Measuring variation : an epistemological account of causality and causal modelling
Abstract (Summary)This doctoral dissertation deals with causal modelling in the social sciences. The specific question addressed here is: what is the notion, or the rationale, of causality involved in causal models? The answer to that epistemological query emerges from a careful analysis of the social science methodology, of a number of paradigmatic case studies and of the philosophical literature. The main result is the development of the rationale of causality as the measure of variation. This rationale conveys the idea that to test – i.e. to confirm or disconfirm – causal hypotheses, social scientists test specific variations among variables of interest. The notion of variation is shown to be embedded in the scheme of reasoning of probabilistic theories of causality, in the logic of structural equation models and covariance structure models, and is also shown to be latent in many philosophical accounts. Further, the rationale of causality as measure of variation leaves room for a number of epistemological consequences about the warranty of the causal interpretation of structural models, the levels of causation, and the interpretation of probability. Firstly, it is argued that what guarantees the causal interpretation is the sophisticated apparatus of causal models, which is made of statistical, extra-statistical and causal assumptions, of a background context, of a conceptual hypothesis and of a hypothetico-deductive methodology. Next, a novel defence of twofold causality is provided and a principle to connect population-level causal claims and individual-level causal claims is offered. Last, a Bayesian interpretation of probability is defended, in particular, it is argued that empirically-based Bayesianism is the interpretation that best fit the epistemology of causality here presented. The rationale of variation is finally shown to be involved or at least consistent with a number of alternative accounts of causality; notably, with the mechanist and counterfactual approach, with agency-manipulability theories and epistemic causality, with singularist accounts and with causal analysis by contingency tables.
Source Type:Master's Thesis
Keywords:probability causal modelling causality
Date of Publication:06/17/2005