Rethinking the Fundamental Problem of Causal Inference in Political Science

• In Event: Tackling Causality and Causal Inference across the Social Sciences

Abstract

The fundamental problem of causal inference (FPCI) is one of the most widely-touted methodological tenets in contemporary political science. It is at the core of a wide range of comparative and statistical approaches geared towards accounting for causal relations and it informs how students are introduced to the key task of causal analysis in the study of politics.

Our paper takes a different perspective and challenges the oft assumed scope of the problem. We demonstrate that rather than being a problem intrinsically tied to all efforts to make causal inferences, it only relates to a limited subset of counterfactual perspectives on causation and the use of the potential outcomes framework as the guiding principle for making causal inferences in open systems. In particular, processual and generative accounts of causation, along with some counterfactual ones, employed across a variety of approaches in political science do not face the FPCI. Here causal inferences are, for instance, made by tracing the process itself binding a cause and outcome together in some form of productive relationship or by accounting for the configuration of factors that generate a particular outcome.

We discuss these assumptions with a view to developing a clearer perspective on the scope of the FPCI and present a range of causal perspectives that are important for knowledge production in the social sciences and from which the problem does not arise. We then discuss the implications of a more bounded notion of the FPCI for causal analysis in political science, especially as concerns where different research designs and methods are placed in an evidential hierarchy. When the FPCI is treated as fundamental for all methods for making causal inferences the randomized controlled experiment is invariably placed at the top of the hierarchy. A clearer and more narrowly defined scope of the problem reveals several other ‘gold standards’ associated to particular perspectives on causal inference.

Authors

• Ludvig Norman, Stockholm University

• Derek Beach, University of Aarhus