2 - THE CASE AGAINST CAUSAL MODELING

Abstract

This chapter presents the case against causal modeling in artificial intelligence and nonexperimental science. Causal modeling of nonexperimental data has been controversial since its beginning, and it is no less controversial today. The controversies are very philosophical, and they involve fundamental differences about what makes for science. The social and behavioral sciences are inevitably compared with the natural sciences, and disputes about methodology in social and behavioral science are often disputes about what it is that has made the natural sciences so successful in understanding, predicting, and controlling the physical world. The chapter discusses the following criticisms of causal modeling: (1) causal modeling involves a mistaken, or incoherent, conception of causal relations; (2) theories with latent variables should be rejected on methodological or semantic grounds; (3) only experimental data can contribute to the knowledge of causal relationships; (4) those who advocate causal models do not, and presumably cannot, make a case for the assumptions of their models; and (5) linear causal models should be rejected because they are always literally false.