Two algorithms for inducing causal models from data
Dawn E. Gregory, Paul R. Cohen · 1994
Many methods have been developed for inducing cause from statistical data. Those employing linear regression have historically been discounted, due to their inability to distinguish true from spurious cause. We present a regression-based statistic that avoids this problem by separating direct and indirect influences. We use this statistic in two causal induction algorithms, each taking a different approach to constructing causal models. We demonstrate empirically the accuracy of these algorithms. This work is supported by ARPA/Rome Laboratory under contract #'s F30602-91-C-0076 and F30602-93-C-0100. To appear in Proceedings of AAAI-94 Workshop on Knowledge Discovery in Databases. 1 Causal Modeling Causal modeling is a method for representing complex causal relationships within a set of variables. Often, these relationships are presented in a directed, acyclic graph. Each node in the graph represents a variable in the set, while the links between nodes represent direct, causal relati...