the CP approach with its composition and decomposition rules can deal with various kinds of general causal systems, and explicate an important distinction between pure-“or” and pure-“and” causal systems. We argue that the CP approach provides a promising and quite general approach to causal analysis in the social sciences where probabilistic causation is involved in a causal system. Approaches such as path analysis and structural-equation methods have been recommended widely for analyzing causal systems that contain variables having direct and indirect effects on other variables and where probabilistic causation is involved. We illustrate some of the features of the CP approach by comparing it with these approaches and by showing that the path-analysis/structural-equation approaches lead to the wrong answers. Finally, we illustrate some of the features of the CP approach by comparing it with the approach developed by Birnbaum (1982). Birnbaum holds that his approach is a method for giving a causal analysis of contingency tables, and claims that the method allows the strength and the form of causality to be estimated from a contingency table. We give several reasons for holding that, overall, Birnbaum’s approach is inadequate.