Talk (Work in Progress): Alessio Moneta (Scuola Superiore Sant'Anna)

Title:

Higher-level Causation and Causal Inference (with a Glance to Instrument Validity)

Abstract:

Experimental methods for causal inference (randomized controlled trials) are believed to conclusively identify causal relations in virtue of realizing ideal conditions (Woodward 2003) that avoid confounding. We observe that many high-level aggregate variables have potentially ambiguous effects on other variables due to their heterogeneous causal role in the population of interest (Spirtes and Scheines 2004). We argue that, when heterogeneity is present and when data on individual units are unavailable, experiments provide a much weaker inferential leverage. The reason is that the ideal conditions on which a conclusive inference would depend are in principle unrealizable. Contrary to the case of variables with homogeneous causal roles, the evidence may not conclusively validate an experiment because confounding may never be ruled out. Granting that causal inference may be warranted in such contexts, the problem arises of how exactly it should be justified. We propose a rationalization based on a form of abductive reasoning. Finally, looking at non-experimental settings, we show how high-level "ill-defined'' variables undermine instrument validity.

The paper I am going to present is a joint work with Lorenzo Casini (University of Bologna).